Integrating biomarkers to predict renal and cardiovascular drug efficacy Schievink, Bauke

Transcription

1 University of Groningen Integrating biomarkers to predict renal and cardiovascular drug efficacy Schievink, Bauke IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below. Document Version Publisher's PDF, also known as Version of record Publication date: 2016 Link to publication in University of Groningen/UMCG research database Citation for published version (APA): Schievink, B. (2016). Integrating biomarkers to predict renal and cardiovascular drug efficacy: PRE score applications from drug registration to personalized medicine. [Groningen]: University of Groningen. Copyright Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons). Take-down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Downloaded from the University of Groningen/UMCG research database (Pure): For technical reasons the number of authors shown on this cover page is limited to 10 maximum. Download date:

2 Integrating biomarkers to predict renal and cardiovascular drug efficacy PRE score applications from drug registration to personalized medicine Bauke Schievink

3 The studies presented in this thesis were partly funded by the Escher Value Creation project (T6-503) of the Dutch Top Institute Pharma. Printing of this thesis was supported by the University of Groningen, the research institute GUIDE and the University Medical Center Groningen. ISBN (printed version): ISBN (digital version): Printed by: Ipskamp Printing Bauke Schievink No parts of this thesis may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system, without permission of the author.

4 Integrating biomarkers to predict renal and cardiovascular drug efficacy PRE score applications from drug registration to personalized medicine PhD thesis to obtain the degree of PhD at the University of Groningen on the authority of the Rector Magnificus Prof. E. Sterken and in accordance with the decision by the College of Deans. This thesis will be defended in public on Wednesday 9 March 2016 at hours by Bauke Hendrik Schievink born on 2 January 1988 in Marum

5 Supervisors Prof. D. de Zeeuw Prof. H.J. Lambers Heerspink Assessment Committee Prof. D.E. Grobbee Prof. H.G.M. Leufkens Prof. P. Rossing

6 Table of contents Page Chapter 1 Introduction and Scope of the Thesis 7 Current Opinion in Nephrology and Hypertension, Chapter 2 Early renin-angiotensin-system intervention is more beneficial 17 than late intervention in delaying end-stage renal disease in type 2 diabetes Diabetes, Obesity and Metabolism, 2015 Chapter 3 Prediction of heart failure outcomes in patients treated with 37 aleglitazar based on short-term changes in multiple risk markers Submitted for publication Chapter 4 Prediction of the effect of atrasentan on renal and heart failure 53 outcomes based on short-term changes in multiple risk markers European Journal of Preventive Cardiology, Chapter 5 The renal protective effect of Angiotensin Receptor Blockers 73 depends on intra-individual response variation in multiple risk markers British Journal of Clinical Pharmacology, Chapter 6 The Use of Surrogate Endpoints in Regulating Medicines for 97 Cardio-Renal Disease: Opinions of Stakeholders PLoS One, Chapter 7 Summary and Future Perspectives 117 Nederlandse samenvatting 127 Acknowledgements 137

7

8 Chapter 1 Introduction and scope of the thesis Adapted from: Current Opinion in Nephrology and Hypertension, 2015.

9 Introduction and scope of the thesis Introduction The prevalence of chronic kidney disease (CKD) is increasing worldwide, and forecasts for 2030 indicate that the number of patients requiring renal replacement therapies will more than double (1). The increase in requirement of renal replacement therapies and the availability of only few proven effective therapies highlight the need to develop new drugs and intervention strategies. To develop new drug interventions, regulatory authorities (i.e. Food and Drug Administration, European Medicine Agency) require that the demonstrated benefits outweigh risks. To this end, new drugs have to show a beneficial effect in well-designed clinical trials on accepted clinically meaningful endpoints, and these benefits must offset any adverse effects the patient may experience during the use of the drug. In trials of CKD progression, end-stage renal disease (ESRD) is an accepted, clinically meaningful endpoint because it is accompanied by a large disease burden and shortened survival. However, since progression to ESRD may take decades, large and complex clinical trials are needed to demonstrate drug efficacy (2-4). This results in large financial and human investments to test new drugs for patients with CKD. The increasing size of clinical trials and related investments in combination with high drug attrition rates in late phase clinical trials has fostered exploration of alternative approaches to test the efficacy and safety of new drugs (5-7). One intuitive alternative is replacing clinically meaningful endpoints by surrogate endpoints. A surrogate endpoint is an intermediate outcome, usually a laboratory measurement of a relevant risk marker, which substitutes the clinically meaningful endpoint (8). An example of an accepted surrogate endpoint in trials of CKD progression is doubling of serum creatinine, equivalent to a halving of kidney function. However, doubling of serum creatinine is still a late event in progression of kidney disease and therefore trials still require large sample sizes and long duration of follow-up to determine drug efficacy. Therefore, there is interest in exploring alternative surrogate endpoints that can be ascertained earlier in the course of renal disease, leading to shorter durations of follow-up in clinical trials. Examples of such surrogates are lesser declines than a halving in estimated glomerular filtration rate (egfr), transition in egfr classes, or changes in albuminuria. egfr represents the filtration power of the kidney and is accepted as 8

10 Chapter 1 the best index of overall kidney function. egfr decline is a necessary intermediate on the pathway of ESRD and it is therefore not surprising that various studies showed a strong and graded association between small reductions in egfr and risk of developing ESRD. Based on the strong mathematical and biological association between egfr and ESRD, lesser declines than a halving of GFR have received ample attention as potential surrogate endpoint in a series of meta-analyses of observational studies and clinical trials (9-14). As an alternative to GFR, albuminuria is proposed as a surrogate endpoint (15). The difference with egfr, which is a direct marker of kidney function, is that albuminuria, just like blood pressure, is not only a marker of kidney damage but also causally implicated in renal disease progression (15,16). Treatment with angiotensin converting enzyme inhibitors (ACEi) or angiotensin receptor blockers (ARB) decreases albuminuria and confers renoprotection. Post-hoc analyses from clinical trials repeatedly showed that the initial reduction in albuminuria with ACEi or ARB is the driving parameter for renoprotection (17-19). Emerging clinical trial data show that ACEi or ARBs are not the only drugs that decrease albuminuria and slow the progression of CKD. First, a randomized clinical trial showed that pentoxifylline, a xanthine derivative registered for treatment of peripheral vascular disease, decreases albuminuria in patients with diabetes and nephropathy and slows the progression of renal function decline relative to placebo (20). In addition, the recent PLANET trials reported that atorvastatin, but not rosuvastatin, decreased proteinuria after 14 weeks treatment (21). However, developing novel surrogate endpoints based on a single renal risk marker such as egfr decline or albuminuria may not be optimal. In the current drug development and registration process a single renal risk marker is selected and a drug is targeted towards that risk marker. However, there are multiple causes of renal disease that may not all be captured by a single risk marker alone, and drugs have multiple effects beyond the target risk marker (so-called off-target effects). These additional drug effects may also influence the ultimate renal endpoint; they either contribute to or counteract the on-target risk marker effect. For example, in addition to blood pressure lowering, RAS intervention also lowers albuminuria which contributes to the renal protective effect. However, these drugs also increase serum potassium, which is associated with increased renal risk. Hence, the rise in serum potassium may blunt the beneficial effect of blood pressure lowering (and 9

11 Introduction and scope of the thesis albuminuria reduction) (22,23). Several recent clinical trials showed no benefit on clinically meaningful outcomes despite the drug exerting beneficial effects on the ontarget parameter (4,24-28). In these cases off-target drug effects may confound the relationship between a single risk marker or surrogate and renal outcome. This suggests that a score that integrates all known drug-induced effects is potentially more accurate in predicting the ultimate drug effect than using a single risk marker alone (29). Using such multiple risk parameter scores to establish drug efficacy in clinical trials is not much different from the risk prediction scores that are used to establish a patient's renal risk. When we aim to predict the risk of an individual we accept that multiple risk markers are of relevance to the clinical outcome, and thus risk prediction scores are developed consisting of multiple risk markers. In drug development a single risk marker is selected as target for therapy. However, nearly all drugs have effects on multiple risk markers. Therefore it seems logical to integrate these multiple drug effects to better predict the ultimate drug effect. Recently, a score was developed that integrates multiple short-term drug effects in order to predict the long-term drug effect on renal and cardiovascular outcomes. This socalled PRE score was used to establish the effect of the angiotensin receptor blockers losartan and irbesartan in patients with type 2 diabetes. Interestingly, the ARBs significantly changed 7 out of 11 measured renal risk markers. Integrating all risk markers in a PRE score showed that it provided a better prediction of the drug effect on hard renal outcomes than any change in single markers (30,31). External validation studies confirmed these results (32). Therefore, such scores may be better suited as surrogate endpoint in clinical trials than using a single marker alone. Although using changes in multiple parameters to establish drug efficacy is intuitively appealing, the approach is still in its infancy and requires more prospective validation. In particular, the validity of the multiple parameter drug response score has been ascertained for drugs intervening in the RAS but it is unclear whether the score will be equally valid for other drugs. Especially whether the score can be used for novel drugs targeting for example inflammatory pathways requires further investigation. Moreover, the risk markers currently included in the score are limited to what is measured and recorded in the trials: physical measurements and standard biochemical measurements. Novel risk markers may be identified and integrated to improve its accuracy. 10

12 Chapter 1 Another concern is that recent clinical drug trials in type 2 diabetes were terminated early for safety reasons, due to an increase in hospitalization due to heart failure in the active treatment arm. This was the case for the Nrf2 activator bardoxolone methyl, the endothelin antagonist avosentan and the peroxisome proliferator-activated receptor activator rosiglitazone (24,33,34). These drug failures illustrate that the validity of the score needs to be ascertained for other outcomes than ESRD, in order to provide a more complete profile of the benefit-risk profile of a given drug. Scope of the thesis Better prediction of drug effects on clinical outcomes may be achieved by risk scores that integrate the effect on multiple risk markers, instead of using single markers alone. Such an approach, with the so-called PRE score, was previously used for drugs intervening in the RAS, but before it can be implemented in practice more validation is required. In this thesis we determined whether this multiple parameter drug response score could also be used for other drug classes that are used to treat patients with type 2 diabetes. In addition, we aimed to address both safety and efficacy by predicting ESRD and heart failure outcomes. Another important aspect is that drug effects are frequently assessed on a group level, but whether the response in multiple risk markers in response to therapy can improve prediction of who is likely to benefit from treatment is not yet known. And lastly, a multiple parameter drug response score such as proposed in this thesis needs to be accepted by all stakeholders in drug development. This includes academics, the pharmaceutical industry and regulatory authorities, and their willingness to accept such a score needs to be ascertained. In Chapter 2 we investigated whether early intervention with drugs that intervene in the RAS is more beneficial in delaying ESRD in patients with type 2 diabetes than intervention in later stages of the disease. Because ESRD can take decades to manifest, prospective clinical trial data is not available to answer this question. Therefore we built a model with patient data from completed clinical trials in nephrology from all stages of CKD. The model is based on disease stages defined by a combination of the risk markers albuminuria and egfr. 11

13 Introduction and scope of the thesis Figure 1. Schematic representation of the PRE score. In step A the associations between multiple risk markers (e.g. blood pressure, albuminuria) and clinical outcomes (e.g. ESRD) are established. In step B these associations are applied to the baseline and follow-up risk marker measurements in each patient. The PRE score then predicts the individual risk change of clinical outcomes induced by treatment. In Chapter 3 we performed a post-hoc analysis of the AleCardio trial in which patients with type 2 diabetes were treated with the dual PPAR agonist aleglitazar. The trial was stopped early due to futility and an increase in hospitalization due to heart failure in the treatment arm. Our aim was to predict whether the adverse heart failure outcomes could have been prevented by more stringent baseline inclusion criteria by using individual or multiple risk markers for heart failure. Secondly, we investigated whether the observed heart failure risk could have been predicted based on short-term response in individual risk markers, or a composite consisting of multiple risk markers. In Chapter 4 we applied the multiple parameter drug response score to a clinical trial of the endothelin antagonist atrasentan, to prospectively 12

14 Chapter 1 predict the outcome of an ongoing phase III trial on the endpoints ESRD and hospitalization due to heart failure. In Chapter 5 we applied the multiple parameter drug response score on individual patients that were subjected to drugs that intervene in the RAS. We determined whether integrating the effect on multiple risk markers in response to RAS intervention would improve the prediction of who will benefit from treatment, compared to using single risk markers alone. In Chapter 6 we performed a questionnaire to investigate whether stakeholders in drug development would be willing to accept novel surrogate endpoints, and whether they deemed surrogate endpoints based on multiple risk markers more accurate than using single markers alone. 13

15 Introduction and scope of the thesis References (1) Liyanage T, Ninomiya T, Jha V, Neal B, Patrice HM, Okpechi I, et al. Worldwide access to treatment for end-stage kidney disease: a systematic review. The Lancet 2015; 385(9981): (2) Brenner BM, Cooper ME, de Zeeuw D, Keane WF, Mitch WE, Parving H, et al. Effects of Losartan on Renal and Cardiovascular Outcomes in Patients with Type 2 Diabetes and Nephropathy. N Engl J Med 2001; 345(12): (3) Lewis EJ, Hunsicker LG, Clarke WR, Berl T, Pohl MA, Lewis JB, et al. Renoprotective Effect of the Angiotensin-Receptor Antagonist Irbesartan in Patients with Nephropathy Due to Type 2 Diabetes. N Engl J Med 2001; 345(12): (4) Parving H, Brenner BM, McMurray JJV, de Zeeuw D, Haffner SM, Solomon SD, et al. Cardiorenal End Points in a Trial of Aliskiren for Type 2 Diabetes. N Engl J Med 2012; 367(23): (5) Palmer SC, Sciancalepore M, Strippoli GFM. Trial Quality in Nephrology: How Are We Measuring Up? American Journal of Kidney Diseases 2011; 58(3): (6) DiMasi JA, Hansen RW, Grabowski HG. The price of innovation: new estimates of drug development costs. J Health Econ 2003; 22(2): (7) DiMasi JA, Feldman L, Seckler A, Wilson A. Trends in Risks Associated With New Drug Development: Success Rates for Investigational Drugs. Clin Pharmacol Ther 2010; 87(3): (8) Biomarkers Definitions Working Group. Biomarkers and surrogate endpoints: Preferred definitions and conceptual framework. Clin Pharmacol Ther 2001; 69(3): (9) Inker LA, Lambers Heerspink HJ, Mondal H, Schmid CH, Tighiouart H, Noubary F, et al. GFR Decline as an Alternative End Point to Kidney Failure in Clinical Trials: A Meta-analysis of Treatment Effects From 37 Randomized Trials. Am J Kidney Dis 2015; 64(6): (10) Coresh J, Turin T, Matsushita K,et al. Decline in estimated glomerular filtration rate and subsequent risk of end-stage renal disease and mortality. JAMA 2014; 311(24): (11) Lambers Heerspink HJ, Weldegiorgis M, Inker LA, Gansevoort R, Parving H, Dwyer JP, et al. Estimated GFR Decline as a Surrogate End Point for Kidney Failure: A Post Hoc Analysis From the Reduction of End Points in Non Insulin-Dependent Diabetes With the Angiotensin II Antagonist Losartan (RENAAL) Study and Irbesartan Diabetic Nephropathy Trial (IDNT). Am J Kidney Dis 2014; 63(2): (12) Sontrop JM, Weir MA, Garg AX. Surrogate Outcomes for ESRD Risk: The Case for a 30% Reduction in Estimated GFR Over 2 Years. Am J Kidney Dis 2014; 64(6): (13) Lambers Heerspink HJ, Tighiouart H, Sang Y, Ballew S, Mondal H, Matsushita K, et al. GFR Decline and Subsequent Risk of Established Kidney Outcomes: A Meta-analysis of 37 Randomized Controlled Trials. Am J Kidney Dis 2015; 64(6): (14) Greene T, Teng C, Inker LA, Redd A, Ying J, Woodward M, et al. Utility and Validity of Estimated GFR-Based Surrogate Time-to-Event End Points in CKD: A Simulation Study. Am J Kidney Dis 2015; 64(6): (15) Roscioni SS, Lambers Heerspink H,J., de Zeeuw D. Microalbuminuria: target for renoprotective therapy PRO. Kidney Int 2014; 86(1):

16 Chapter 1 (16) Parving H, Persson F, Rossing P. Microalbuminuria: A parameter that has changed diabetes care. Diabetes Res Clin Pract 2015; 107(1): 1-8. (17) de Zeeuw D, Remuzzi G, Parving H, Keane WF, Zhang Z, Shahinfar S, et al. Proteinuria, a target for renoprotection in patients with type 2 diabetic nephropathy: Lessons from RENAAL. Kidney Int 2004; 65(6): (18) Hellemons ME, Persson F, Bakker SJL, Rossing P, Parving H, De Zeeuw D, et al. Initial Angiotensin Receptor Blockade Induced Decrease in Albuminuria Is Associated With Long-Term Renal Outcome in Type 2 Diabetic Patients With Microalbuminuria: A post hoc analysis of the IRMA-2 trial. Diabetes Care 2011; 34(9): (19) Atkins RC, Briganti EM, Lewis JB, Hunsicker LG, Braden G, Champion de Crespigny PJ, et al. Proteinuria reduction and progression to renal failure in patients with type 2 diabetes mellitus and overt nephropathy. Am J Kidney Dis 2005; 45(2): (20) Navarro-González JF, Mora-Fernández C, Muros de Fuentes M, Chahin J, Méndez ML, Gallego E, et al. Effect of Pentoxifylline on Renal Function and Urinary Albumin Excretion in Patients with Diabetic Kidney Disease: The PREDIAN Trial. J Am Soc Nephrol 2015; 26(1): (21) de Zeeuw D, Anzalone DA, Cain VA, Cressman MD, Heerspink HJL, Molitoris BA, et al. Renal effects of atorvastatin and rosuvastatin in patients with diabetes who have progressive renal disease (PLANET I): a randomised clinical trial. Lancet Diabetes Endocrinol 2015/; 3(3): (22) Miao Y, Dobre D, Lambers Heerspink HJ, Brenner BM, Cooper ME, Parving H, et al. Increased serum potassium affects renal outcomes: a post hoc analysis of the Reduction of Endpoints in NIDDM with the Angiotensin II Antagonist Losartan (RENAAL) trial. Diabetologia 2011; 54(1): (23) Heerspink HJL, Gao P, Zeeuw Dd, Clase C, Dagenais GR, Sleight P, et al. The effect of ramipril and telmisartan on serum potassium and its association with cardiovascular and renal events: Results from the ONTARGET trial. Eur J Prev Cardiol 2014; 21(3): (24) Mann JFE, Green D, Jamerson K, Ruilope LM, Kuranoff SJ, Littke T, et al. Avosentan for Overt Diabetic Nephropathy. J Am Soc Nephrol 2010; 21(3): (25) Telmisartan, Ramipril, or Both in Patients at High Risk for Vascular Events. N Engl J Med 2008; 358(15): (26) Fried LF, Emanuele N, Zhang JH, Brophy M, Conner TA, Duckworth W, et al. Combined Angiotensin Inhibition for the Treatment of Diabetic Nephropathy. N Engl J Med 2013; 369(20): (27) James WP, Caterson ID, Coutinho W, Finer N, Van Gaal LF, Maggioni AP, et al. Effect of Sibutramine on Cardiovascular Outcomes in Overweight and Obese Subjects. N Engl J Med 2010; 363(10): (28) Barter PJ, Caulfield M, Eriksson M, Grundy SM, Kastelein JJP, Komajda M, et al. Effects of Torcetrapib in Patients at High Risk for Coronary Events. N Engl J Med 2007; 357(21): (29) Heerspink HJL, Grobbee DE, de Zeeuw D. A novel approach for establishing cardiovascular drug efficacy. Nat Rev Drug Discov 2014; 13(12): 942. (30) Smink PA, Miao Y, Eijkemans MJC, Bakker SJL, Raz I, Parving H, et al. The Importance of Short-Term Off-Target Effects in Estimating the Long-Term Renal and Cardiovascular Protection of Angiotensin Receptor Blockers. Clin Pharmacol Ther 2014; 95(2):

17 Introduction and scope of the thesis (31) Schievink B, de Zeeuw D, Parving H, Rossing P, Lambers Heerspink HJ. The renal protective effect of Angiotensin Receptor Blockers depends on intra-individual response variation in multiple risk markers. Br J Clin Pharmacol 2015, advance online publication. DOI: /bcp (32) Smink P, Hoekman J, Grobbee D, Eijkemans M, Parving H, Persson F, et al. A prediction of the renal and cardiovascular efficacy of aliskiren in ALTITUDE using short-term changes in multiple risk markers. Eur J Prev Cardiol 2014; 21(4): (33) de Zeeuw D, Akizawa T, Audhya P, Bakris GL, Chin M, Christ-Schmidt H, et al. Bardoxolone Methyl in Type 2 Diabetes and Stage 4 Chronic Kidney Disease. N Engl J Med 2013; 369(26): (34) Komajda M, McMurray JJV, Beck-Nielsen H, Gomis R, Hanefeld M, Pocock SJ, et al. Heart failure events with rosiglitazone in type 2 diabetes: data from the RECORD clinical trial. Eur Heart J 2010; 31(7):

18 Chapter 2 Early renin-angiotensin-system intervention is more beneficial than late intervention in delaying end-stage renal disease in type 2 diabetes Bauke Schievink Tobias Kröpelin Skander Mulder Hans-Henrik Parving Giuseppe Remuzzi Jamie Dwyer Pepijn Vemer Dick de Zeeuw Hiddo Lambers Heerspink Diabetes, Obesity and Metabolism, 2015

19 Early RAS intervention is more beneficial than late intervention in delaying ESRD Abstract Aims: Intervening in the renin-angiotensin-system (RAS) early in the course of diabetic kidney disease (DKD) may be more beneficial in delaying end-stage-renaldisease (ESRD) than late intervention. However, prospective randomized controlled trial data are lacking. We developed and validated a model to simulate progression of DKD from early onset until ESRD, and assessed the effect of RAS intervention in early, intermediate and advanced stages of DKD. Methods: We used data from BENEDICT, IRMA-2, RENAAL and IDNT trials that assessed effects of RAS intervention in patients with type 2 diabetes. We built a model with discrete disease stages based on albuminuria and egfr. Using survival analyses we assessed the effect of RAS intervention on delaying ESRD in early (egfr >60 ml/min/1.73m 2 and albumin:creatinine ratio (ACR) <30 mg/g), intermediate (egfr ml/min/1.73m 2 or ACR mg/g) and advanced (egfr <30 ml/min/1.73m 2 or ACR >300 mg/g) stages of DKD for patients in different age groups. Results: For patients at early, intermediate and advanced stage of disease being 60 years on average and receiving placebo, median time to ESRD was 21.4, 10.8 and 4.7 years, respectively. RAS intervention delayed the predicted time to ESRD by 4.2, 3.6 and 1.4 years, respectively. Benefit of early RAS intervention was more pronounced in younger patients. For example for patients aged 45 years on average, RAS intervention at early, intermediate or advanced stage delayed ESRD by 5.9, 4.0 and 1.1 years versus placebo. Conclusions: RAS intervention early in the course of proteinuric DKD is more beneficial than late intervention in delaying ESRD. 18

20 Chapter 2 Introduction It has been suggested that intervention in the renin-angiotensin system (RAS) early in the course of type 2 diabetic kidney disease (DKD) might be more beneficial than intervention in later stages of disease, in order to prevent progression to end-stage renal disease (ESRD) [1,2]. Unfortunately, there are no prospective randomized controlled trials that have tested the effect of early RAS intervention on hard renal endpoints, because progression of DKD to end-stage renal disease (ESRD) can take decades to manifest. Clinical trials would therefore require an unfeasibly long followup time. Progression of DKD is characterized by several stages [3]. Initially, the harmful hyperglycemic effects in type 2 diabetes may yield a compensatory response in the kidney by increasing glomerular pressure, leading to hyperfiltration. The hyperfiltrating nephrons cause an increase in the filtration of plasma proteins, including albumin, that leads to microalbuminuria. In later stages of disease glomerular filtration rate declines due to progressive kidney damage and loss of functional nephrons, often exacerbated by hypertension and increasing levels of albuminuria, ultimately culminating in ESRD. The current classification of DKD is based on both albuminuria and glomerular filtration rate (GFR) [4]. Past clinical trials have been conducted at different stages of DKD [5-11]. These trials recorded transition in egfr or albuminuria categories and determined the effect of RAS intervention using transitions in albuminuria stages (i.e. micro or macroalbuminuria) or ESRD as endpoint. One way to determine the treatment effect of RAS intervention early in the course of DKD would be to connect data from these past clinical trials in order to simulate the progression of DKD from early onset to ESRD and to assess the effect of RAS intervention at different stages of DKD. This would provide insight as to whether treatment initiation in early stages of DKD is more beneficial in delaying ESRD than initiation in advanced stages. The first aim of our study was therefore to develop and validate a statistical model to simulate progression of DKD from early onset to ESRD, by connecting data from past clinical trials in early, intermediate and advanced disease stages. Secondly, we assessed the effect of RAS treatment on ESRD in early, intermediate and advanced stages of DKD. Since the incidence of type 2 diabetes is increasing 19

21 Early RAS intervention is more beneficial than late intervention in delaying ESRD strikingly among individuals aged below 40 years [12-15], we also assessed the impact of treatment initiation at different stages of disease in age-specific subgroups. Thirdly, we compared the treatment effect of RAS inhibition in patients responding to RAS treatment (based on a >30% initial decrease in albuminuria) versus patients who do not respond to RAS intervention. Methods Databases and data selection We used data from the following completed clinical trials: BErgamo NEphrologic DIabetes Complications Trial (BENEDICT), Irbesartan in Patients with Type 2 Diabetes and Microalbuminuria Study (IRMA-2), Reduction of Endpoints in NIDDM with the Angiotensin II Antagonist Losartan (RENAAL) and Irbesartan Diabetic Nephropathy Trial (IDNT) which included patients with type 2 diabetes. Design and results have been published elsewhere [5-8]. In all trials patients gave informed consent. Our study was conducted in accordance with the principles of the Declaration of Helsinki as revised in All trials investigated the effect of RAS inhibition (ACE-I/ARB). In BENEDICT, 1209 hypertensive patients with normoalbuminuria (<20μg/min urinary albumin excretion) and serum creatinine <= 1.5mg/dL were randomly allocated to treatment with trandolapril, verapamil, their combination or placebo. Primary outcome was transition from normo- to microalbuminuria. The median follow-up time was 3.6 years. In IRMA-2, 590 hypertensive patients with microalbuminuria (20 to 200 μg/min urinary albumin excretion) and serum creatinine levels <1.5 mg/dl (men) and <1.1mg/dL (women) were enrolled. Patients were randomly allocated to either placebo or irbesartan (150 or 300 mg) treatment. Primary outcome was the transition to macroalbuminuria. The median follow-up time was 2.0 years. RENAAL and IDNT both enrolled patients (RENAAL: 1513 patients, IDNT: 1715 patients) with type 2 diabetes and macroalbuminuria (>300mg/g albumin:creatinine ratio in RENAAL and >900mg/24h proteinuria in IDNT), with serum creatinine levels between 1.0 and 3.0 mg/dl. Patients were randomly allocated to losartan or placebo in RENAAL, or irbesartan, amlodipine or placebo in IDNT. Primary outcome was time to first event of a composite renal endpoint including doubling of serum creatinine, ESRD or death. 20

22 Chapter 2 IDNT included serum creatinine >6mg/dL as an additional component to the primary outcome. The median follow-up time was 3.7 years for RENAAL and 3.4 years for IDNT. Albuminuria and egfr were measured at baseline and every 3 months in RENAAL and IDNT and every 6 months in BENEDICT and IRMA-2. Classification and modeling of diabetic kidney disease progression To simulate the progression of DKD we built disease stages based on albuminuria and egfr classes [16]. To this end we used the following albuminuria strata: 0-15, >15-30, >30-150, > , > and >1000 mg/g albumin:creatinine ratio. EGFR strata were: >90, 90- >60, 60- >30, 30- >15 and <15 ml/min/1.73m 2. Due to low numbers in some strata, all patients with egfr below 15ml/min/1.73m 2 were merged in one group irrespective of their albuminuria and patients with albuminuria 0-30mg/g and 15-30ml min/1.73m 2 egfr were merged. Occurrence of ESRD, defined as the need for renal replacement therapy (dialysis or transplantation), was recorded as the renal endpoint. All-cause mortality was used as a censoring event in the model. Albuminuria and egfr follow-up data was used to determine individual course of kidney disease over time. If more than two subsequent albuminuria or egfr values were missing during follow-up: those values were imputed using a last observation carried forward approach. If there were more than two subsequent missing values the patient was censored. Progression was defined as transition to a worse stage in renal disease (either a worsening in albuminuria, egfr or both). A transition to the next category had to be accompanied with at least a 30% increase in albuminuria or confirmed by the next follow-up measurement. Modeling of diabetic kidney disease progression was performed in two steps. Firstly, time to a transition in disease stage was estimated using survival analysis. Secondly, we used multinomial regression to calculate patient-specific probabilities for every possible transition from each disease stage (first event of worsening in albuminuria stage, worsening of egfr stage, worsening in both or death). The models included treatment allocation, age, gender and systolic blood pressure as covariates. These covariates were selected because they provided the best overall model fit, as determined by AIC. The multinomial regression models contained calculated time-to-event as determined in step 1 as an additional covariate. For model building purposes nonparametric data were log transformed and log values were used in further analyses. Statistical analysis were conducted using R version 21

23 Early RAS intervention is more beneficial than late intervention in delaying ESRD (R Project for Statistical Computing, with a two sided P value <0.05 considered significant. Simulating diabetic kidney disease progression Two steps were performed to simulate patient-specific disease progression. Firstly, actual patient-specific time to transition from the survival model was based on a random pick from the 95% confidence interval around the patient-specific point estimate. We introduced this form of randomness to take into account patientspecific variability. Secondly, the transition direction (i.e. progression in albuminuria, progression in egfr, progression in both or death) was determined by a random weighted pick based on the probabilities derived from the multinomial regression model (i.e. transitions with a higher probability are more likely). After time (step 1) and direction (step 2) were calculated, the simulated patient entered a new disease stage which was used as starting point for a new simulation cycle. Calculations were repeated until the patient reaches the endpoint ESRD or death, while accumulated time (sum of all transition times) is recorded. Bootstrapping was used (100 iterations) to assess reliable point estimates. Simulations were performed for separate patient groups by classifying patients into early, intermediate or advanced stages of DKD, and by different age categories. Our definition of early, intermediate and advanced stage of disease is based on the KDIGO guidelines [16] and displayed in Supplemental Figure S1. Early DKD was defined as egfr>60 ml/min/1.73m 2 and albumin:creatinine ratio (ACR) <30 mg/g, intermediate DKD defined as egfr ml/min/1.73m 2 or ACR mg/g, and advanced DKD as egfr <30 ml/min/1.73m 2 or ACR >300 mg/g. Age categories ranged from 25 to 65 with 5-year intervals. The age distribution for each age category was similar to the age distribution in the trials used to develop the model. Additionally, we assessed the effect of RAS intervention on delaying ESRD in patients who responded to RAS interventions (defined as a regression in albuminuria stage accompanied with at least 30% reduction in albuminuria after 6 months of treatment) and non-responders. Patients with baseline albuminuria levels <15 mg/g were excluded from this analysis because they could not regress in albuminuria stage. 22

24 Chapter 2 For simulation purposes we added an age-specific mortality probability for patients older than 65 years, on top of the mortality probabilities observed in the dataset. This takes into account that as patients age their probability to die increases. These calculations were based on age- and sex-adjusted mortality rates for patients with type 2 diabetes as previously reported (Supplementary Table S1) [17]. Model validation Internal and external validity was assessed by comparing the proportion of events derived from our model with observed proportion of events in the trials. For internal validation, we applied the model to all patients from trials included in the training database. The time to ESRD for each individual was calculated using baseline characteristics of each individual. For external validation we applied the model to the individual patient-level data of clinical trials in diabetes not included in our training dataset: LIFE, SUN-MACRO and ALTITUDE. Their rationale, study design and results have been published elsewhere [18-20]. Additionally, we compared the proportion of ESRD events derived from our model with the observed proportion of ESRD events in trials of which no individual patient data was available. For these trials we used aggregated trial level data. We used this approach for ADVANCE, ACCORD, TREAT and ORIENT. The results and design of these trials have been published elsewhere [9,21-23], and are summarized in Supplementary Table S2. Results Characteristics of patients included in the dataset An overview of the baseline characteristics of included trials are presented in Table 1. In all included datasets, participants were diagnosed with type 2 diabetes were and on average around 60 years of age. Albuminuria levels were in the normoalbuminuric range (N=1209), microalbuminuric range (N=590) and macroalbuminuric range (N=3228). Renal function (egfr) ranged from normal (>90ml/min/1.73m 2 ) to severely impaired (15-30 ml/min/1.73m 2 ). The final dataset included 5027 patients. In this dataset, a total of 628 ESRD events and 576 death events were recorded during follow-up. The majority of deaths (357; 62%) were recorded in patients with egfr <45/ml/min/1.73m 2 and albumin:creatinine ratio 23

25 Early RAS intervention is more beneficial than late intervention in delaying ESRD >300mg/g at baseline. For modeling purposes, we used all available transitions that patients experienced during follow-up, resulting in a median of 551 transitions (interquartile range: ) for each disease stage (Supplemental Figure S1). Table 1. Baseline characteristics of patients in the included clinical trials BENEDICT N=1209 IRMA-2 N=590 RENAAL N=1513 IDNT N=1715 Age (years) 61.9 (8.1) 58.0 (8.2) 60.2 (7.4) 58.9 (7.8) Gender (% male) 53% 68% 63% 66% Systolic BP (mmhg) (14.2) (14.4) (19.3) (19.7) Diastolic BP (mmhg) 87.5 (7.6) 90.1 (9.2) 82.4 (10.5) 86.9 (11.0) egfr (ml/min/1.73m 2 ) 81.2 (15.0) 72.2 (13.8) 39.8 (12.3) 47.3 (17.6) Albuminuria (mg/g) 5.9 [ ] [ ] 1246 [ ] 1500 [ ] HbA1C (%) 5.8 (1.4) 6.9 (1.7) 8.5 (1.6) 8.1 (1.7) Potassium (mmol/l) 4.3 (0.4) 4.7 (0.5) 4.6 (0.5) 4.6 (0.5) LDL cholesterol (mg/dl) (36.1) (40.3) (45.8) (46.5) HDL cholesterol (mg/dl) 46.9 (12.1) 43.6 (11.6) 45.1 (15.1) 42.4 (14.1) Baseline characteristics of all the clinical trials that were included in the model building. Numbers represent mean (sd) unless otherwise indicated. Albuminuria (albumin:creatinine ratio) is calculated as median + interquartile range. Model validation The predicted survival probabilities (with ESRD as endpoint and death as censoring event) corresponded well to the observed probabilities seen in BENEDICT, IRMA-2, RENAAL and IDNT, with predictions being within the 95% confidence intervals of the observed probabilities for almost all years of follow-up with the only exception being the RENAAL trial after 2 years follow-up (Figure 1). We subsequently validated our model using past clinical trials not included in our training database. The predictions from our DKD model showed very good agreement with the observed probabilities of ESRD events in each treatment arm in each trial (Figure 2). The predicted and observed proportion of ESRD events appeared to be closer to the line of identity for trials where individual patient data was available compared to trials with aggregated trial level data. 24

26 Chapter 2 Figure 1. Kaplan Meier plot showing the observed versus simulated renal events (with death as censoring event) over time in the BENDICT, IRMA-2, RENAAL and IDNT studies. For BENEDICT and IRMA-2 a black horizontal line is drawn because no ESRD events were observed in the trials. For RENAAL and IDNT the 95% confidence intervals are the shaded areas for placebo (red) and treatment (blue). Effect of RAS intervention in early, intermediate or advanced stage disease We subsequently assessed the effect of RAS intervention at early, intermediate or advanced stages of DKD. Figure 3 shows that the predicted time to ESRD was 21.4, 10.8 and 4.7 years for patients at early, intermediate, and advanced stage of disease respectively, being on average 60 years of age (the average age in most type 2 25

27 Early RAS intervention is more beneficial than late intervention in delaying ESRD diabetes trials) and receiving placebo treatment. RAS intervention delayed the predicted time to ESRD by 4.2, 3.6 and 1.4 years, respectively (P values < for pairwise comparisons between early, intermediate and advanced). The beneficial effect of RAS intervention in early stages of DKD became more apparent when treatment is initiated at younger age (Table 2). For example, among patients with an average of 45 years, RAS intervention in early, intermediate and advanced stages of disease delayed the median time to ESRD by 5.9 years 4.0, and 1.1 years respectively (P values < for pairwise comparisons between early, intermediate and advanced). Figure 2. The agreement plot shows the observed and simulated renal events for several clinical trials in nephrology. The percentage of events based on simulated data is shown on the Y axis and the percentage of events derived from trials on the X axis. Blue dots indicate that simulations were performed with patient level data. Red squares indicate that simulations were performed with trial level data. The diagonal line shows the line of exact agreement. 26

28 Chapter 2 Effect of treatment response on time to ESRD We finally assessed the impact of treatment response (defined as a >30% reduction in albuminuria and an improvement in albuminuria staging from baseline to 6 months of treatment) on time to ESRD. Again, analyses were performed for treatment initiated in early, intermediate or advanced stages of DKD. As expected, treatment responders benefitted more from treatment than non-responders and this effect was particularly striking when treatment was initiated at early stages of disease (Figure 4). For patients who responded to RAS intervention aged 60 years, treatment in early stages of disease delayed the predicted time to ESRD by 11.8 years and 13.3 years compared to the non-responder subgroup and placebo group, respectively (P<0.001), while the model predicted a delay in ESRD in responders to RAS intervention in intermediate and advanced stages of 4.9 and 3.5 years compared to non-responders and placebo (both P<0.001). Discussion We have developed and validated a model for patients with type 2 diabetes that can accurately simulate DKD progression and assess long-term treatment effects of RAS inhibition from the earliest stages of disease until ESRD. Our model showed that RAS intervention in the earliest stages of disease is most beneficial in delaying ESRD, and that this treatment effect is even more pronounced among younger patients. The beneficial treatment effect was attributed to a large extent to the initial albuminuria lowering response. ESRD was markedly delayed among patients with an initial response in albuminuria whereas non-responders showed only little benefit compared to placebo, highlighting the importance of monitoring albuminuria during RAS intervention. Our model predicted that half of the patients with normoalbuminuria and hypertension remain free of ESRD for approximately 21 years, while RAS intervention delayed this to approximately 26 years, confirming that progression from early stage to ESRD takes decades to manifest. This is in line with other studies that reported similar time frames. For example, the United Kingdom Prospective Diabetes Study (UKPDS) showed that patients with normoalbuminuria take a median of 19 years to develop nephropathy (defined as microalbuminuria or macroalbuminuria), and patients with macroalbuminuria take a median of 9.7 years to reach ESRD, 27

29 Early RAS intervention is more beneficial than late intervention in delaying ESRD suggesting that progression from normoalbuminuria to ESRD takes approximately three decades [24]. The longer time to reach ESRD in the UKPDS model can be attributed to the inclusion of newly diagnosed diabetes population in the UKPDS whereas the normoalbuminuric population in our study had hypertension and a mean diabetes duration of approximately 8 years. An older, retrospective study in patients with type 1 diabetes reported that the onset of renal failure takes on average 21.6 years from diagnosis [25]. Table 2. Treatment effect of RAS intervention on delaying ESRD Delay of ESRD (years) compared to placebo Age at which RAS Disease stage at which RAS treatment is initiated treatment is initiated Early Intermediate Advanced 35 years years years years years years years Numbers indicate years that treatment with RAS inhibition delays ESRD compared to placebo. Results are displayed for different age groups, ranging from on average 35 years to 65 years, and for treatment initiation in early, intermediate or late stages of DKD. The finding that early intervention was particularly fruitful in younger patients raises the question as to why older patients benefit to a lesser extent. Our model showed that progression from early stage to ESRD may take several decades. Many patients have already died from advanced age or from comorbidities by the time they would have reached ESRD, and therefore death likely obscures the beneficial effect of RAS intervention when initiated at advanced age. Large observational studies showed that patients with mild chronic kidney disease are much more likely to die before reaching ESRD [26,27], with substantially larger risks for death instead of ESRD in populations with less severe kidney disease [28,29]. Indeed, our model, which censored patients in case of death, showed that with increasing patient age at treatment initiation the death/esrd ratio increased substantially, especially in patients with less severe kidney disease. 28

30 Chapter 2 Figure 3. Difference in median time to ESRD for treatment versus placebo for intervention starting in early, intermediate or advanced stage of DKD. The presented patient populations are on average 60 and 45 years old. The proportion of survival is presented on the Y axis and the time in years is presented on the X axis. The necessity of investigating the advantages of intervention in early versus advanced stages of DKD for different age groups is prompted by the rapid increase in type 2 diabetes in younger populations. For example, a recent study showed that incidence of type 2 diabetes is increasing dramatically at ages <40 years [30]. Likewise, the incidence of type 2 diabetes is markedly increasing in pediatric and adolescent populations [14,15,31]. We have shown that the benefits of RAS intervention in early DKD stages becomes more apparent at younger age. We also showed that the ultimate treatment effect depends to a large extent on the initial albuminuria response, with more treatment benefit attributed to early intervention for patients classified as responders. Ideally, this should be confirmed in a prospective randomized clinical trial. However, given that the median time to reach ESRD takes 29

31 Early RAS intervention is more beneficial than late intervention in delaying ESRD more than two decades for patients in early DKD stages, it would require unfeasibly large patient populations and follow-up times, which makes it unlikely that such a trial will ever be performed. Figure 4. This figure is similar to Figure 3. In this analysis we assessed the effect of treatment response (Resp; defined as a regression in albuminuria stage accompanied with at least 30% reduction in albuminuria after 6 months of treatment) on time to ESRD. To our knowledge this is the first study that investigated the entire course of proteinuric DKD and compared treatment effect in early, intermediate and advanced stages of disease. A previous study by Palmer et al with a Markov model compared intermediate and late intervention with data from the IRMA-2 and IDNT trials and showed that intervention in intermediate stages delays onset of ESRD compared to intervention in advanced disease stages [32]. However, our model included more disease stages, a larger population, and covered the full range of DKD with smaller gaps between different disease stages thereby increasing precision and power. In 30

32 Chapter 2 addition, our survival analysis enabled us to calculate patient-specific time to event, which is not possible with a Markov model, and uses individual patient characteristics, therefore providing the possibility to determine whether these characteristics modify treatment effect. Our study has limitations. Firstly, our resolution is limited by the number of defined disease stages used to develop the model. Larger numbers of patients, in particular those with low egfr and low albuminuria, will increase the accuracy and precision of the model. Secondly, our model was developed for RAS intervention but is in principle applicable to other drug classes. This however requires validation. Thirdly, the model does not consider improvement of disease stages during simulation. Instead of taking improvement into account, our model assumes patients stay in the same disease stage until worsening is observed. However, the model records time until worsening in albuminuria or egfr stages occurs and takes it into account in the survival analysis. We used this approach to make sure our model does not include unfeasibly large numbers of possible transition directions. In conclusion, we have built a model that is capable of simulating the entire course of DKD. Using this model, we showed that early intervention with RAS inhibitors is more beneficial in delaying ESRD than intervention in later stages. 31

33 Early RAS intervention is more beneficial than late intervention in delaying ESRD Supplement Supplemental Figure S1. Overview of the different disease stages characterized in the model, including definitions for early, intermediate and advanced stages of DKD. Supplementary Table S1. Age-adjusted mortality rates used in the model Age (years) Mortality rate per year Female Male

34 Chapter 2 Supplementary Table S2. Characteristics of the clinical trials used for validation Trial name Inclusion criteria N patients ACCORD type 2 diabetes, HbA1c>7.5mmol/L, cardiovascular risk factors Treatment allocation intensive HbA1c targeting (<6.0mmol/L) vs. conventional therapy Primary endpoint definition first occurrence of nonfatal myocardial infarction or nonfatal stroke or death from cardiovascular causes perindopril and indapamide or type 2 diabetes, matching placebo, composite of ADVANCE history or risk of cardiovascular and intensive HbA1c targeting macrovascular events and a composite of disease (<6.5mmol/L) vs. microvascular events conventional therapy type 2 diabetes and chronic kidney aliskiren vs. first occurrence of a ALTITUDE disease or 8561 placebo on top of cardiovascular or renal cardiovascular RAS treatment event disease type 2 diabetes, previously LIFE untreated or treated stage II III hypertension with 1195* losartan vs. atenolol cardiovascular morbidity and mortality ECG left ventricular hypertrophy NEPHRON-D type 2 diabetes and nephropathy 1448 losartan and lisinopril or matching placebo decline in egfr (>30ml/min/1.73m 2 or >50%), ESRD or death. first occurrence of ORIENT type 2 diabetes and nephropathy 577 olmesartan vs. placebo doubling of serum creatinine, ESRD or death TREAT type 2 diabetes and chronic kidney disease 4038 darbepoetin alfa vs. placebo first occurrence of cardiovascular event, ESRD or death * The overall LIFE population consisted of 9194 participants of whom 1195 had diabetes. Only participants with diabetes were used for validation purposes. 33

35 Early RAS intervention is more beneficial than late intervention in delaying ESRD References (1) Levey AS, Coresh J. Chronic kidney disease. The Lancet 2012; 379(9811): (2) De Jong PE, Navis G, de Zeeuw D. Renoprotective therapy: titration against urinary protein excretion. The Lancet 1999; 354(9176): (3) Metcalfe W. How does early chronic kidney disease progress?: A Background Paper prepared for the UK Consensus Conference on Early Chronic Kidney Disease. Nephrol Dial Transplant 2007; 22(suppl 9): ix26-ix30. (4) Levey AS, de Jong P,E., Coresh J, et al. The definition, classification, and prognosis of chronic kidney disease: a KDIGO Controversies Conference report. Kidney Int 2011; 80(1): (5) Ruggenenti P, Fassi A, Ilieva AP, et al. Preventing Microalbuminuria in Type 2 Diabetes. N Engl J Med 2004; 351(19): (6) Brenner BM, Cooper ME, de Zeeuw D, et al. Effects of Losartan on Renal and Cardiovascular Outcomes in Patients with Type 2 Diabetes and Nephropathy. N Engl J Med 2001;345(12): (7) Lewis EJ, Hunsicker LG, Clarke WR, et al. Renoprotective Effect of the Angiotensin-Receptor Antagonist Irbesartan in Patients with Nephropathy Due to Type 2 Diabetes. N Engl J Med 2001; 345(12): (8) Parving H-H, Lehnert H, Bröchner-Mortensen J, Gomis R, Andersen S, Arner P. The Effect of Irbesartan on the Development of Diabetic Nephropathy in Patients with Type 2 Diabetes. N Engl J Med 2001; 345(12): (9) Imai E, Chan JCN, Ito S, et al. Effects of olmesartan on renal and cardiovascular outcomes in type 2 diabetes with overt nephropathy: a multicentre, randomised, placebo-controlled study. Diabetologia 2011; 54(12): (10) Haller H, Ito S, Izzo JL, et al. Olmesartan for the Delay or Prevention of Microalbuminuria in Type 2 Diabetes. N Engl J Med 2011; 364(10): (11) Makino H, Haneda M, Babazono T, et al. Prevention of Transition From Incipient to Overt Nephropathy With Telmisartan in Patients With Type 2 Diabetes. Diabetes Care 2007; 30(6): (12) Koopman RJ, Mainous AG, Diaz VA, Geesey ME. Changes in Age at Diagnosis of Type 2 Diabetes Mellitus in the United States, 1988 to Ann Fam Med 2005; 3(1): (13) Cheung BMY, Ong KL, Cherny SS, Sham P, Tso AWK, Lam KSL. Diabetes Prevalence and Therapeutic Target Achievement in the United States, 1999 to Am J Med 2009; 122(5): (14) D Adamo E, Caprio S. Type 2 Diabetes in Youth: Epidemiology and Pathophysiology. Diabetes Care 2011; 34(Supplement 2): S161-S165. (15) Dabelea D, Mayer-Davis E, Saydah S, et al. Prevalence of Type 1 and Type 2 Diabetes Among Children and Adolescents From 2001 to JAMA 2014; 311(17): (16) Kidney Disease: Improving Global Outcomes (KDIGO) CKD Work Group. KDIGO 2012 Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease. Kidney inter. Suppl. 2013; 3:

36 Chapter 2 (17) Zhuo X, Zhang P, Barker L, Albright A, Thompson TJ, Gregg E. The Lifetime Cost of Diabetes and Its Implications for Diabetes Prevention. Diabetes Care 2014; 37(9): (18) Parving H-H, Brenner BM, McMurray JJV, et al. Cardiorenal End Points in a Trial of Aliskiren for Type 2 Diabetes. N Engl J Med 2012; 367(23): (19) Packham DK, Wolfe R, Reutens AT, et al. Sulodexide Fails to Demonstrate Renoprotection in Overt Type 2 Diabetic Nephropathy. J Am Soc Nephrol 2012; 23(1): (20) Wachtell K, Olsen MH, Dahlöf B, et al. Microalbuminuria in hypertensive patients with electrocardiographic left ventricular hypertrophy: The LIFE Study. J Hypertens 2002; 20(3): (21) The Action to Control Cardiovascular Risk in Diabetes Study Group. Effects of Intensive Glucose Lowering in Type 2 Diabetes. N Engl J Med 2008; 358(24): (22) The ADVANCE Collaborative Group. Intensive Blood Glucose Control and Vascular Outcomes in Patients with Type 2 Diabetes. N Engl J Med 2008; 358(24): (23) Pfeffer MA, Burdmann EA, Chen C, et al. A Trial of Darbepoetin Alfa in Type 2 Diabetes and Chronic Kidney Disease. N Engl J Med 2009; 361(21): (24) Adler AI, Stevens RJ, Manley SE, Bilous RW, Cull CA, Holman RR. Development and progression of nephropathy in type 2 diabetes: The United Kingdom Prospective Diabetes Study (UKPDS 64). Kidney Int 2003; 63(1): (25) Kussman M.J., Goldstein H, Gleason RE. The Clinical Course of Diabetic Nephropathy. JAMA 1976; 236: (26) Dalrymple L, Katz R, Kestenbaum B, et al. Chronic Kidney Disease and the Risk of End-Stage Renal Disease versus Death. J Gen Intern Med 2011; 26(4): (27) Keith DS, Nichols GA, Gullion CM, Brown J,Smith DH. Longitudinal follow-up and outcomes among a population with chronic kidney disease in a large managed care organization. Arch Intern Med 2004; 164(6): (28) Sud M, Tangri N, Levin A, Pintilie M, Levey AS, Naimark DM. CKD Stage at Nephrology Referral and Factors Influencing the Risks of ESRD and Death. Am J Kidney Dis 2014; 63(6): (29) Packham DK, Alves TP, Dwyer JP, et al. Relative Incidence of ESRD Versus Cardiovascular Mortality in Proteinuric Type 2 Diabetes and Nephropathy: Results From the DIAMETRIC (Diabetes Mellitus Treatment for Renal Insufficiency Consortium) Database. Am J Kidney Dis 2012; 59(1): (30) Holden SE, Barnett AH, Peters JR, et al. The incidence of type 2 diabetes in the United Kingdom from 1991 to Diabetes Obes Metab 2013; 15(9): (31) Pinhas-Hamiel O, Zeitler P. The global spread of type 2 diabetes mellitus in children and adolescents. J Pediatr 2005; 146(5): (32) Palmer AJ, Annemans L, Roze S, et al. Cost-Effectiveness of Early Irbesartan Treatment Versus Control or Late Irbesartan Treatment in Patients With Type 2 Diabetes, Hypertension, and Renal Disease. Diabetes Care 2004; 27(8):

37

38 Chapter 3 Heart failure induced by aleglitazar treatment can be predicted based on short-term response in multiple risk markers Bauke Schievink Diederick Grobbee A. Michael Lincoff Dick de Zeeuw Hiddo Lambers Heerspink On behalf of the AleCardio Steering Committee Submitted for publication

39 Heart failure induced by aleglitazar can be predicted based on short term risk marker response Abstract Introduction: Hospitalization for heart failure (HF) is a common safety problem in clinical drug trials in type 2 diabetes. Novel strategies are needed to mitigate risk for HF and prevent early termination of these trials. We performed a post-hoc analysis of the AleCardio trial to assess whether the increased risk for HF associated with aleglitazar treatment could have been prevented by more stringent baseline criteria, or predicted based on short-term response in risk markers. Methods: We calculated hazard ratios for hospitalization due to HF associated with aleglitazar treatment by restricting the baseline population in the AleCardio trial based on body weight, hemoglobin and NT pro-bnp levels. We also calculated risk for HF based on changes in these risk markers, and by integrating changes in all measured cardiovascular risk markers by using the so-called PRE score. Results: With baseline restrictions based on body weight, hemoglobin and NT pro- BNP the hazard ratio for hospitalization due to HF associated with aleglitazar did not attenuate. For NT pro-bnp we found a strong linear relationship between short-term change and predicted risk for hospitalization due to HF. However, prediction of HF risk was more accurate with using changes in all measured risk markers (PRE score predict risk: +20.5% vs % observed), compared to using only NT pro-bnp or any other single risk marker alone. Discussion: Integrating short-term changes in all measured cardiovascular risk markers predicted HF risk in the AleCardio trial. 38

40 Chapter 3 Introduction Hospitalization for heart failure (HF) is a common safety problem in clinical drug trials in type 2 diabetes leading in some cases to early discontinuation of several clinical drug trials. Recent examples of drugs that increase risk for HF include the Nrf2 activator bardoxolone-methyl, the endothelin antagonist avosentan, the DPP-4 inhibitor saxagliptin and the PPAR agonist rosiglitazone (1-4). A common feature of these drugs is that they induce fluid retention, which increases risk of HF. In order to mitigate HF risk and early termination of future trials, it is important to exclude patients that are at risk for HF. The current inclusion/exclusion strategies are insufficient in mitigating this risk, and therefore new risk assessment strategies are needed. Several risk markers have been proposed as proxies for HF, including body weight, hemoglobin and N-terminal pro-brain natriuretic peptide (NT pro-bnp) (5-7). However, it is unclear whether patients at risk of hospitalization for HF can reliably be identified based on baseline levels of a single or combination of these risk markers, or based on short-term treatment-induced response in these risk markers. We performed a post-hoc analysis of the AleCardio trial in which patients with type 2 diabetes and a recent cardiovascular event were treated with the dual PPAR agonist aleglitazar. The study was terminated early due to futility and a higher incidence of hospitalization for HF in the aleglitazar treatment arm (8). We questioned whether the observed increase in HF incidence due to aleglitazar treatment could have been prevented if patient inclusion had been restricted by first excluding high risk patients on the basis of a single or combination of cardiovascular risk markers at baseline. Secondly, we questioned whether the observed HF risk could have been predicted on the basis of short-term (6 months) changes in risk markers associated with fluid retention (i.e. body weight, hemoglobin or NT pro- BNP), or by using an integrated response score with all measured cardiovascular risk markers. For the latter approach, we used the previously validated PRE score, which integrates the short-term drug effect on all measured risk markers and translates this to drug-induced risk probability of clinical outcomes (9,10). 39

41 Heart failure induced by aleglitazar can be predicted based on short term risk marker response Methods AleCardio trial design and patient population AleCardio (ClinicalTrials.gov identifier: NCT ) was a phase III trial in which 7226 patients with type 2 diabetes were randomized to treatment with either the dual PPAR agonist aleglitazar (150µg daily) or placebo, on top of conventional care. Design and outcome of the trial have been described elsewhere (8). In brief, patients were eligible if they were hospitalized for acute coronary syndrome (ACS), defined as myocardial infarction with or without ST segment elevation or biomarker-negative unstable angina. Randomization took place within 12 weeks of hospital discharge of the index ACS event, in order to allow for stabilization of the clinical condition of the patient. Exclusion criteria included hospitalization for HF currently or in the previous 12 months, peripheral edema, and chronic kidney disease, defined as an estimated glomerular filtration rate (egfr) of less than 45ml/min/1.73m 2. Risk marker selection From the AleCardio dataset we selected all measured risk markers that were previously identified as predictors for cardiovascular disease. Firstly, we made subsets of the AleCardio patient population by excluding those at higher baseline risk of hospitalization for HF, based on known cardiovascular risk markers. Secondly, we assessed short-term (6-month) changes in the risk markers hemoglobin and NT pro-bnp, while body weight was assessed as a 1-month change to specifically capture fluid retention and not changes in body composition. We divided the AleCardio population in quartiles with respect to their short-term response in these risk markers. We also calculated hospitalization for HF risk based on all measured cardiovascular risk markers with the PRE score. For this analysis we included the following markers in addition to body weight, hemoglobin and NT pro-bnp: HbA1c, urinary albumin:creatinine ratio (UACR), systolic blood pressure (SBP), HDL cholesterol, LDL cholesterol, serum albumin, serum calcium and serum potassium. Development of the PRE score We predicted the effect of aleglitazar on HF risk compared to placebo by using the PRE score. The PRE score integrates the effect of short-term changes in multiple 40

42 Chapter 3 risk markers and translates this to a probability of long term clinical outcomes in three distinct steps, and has been described and validated previously (9,10). In brief, the PRE score was first used to establish the relationship between the selected risk markers and the clinical outcome of hospitalization for HF by using baseline patient data and HF events as observed in the AleCardio trial. Secondly, the established risk marker-outcome relationships (with a median follow-up of 2 years) were applied to baseline and month 6 (exception: 1 month for body weight) risk marker levels as observed in AleCardio, in order to establish the risk difference between the two time points. PRE scores were then calculated by subtracting the baseline risk from month 6 correcting for placebo effects. The mean difference between the two time points is the relative risk change conferred by aleglitazar treatment over 2 years. Statistical analysis Variables were expressed as mean (SD) with log transformation for non-normal data. Statistical significance in changes in risk markers between the treatment and placebo group were calculated by ANCOVA adjusted for baseline values of each risk marker. To assess if the HF outcome could be prevented by applying a more stringent baseline selection of participations we assessed the risk of HF in subsets of patients by Cox proportional hazard models. For participants who experienced more than one HF event during follow-up, survival time to the first relevant endpoint was used in each analysis. Participants were censored at their date of death or, for those still alive at the end of follow-up, the date of their last clinic visit before the termination of this study arm. Patients with unknown vital status were censored when they were last known to be alive. To assess whether the observed HF risk could have been predicted on the basis of single short-term risk marker responses we first established the associations between body weight, hemoglobin, and NT pro-bnp changes between baseline and 6 months and HF outcomes by Cox proportional hazard regression. Body weight, hemoglobin, and NT pro-bnp changes were divided into quartiles and entered in the Cox regression model. Each Cox regression model was adjusted for baseline values of each respective risk marker. In an additional analysis we adjusted each Cox proportional hazard model for the following covariates: age, gender, history of HF and baseline hemoglobin, body weight, NT pro-bnp, egfr, systolic blood pressure and HbA1c. 41

43 Heart failure induced by aleglitazar can be predicted based on short term risk marker response To assess the PRE score we established Cox proportional hazard models with all included cardiovascular risk markers at baseline. We used the coefficients for each risk marker in the model as weights for the risk equation for hospitalization for HF. These risk equations were applied to baseline and month 6 values of the selected risk markers as observed in AleCardio to establish a difference in risk for HF for each patient. The mean risk difference, after subtracting the risk difference in the placebo arm, represented the PRE score. A two-tailed p value of 0.05 was used as border for statistical significance. All statistical analyses were performed with R version (R Project for Statistical Computing, Results Baseline characteristics All 7226 randomized patients (n=3616 on aleglitazar, n=3610 on placebo) were included in this post-hoc analysis. Characteristics were well balanced between the aleglitazar and placebo group (Table 1). Mean baseline body weight and hemoglobin values were 83.0 kg and 13.7 g/dl respectively. Median NT pro-bnp was 382 pmol/l. Patient selection at baseline and HF risk We assessed whether the hazard ratio for HF associated with aleglitazar treatment could have been attenuated by restricting the patient population on baseline, prior to initiating treatment, by using a single or combination of risk markers. Restricting the baseline population based on individual risk markers or combination of risk markers did not attenuate the observed hazard ratio of 1.22 (95% confidence interval (CI) 0.94 to 1.59) for hospitalization for HF, although the confidence intervals were wide for some selections (Table 2). Changes in body weight, hemoglobin and NT pro-bnp and HF risk We subsequently assessed whether short-term changes in individual risk markers (i.e. body weight, hemoglobin, and NT pro-bnp) after treatment initiation (up to 6 months) could have predicted HF risk. Relative to placebo, treatment with aleglitazar significantly (P<0.001) increased body weight by 0.81 kg (95% CI: 0.70 to 0.92), NT 42

44 Chapter 3 pro-bnp with 38.3% (95% CI: 32.8 to 44.6), and decreased hemoglobin by 0.65 g/dl (95% CI: 0.59 to 0.71; Figure 1A). Table 1. Baseline characteristics of the patients included in the AleCardio trial. Numbers are reported in mean (SD), unless otherwise indicated. Aleglitazar (N=3616) Placebo (N=3610) Age, years 61 (10) 61 (10) Sex, men (N, %) 2641 (73.1) 2619 (72.5) Race (N, %) White 2427 (67.2) 2391 (66.3) Asian 942 (26.1) 942 (26.1) Body weight, kg 82.9 (18.9) 83.3 (19.1) Hemoglobin, g/dl 13.7 (1.5) 13.7 (1.5) NT pro-bnp, pmol/l (median, IQR) 383 [ ] 378 [ ] HbA1c, % 7.8 (1.7) 7.8 (1.6) SBP, mmhg 128 (17.4) 128 (17.6) Albuminuria, mg/g (median, IQR) 12 [6-38] 12 [6-36] egfr, ml/min/1.73m 2 78 (20.3) 78 (20.4) HDL cholesterol, mg/dl 42 (11) 42 (11) LDL cholesterol, mg/dl 79 (31) 80 (31) There was no linear relationship between the degree of body weight change and the hazard ratio for hospitalization for HF, both in the placebo and aleglitazar group (Figure 2). Only for the highest quartile in the aleglitazar group we found a significant increase in hazard ratio for HF compared to the reference category (HR: 2.22 [95% CI: 1.31 to 3.78]). For hemoglobin, we observed that a decrease after 6 months of therapy was associated with a higher hazard ratio for hospitalization due to HF, both in the placebo (HR: 4.53 [95% CI: 2.23 to 9.20, P<0.001] in first quartile compared to reference category) and aleglitazar group (HR: 2.33 [95% CI: 1.22 to 4.47, P=0.01]) group. Lastly, we observed a strong positive linear association between increases in NT pro-bnp and hazard ratio for hospitalization due to HF (P for trend <0.001; Figure 2). Multivariate adjustments of these associations did not alter the results (Supplementary Figure S1). 43

45 Heart failure induced by aleglitazar can be predicted based on short term risk marker response Table 2. Relationship between baseline levels of body weight, hemoglobin and NT pro-bnp versus hazard ratio for HF associated with aleglitazar treatment. Select only patients with: HR for aleglitazar # events No restriction 1.22 ( ) 222 NT pro-bnp < ( ) 13 NT pro-bnp < ( ) 34 NT pro-bnp < ( ) 84 egfr > ( ) 135 No CV history 1.21 ( ) 159 Weight < ( ) 154 Hemoglobin < ( ) 61 Combinations of risk markers NT pro-bnp <500 & weight < ( ) 13 NT pro-bnp <500 & egfr > ( ) 22 NT pro-bnp <500 & no CV history 1.38 ( ) 26 Changes in multiple cardiovascular risk markers and HF risk We finally assessed whether integrating changes in single cardiovascular risk markers into a PRE score could yield a better prediction hospitalization for HF risk due to aleglitazar treatment. In addition to body weight, hemoglobin and NT pro- BNP, treatment with aleglitazar resulted in a significant decrease in HbA1c and systolic blood pressure, and an increase in HDL cholesterol, LDL cholesterol, albumin and calcium, as shown in Figure 1B. Figure 3 shows the individual contribution of the aleglitazar-induced changes in each cardiovascular risk marker on risk change of hospitalization for HF. Changes in single risk markers underestimated the actual observed effect of aleglitazar on heart failure. However, integrating the changes in all measured risk markers with the PRE score provided the most accurate prediction of the HF risk, with a predicted risk increase of 20.5% (16.4% to 24.7%), close to the observed relative risk increase of 21.4% in the AleCardio trial. Changes in NT pro-bnp provided the largest individual contribution to the observed risk increase for HF associated with aleglitazar treatment, with a predicted risk increase of +16.6% (95% CI: +13.5% to +19.7). 44

46 Chapter 3 Figure 1. Panel A: risk marker responses after 6 months of aleglitazar or placebo treatment (1 month for body weight) in AleCardio. All aleglitazar vs. placebo comparisons are P< Panel B: other risk marker responses in AleCardio. All aleglitazar vs. placebo comparisons are P<0.001 with the exception of potassium (P=0.17). Figure 2. Relationship between response to aleglitazar or placebo on body weight, hemoglobin and NT pro BNP versus hazard ratios for HF. 45

47 Heart failure induced by aleglitazar can be predicted based on short term risk marker response Discussion The AleCardio trial revealed safety problems due to increased risk of hospitalization for HF, which was among the reasons for early termination of the trial. We found that applying more stringent inclusion criteria using available cardiovascular risk markers could not have prevented the observed outcome. However, using drug-induced changes in multiple risk markers during the first months of treatment and integrating them with the PRE score provided an accurate prediction of the effect of aleglitazar on hospitalization for HF. Use of this score may be a way forward to obtain an accurate estimate of a drug s efficacy and safety during early stage drug development and avoid drug failures during late stage development as seen with recent clinical trials. Figure 3. Relative risk for HF as calculated by the response in individual cardiovascular risk markers and by integrating the response into the PRE score. Nowadays, risk assessment for identifying patients at risk for HF is performed by applying restrictions for enrolment in clinical trials such as excluding patients who have been hospitalized previously with HF or have signs of sodium/water retention 46

48 Chapter 3 (i.e. edema). We have tried to narrow the inclusion and exclusion criteria of the AleCardio population to ameliorate HF risk based on more stringent entry criteria using baseline single or combinations of risk markers. However, this approach did not attenuate the HF risk of aleglitazar treatment, although the confidence intervals were wide in some subpopulations. This finding is in contrast with an analysis recently performed in the BEACON trial with bardoxolone methyl. That trial was also prematurely terminated due to HF. A post-hoc analyses in BEACON revealed that using more stringent entry criteria by excluding patients with a high BNP (>200pg/mL) and a history of HF mitigated HF risk induced by bardoxolone methyl (11). The difference between our findings and those in the BEACON trial may be attributed to the different populations enrolled in the trials and the different drugs with different mechanisms of action. Apparently, in the AleCardio trial patients first needed to be exposed to aleglitazar in order to assess who is at risk for HF. Indeed, when we tested individual risk markers associated with fluid retention, we found that a decrease in hemoglobin is associated with increased HF risk. Furthermore, there was a consistent relationship between drug-induced change in NT pro-bnp and hazard ratio for HF. NT pro-bnp is predominantly produced by cardiac myocytes in response to myocardial stretching, which is a sign of increased fluid levels and increased preload, either as a result of increased sodium/water retention or impaired functioning of the cardiac ventricles (12-14). This physiological phenomenon suggests that NT pro-bnp may indeed be used as a predictor for fluid retention and risk for HF. In support, previous studies have showed that levels of NT pro-bnp can be used in the diagnosis and prognosis for patients with HF (15). Whether treatmentinduced changes in NT pro-bnp can also be used to predict the risk of hospitalization for HF is less well established. Our study adds that NT pro-bnp can also be used to monitor treatment effects, and that NT pro-bnp is a better predictor of HF risk than other individual risk markers, such as body weight and hemoglobin. Next to its effect on fluid parameters, aleglitazar has effects on multiple other cardiovascular risk markers. Changes in these risk markers, such as albuminuria, blood pressure and HbA1c, have been associated with HF (16-18). Responses in these multiple risk markers may vary within individuals. For example, HbA1c levels may fall and NT pro-bnp levels may rise or the other way around. Integrating these short term changes in multiple risk markers in response to therapy may therefore 47

49 Heart failure induced by aleglitazar can be predicted based on short term risk marker response provide a more accurate prediction of HF risk compared to using changes in single risk markers alone. Indeed, in the present study we showed that integrating short term risk marker changes into a PRE score provided the most accurate assessment of HF risk. These results are in line with previous analyses, which showed that the PRE score is more accurate than single risk markers alone in predicting the clinical outcomes of the direct renin inhibitor aliskiren and of the angiotensin receptor blockers losartan and irbesartan (9,10). Identification and exclusion of patients at risk of HF before randomization into a long term trial by characterizing short-term responses in multiple risk markers during a so-called enrichment period could prevent early termination of clinical trials due to safety concerns. Consequently, it may increase the chance of finding a beneficial effect on the clinical outcome of interest. Such an approach with an enrichment design is used for the currently ongoing SONAR phase III trial (ClinicalTrials.gov identifier: NCT ), in which patients with type 2 diabetes and nephropathy are subjected to a six-week enrichment phase in which their response to the endothelin antagonist atrasentan is determined. In a phase II trial atrasentan treatment decreased albuminuria, but also increased body weight and decreased hemoglobin (19). Therefore in the SONAR trial only patients with a >30% decrease in the targeted parameter albuminuria and without unacceptable risk increase for HF (rise in body weight <3 kg or BNP <300 pg/ml) after the enrichment phase will proceed to randomization to either long-term treatment with atrasentan or placebo (on top of conventional care) (20). This study has limitations. First, there was a strong initial reduction in NT pro- BNP after start of the trial. This is likely due to a high baseline level in all patients as a result of the recent ACS event prior to enrollment in the study. Secondly, the increase in HDL cholesterol seen with aleglitazar treatment increased risk for hospitalization due HF. This is somewhat surprising as an increase in HDL cholesterol is generally associated with less cardiovascular outcomes. It is unclear why the HDL component associated with increased HF risk, although our model does not assume causality and we cannot exclude the possibility of confounding. Thirdly, the limited number of HF events resulted in wide confidence intervals of the aleglitazar treatment effect on HF in some subpopulations after excluding patients at risk on the basis of baseline cardiovascular risk markers. The small number of HF events also precluded testing of all possible combinations of risk markers selections. 48

50 Chapter 3 Finally, we acknowledge that this is a post-hoc analysis of a clinical trial with all its inherent limitations. The results can therefore only be interpreted as hypothesis generating. In conclusion, integrating short-term changes in all known and measured cardiovascular risk markers provided the most accurate prediction of the effect of aleglitazar on HF, compared to single risk markers alone. This supports using all available risk markers to monitor the drug-induced responses in clinical trials order to predict treatment-related HF risk. A randomized controlled clinical trial design in which patients are exposed to the drug of interest before randomization in order to identify individuals at risk of HF based on changes in multiple risk markers may facilitate clinical trial conduct and may prevent early termination of clinical trials due to adverse effects. Acknowledgements We acknowledge the supportive role of all AleCardio investigators, support staff, and participating patients. 49

51 Heart failure induced by aleglitazar can be predicted based on short term risk marker response Supplement Supplemental Figure S1. Relationship between response to aleglitazar or placebo on body weight, hemoglobin and NT pro-bnp versus hazard ratios for HF, calculated by a multivariate Cox regression model. 50

52 Chapter 3 References (1) de Zeeuw D, Akizawa T, Audhya P, Bakris GL, Chin M, Christ-Schmidt H, et al. Bardoxolone Methyl in Type 2 Diabetes and Stage 4 Chronic Kidney Disease. N Engl J Med 2013; 369(26): (2) Mann JFE, Green D, Jamerson K, Ruilope LM, Kuranoff SJ, Littke T, et al. Avosentan for Overt Diabetic Nephropathy. J Am Soc Nephrol 2010; 21(3): (3) Komajda M, McMurray JJV, Beck-Nielsen H, Gomis R, Hanefeld M, Pocock SJ, et al. Heart failure events with rosiglitazone in type 2 diabetes: data from the RECORD clinical trial. Eur Heart J 2010; 31(7): (4) Scirica BM, Braunwald E, Raz I, Cavender MA, Morrow DA, Jarolim P, et al. Heart Failure, Saxagliptin and Diabetes Mellitus: Observations from the SAVOR - TIMI 53 Randomized Trial. Circulation 2014;130(18): (5) Hoekman J, Lambers Heerspink HJ, Viberti G, Green D, Mann JFE, de Zeeuw D. Predictors of Congestive Heart Failure after Treatment with an Endothelin Receptor Antagonist. Clin J Am Soc Nephrol 2014; 9(3): (6) Richards M, Troughton RW. NT-proBNP in heart failure: therapy decisions and monitoring. Eur J Heart Fail 2004; 6(3): (7) Tang Y, Katz SD. Anemia in Chronic Heart Failure: Prevalence, Etiology, Clinical Correlates, and Treatment Options. Circulation 2006; 113(20): (8) Lincoff A, Tardif J, Schwartz GG, et al. Effect of aleglitazar on cardiovascular outcomes after acute coronary syndrome in patients with type 2 diabetes mellitus: The AleCardio randomized clinical trial. JAMA; 311(15): (9) Smink P, Hoekman J, Grobbee D, Eijkemans M, Parving H, Persson F, et al. A prediction of the renal and cardiovascular efficacy of aliskiren in ALTITUDE using short-term changes in multiple risk markers. Eur J Prev Cardiol 2014; 21(4): (10) Smink PA, Miao Y, Eijkemans MJC, Bakker SJL, Raz I, Parving H, et al. The Importance of Short-Term Off-Target Effects in Estimating the Long-Term Renal and Cardiovascular Protection of Angiotensin Receptor Blockers. Clin Pharmacol Ther 2014; 95(2): (11) Chin MP, Wrolstad D, Bakris GL, Chertow GM, de Zeeuw D, Goldsberry A, et al. Risk Factors for Heart Failure in Patients With Type 2 Diabetes Mellitus and Stage 4 Chronic Kidney Disease Treated With Bardoxolone Methyl. J Card Fail 2014; 20(12): (12) Bruggink AH, de Jonge N, van Oosterhout MFM, Van Wichen DF, de Koning E, Lahpor JR, et al. Brain Natriuretic Peptide is Produced Both by Cardiomyocytes and Cells Infiltrating the Heart in Patients with Severe Heart Failure Supported by a Left Ventricular Assist Device. J Heart Lung Transplant 2006; 25(2): (13) Hall C. Essential biochemistry and physiology of (NT-pro)BNP. Eur J Heart Fail 2004; 6(3): (14) Houben AJHM, van der Zander K, de Leeuw PW. Vascular and renal actions of brain natriuretic peptide in man: physiology and pharmacology. Fundam Clin Pharmacol 2005; 19(4):

53 Heart failure induced by aleglitazar can be predicted based on short term risk marker response (15) Taylor CJ, Roalfe AK, Iles R, Hobbs FDR. The potential role of NT-proBNP in screening for and predicting prognosis in heart failure: a survival analysis. BMJ Open ; 4(4). (16) Gerstein HC, Mann JE, Yi Q, et al. Albuminuria and risk of cardiovascular events, death, and heart failure in diabetic and nondiabetic individuals. JAMA 2001; 286(4): (17) UK Prospective Diabetes SG. Tight blood pressure control and risk of macrovascular and microvascular complications in type 2 diabetes: UKPDS 38. BMJ 1998; 317(7160): (18) Parry HM, Desmukh H, Levin D, Van Zuydam N, Elder DHJ, Morris AD, et al. Both High and Low HbA1c Predict Incident Heart Failure in Type 2 Diabetes Mellitus. Circ Heart Fail 2015; 8(2): (19) de Zeeuw D, Coll B, Andress D, Brennan JJ, Tang H, Houser M, et al. The Endothelin Antagonist Atrasentan Lowers Residual Albuminuria in Patients with Type 2 Diabetic Nephropathy. J Am Soc Nephrol 2014; 25(5): (20) Schievink B, de Zeeuw D, Smink PA, Andress D, Brennan JJ, Coll B, et al. Prediction of the effect of atrasentan on renal and heart failure outcomes based on short-term changes in multiple risk markers. Eur J Prev Cardiol doi: /

54 Chapter 4 Prediction of the effect of atrasentan on renal and heart failure outcomes based on short-term changes in multiple risk markers Bauke Schievink Dick de Zeeuw Paul A Smink Dennis Andress John J Brennan Blai Coll Ricardo Correa-Rotter Fan Fan Hou Donald Kohan Dalane W. Kitzman Hirofumi Makino Hans-Henrik Parving, Vlado Perkovic, Giuseppe Remuzzi Sheldon Tobe Robert Toto Jarno Hoekman Hiddo Lambers Heerspink European Journal of Preventive Cardiology, 2015.

55 Prediction of the effect of atrasentan based on short term changes in multiple risk markers Abstract Background: A recent phase II clinical trial (RADAR/JAPAN) showed that the endothelin A receptor antagonist atrasentan lowers albuminuria, blood pressure, cholesterol, hemoglobin, and increases body weight in patients with type 2 diabetes and nephropathy. We previously developed an algorithm, the PRE score, which translates short-term drug effects into predictions of long-term effects on clinical outcomes. Design: We used the PRE score on data from the RADAR/JAPAN study to predict the effect of atrasentan on renal and heart failure outcomes. Methods: We performed a post-hoc analysis of the RADAR/JAPAN randomized clinical trials in which 211 patients with type-2 diabetes and nephropathy were randomly assigned to atrasentan 0.75mg/day, 1.25mg/day, or placebo. A PRE score was developed in a background set of completed clinical trials using multivariate Cox models. The score was applied to baseline and week-12 risk marker levels of RADAR/JAPAN participants, to predict atrasentan effects on clinical outcomes. Outcomes were defined as doubling serum creatinine or end-stage renal disease and hospitalization for heart failure. Results: The PRE score predicted renal risk changes of -23% and -30% for atrasentan 0.75 and 1.25 mg/d, respectively. PRE scores also predicted a small nonsignificant increase in heart failure risk for atrasentan 0.75 and 1.25 mg/d (+2% vs. +7%). Selecting patients with >30% albuminuria response from baseline (responders) improved renal outcome to almost 50% risk reduction, whereas nonresponders showed no renal benefit. Conclusions: Based on the RADAR/JAPAN study, with short-term changes in risk markers, atrasentan is expected to decrease renal risk without increased risk of heart failure. Within this population albuminuria responders appear to contribute to the predicted improvements, whereas non-responders showed no benefit. The ongoing hard outcome trial (SONAR) in type 2 diabetic patients with >30% albuminuria response to atrasentan will allow us to assess the validity of these predictions. 54

56 Chapter 4 Introduction Despite the availability of existing proven therapies to slow progression of kidney disease, diabetic nephropathy remains associated with a high risk of end-stage renal disease (ESRD) (1,2). Endothelin-A receptor antagonists (ERA) have been proposed as an addition to blockade of the renin-angiotensin-aldosterone system (RAAS) to delay progression of kidney disease (3-6). Clinical trials with the ERAs avosentan, darusentan and sitaxsentan have shown reductions in albuminuria (7-9). The recent Reducing Residual Albuminuria in Subjects With Diabetes and Nephropathy With AtRasentan trial and an identical trial in Japan (RADAR/JAPAN trials) showed that the addition of the ERA atrasentan in doses of 0.75 and 1.25 mg/day on top of angiotensin-converting enzyme inhibitor (ACEi) or angiotensin receptor blocker (ARB) therapy results in approximately 40% reduction in albuminuria in patients with type 2 diabetes and nephropathy (10). These data suggest that ERAs may confer renoprotective effects, although this has not been confirmed in a long-term hard outcome trial. In the RADAR/JAPAN trials atrasentan not only decreased albuminuria, but also blood pressure and LDL cholesterol (10). These additional effects may enhance the beneficial effect of atrasentan. In contrast, atrasentan dose-dependently increased body weight reflecting sodium retention, which induced edema and may increase the risk of congestive heart failure (CHF) as seen with avosentan. The sodium-retaining effect may thus negatively influence the benefit-risk ratio of ERAs, and has indeed led to early discontinuation of other clinical trials with ERAs (11,12). Because ERAs have effects on multiple renal or cardiovascular risk markers, known risk markers should ideally be measured and integrated in order to reliably predict the long-term effect of atrasentan on clinical outcomes, as opposed to using a single marker (such as albuminuria) alone. We recently developed and validated an algorithm, the multiple Parameter Response Efficacy (PRE) score, that predicts longterm risk of clinical outcomes based on short-term effects of drugs on multiple risk markers (13,14). The PRE score was developed because single drugs have effects on multiple renal/cardiovascular risk markers, and each of these effects may alter the ultimate renal/cardiovascular outcome (15). The aim of the current study is to apply the PRE score to data from the RADAR/JAPAN trials to predict the potential long- 55

57 Prediction of the effect of atrasentan based on short term changes in multiple risk markers term effect of the ERA atrasentan on renal and heart failure outcomes. In addition, the benefit-risk ratio for the 0.75 and 1.25 mg/day atrasentan doses was established. Methods Principles of the PRE score model We predicted the effect of atrasentan on renal and heart failure outcomes by using the PRE score. This score translates the effect of short-term drug-induced changes in multiple risk markers to long term outcomes as previously described (13,14). In brief, first we established the relationship between risk markers and clinical outcomes in a background population of completed clinical trials with similar patient characteristics as the RADAR/JAPAN trials (NCT and NCT ). We matched our background population to be able to calculate risk marker outcome relationships that are representative of patients included in the RADAR study. The background population was selected from completed clinical trials including the Reduction of Endpoints in NIDDM with the AII Antagonist Losartan (RENAAL), Irbesartan Diabetic Nephropathy Trial (IDNT), the ALiskiren Trial In Type 2 Diabetes Using CarDiorenal Endpoints (ALTITUDE) and The Avosentan on Time to Doubling of Serum Creatinine, End Stage Renal Disease or Death (ASCEND) trial. Because patients in the RADAR/JAPAN trials used maximum ACEi or ARB, we selected all patients from the background datasets that used ACEIs or ARBs for at least six months. Secondly, the calculated risk marker-outcome relationships, established in the background population, with a median follow-up of almost 3 years, were applied to the baseline and week-12 risk marker measurements in the atrasentan 0.75 mg/day and 1.25 mg/day treatment arm of the RADAR/JAPAN trials to predict individual patients 3-year renal (defined as a composite of sustained doubling of serum creatinine or ESRD) or heart failure risk (defined as hospitalization due to CHF) at both time points. These endpoints were recorded in the background database and were pre-specified and adjudicated by an independent endpoint committee using rigorous and standardized definitions. Thirdly, we calculated the mean of the difference in the predicted risk at baseline and week 12, adjusted for the mean of the difference in predicted risk in the placebo arm. This represented the PRE score and 56

58 Chapter 4 indicated the 3-year renal or heart failure risk change conferred by atrasentan. In summary, the background database was used to establish the association between multiple risk markers and clinical outcomes using a population which was similar to the RADAR/JAPAN population. Based on the risk marker outcome associations in the background database we predicted renal/heart failure risk change by considering the changes in multiple risk markers as observed in the RADAR/JAPAN trials. The model assumes that the predictive ability of short-term risk marker changes are independent of the drug that induces the change. As such the model is applicable to different interventions as long as all relevant risk marker changes are measured and included. We also applied the risk score to a subset of responders (>30% reduction in albuminuria) and a subset of non-responders (<30% albuminuria reduction) in the RADAR/JAPAN trials to assess whether enriching the patient population (only including responders after a run-in period) would improve the benefit-risk profile of atrasentan. Risk marker selection in the RADAR/JAPAN trials All parameters that were measured in the intention-to-treat population of the RADAR/JAPAN trials and were previously identified as risk markers for cardiovascular or renal clinical outcomes were used for analysis: urine albumin-tocreatinine ratio (albuminuria), body weight, systolic blood pressure (SBP), hemoglobin (Hb), albumin, calcium, HbA1c, LDL cholesterol, HDL cholesterol, serum potassium, phosphate and uric acid. Albuminuria was measured from the geometrical mean of 3 first morning void urine samples in a central laboratory in the United States (Quest Diagnostics Clinical Trials, Valencia, CA, USA) or Japan (BML, Inc., Saitama, Japan). We included all available risk markers to capture all measured short-term effects of atrasentan. PRE scores were calculated for subjects in RADAR/JAPAN in whom all risk markers were measured at baseline and follow-up. Changes in body weight were used as a proxy for the sodium-retaining effects of atrasentan, which likely represents a different heart failure risk than other forms of body weight change (such as fat deposition). Because no data on fluid-specific body weight changes was present in the RENAAL, IDNT and ALTITUDE trial, we included data from the ASCEND trial in which patients were treated with the ERA avosentan. 57

59 Prediction of the effect of atrasentan based on short term changes in multiple risk markers From the ASCEND data we could assess the direct association between ERAinduced body weight changes and renal and heart failure outcomes. Statistical analysis Data are presented as means and standard deviation or counts and percentages. Risk markers were handled as continuous variables. Changes in risk marker levels at week 12 between treatment arms were tested with ANCOVA adjusted for baseline values and Tukey post-hoc tests for pairwise comparisons. A Cox proportional hazards model was used to estimate the coefficients and hazard ratios associated with each risk marker for the first recorded renal or heart failure event. Non-normally distributed data was log-transformed. The regression coefficients for each risk marker were taken and used as weights for the risk equation for renal and heart failure outcomes. The risk equations for renal and heart failure outcomes were applied to the risk markers observed in the RADAR/JAPAN trials at baseline and week 12 to calculate renal and heart failure risk at both time points. The mean difference in risk between the two time points, after subtracting the mean risk difference between both time points in the placebo arm, represented the PRE score for each outcome. We performed additional analyses with a simulated range of atrasentaninduced albuminuria changes and body weight changes. We simulated the albuminuria response because it is an important predictor for renal outcomes (16). We simulated body weight change because it is a strong predictor for heart failure induced by fluid retention (17). Simulations were performed by shifting the distribution of the albuminuria and body weight response. This sensitivity analysis was performed to take into account that atrasentan may have different effects on these risk markers in the long-term hard outcome trial. A two-tailed p value of <0.05 indicated statistical significance. All statistical analyses were conducted with R version (R Project for Statistical Computing, 58

60 Chapter 4 Results The characteristics of all subjects included in the calculations are presented in Table 1. Out of the 211 subjects in the RADAR/JAPAN dataset, 164 (78%) had a complete risk marker profile at baseline and follow-up. Of the 164 subjects derived from the RADAR/JAPAN dataset, 119 were randomized to atrasentan treatment (59 patients 0.75mg/d and 60 patients 1.25mg/d). A total of 63 (52.9%) patients were classified as responders (27 patients 0.75mg/d, 36 patients 1.25mg/d), while 56 (47.1%) patients were classified as non-responders (32 patients 0.75mg/d, 24 patients 1.25mg/d). Table 1. Baseline characteristics of subjects in the background and RADAR/JAPAN dataset. Numbers indicate mean (SD), unless otherwise specified. Background (N=2466) RADAR/JAPAN Complete cases (N=164) Age, years 61 (9) 65 (9) Males N (%) 1630 (66) 124 (76) Race N (%) White 1328 (54) 65 (40) Black 224 (9) 21 (13) Asian 631 (26) 74 (45) Other 283 (11) 4 (2) Systolic blood pressure, mmhg 142 (18) 137 (14) Weight, kg 83 (20) 86 (21) egfr (ml/min/1.73m 2 ) 41.0 (13) 49.3 (14) HbA1c, % 8.1 (1.8) 7.4 (1.4) Hb, (g/dl) 12.5 (1.9) 12.9 (1.7) LDL cholesterol, mg/dl 114 (43) 93 (35) HDL cholesterol, mg/dl 45 (14) 47 (13) Serum K+, meq/l 4.7 (0.6) 4.5 (0.5) Phosphate, mg/dl 3.9 (0.8) 3.6 (0.6) Serum albumin, g/dl 4.0 (0.4) 4.0 (0.4) Calcium, mg/dl 9.2 (0.5) 9.2 (0.5) Uric acid, mg/dl 7.1 (1.8) 7.7 (1.8) UACR (median mg/g, IQR) 1026 ( ) 863 ( ) Abbreviations: SBP, systolic blood pressure; UACR, urinary-albumin-to-creatinine ratio (albuminuria); Hb, hemoglobin; K+, potassium; IQR, interquartile range. 59

61 Prediction of the effect of atrasentan based on short term changes in multiple risk markers A total of 2466 cases were used to calculate risk marker-outcome relationships in the background dataset. In the background dataset 331 patients reached a renal event and 130 patients a hospitalization for heart failure event. Short term risk marker changes Figure 1 shows changes in risk markers after treatment with placebo, atrasentan 0.75 mg/d, or atrasentan 1.25 mg/d for the overall RADAR/JAPAN population. As previously shown, atrasentan at doses of 0.75 mg/d and 1.25 mg/d induced a significant reduction in urinary albumin excretion of approximately 40%. Atrasentan also decreased blood pressure, hemoglobin, LDL cholesterol, and increased body weight (Figure 1). By restricting the population to responders (>30% albuminuria reduction), we found a mean decrease in albuminuria of approximately 60% for both the 0.75 mg/d and 1.25 mg/d atrasentan dose, respectively. Non-responders had no significant change in albuminuria (Supplemental Figure S1). Changes in risk markers other than albuminuria were similar between responders and non-responders (Supplemental Figure S1), suggesting that responses in other risk markers were independent of the albuminuria response. Predicted treatment effect Based on the albuminuria-lowering effect of atrasentan 0.75 mg/d and 1.25 mg/day alone, we predict a relative risk reduction for renal events of 31% and 37% respectively. However, atrasentan induced short-term changes in body weight and hemoglobin could imply an increase in renal risk (Figure 2). The PRE score, that integrates the effect of atrasentan on all risk markers, indicated a relative renal risk change of -23% (95% CI: -47% to +1%) for the 0.75 mg/d dose and -30% (-55% to - 6%) for the 1.25 mg/d dose (Figure 2). For the heart failure endpoint, the PRE score indicated a slightly higher risk increase for the 1.25 mg/d dose (+7%; -13% to +27%) compared to the 0.75 mg/d dose (+2%; -16% to +20%) (Figure 3). Changes in body weight and Hb were the main contributors to the adverse effect prediction for heart failure. Results were similar when missing values in risk markers of RADAR/JAPAN subjects were imputed. In the RADAR/JAPAN trials, we observed a large variability in albuminuria response to atrasentan (5 th to 95 th percentile -74.8% to +48.4%). When we restricted 60

62 Chapter 4 Figure 1. Overview of the changes (mean + 95% CI) in risk markers in RADAR/JAPAN in the placebo, atrasentan 0.75 mg and atrasentan 1.25 mg group after 12 weeks of follow-up. Results are shown for the total population. * P<0.05, ** P<0.001 versus placebo. Abbreviations: SBP, systolic blood pressure, Hb, hemoglobin. the population to only albuminuria responders (>30% albuminuria reduction), the estimated renal risk reduction conferred by atrasentan 0.75 mg/d was -47% (-71% to -23%), with similar results in the 1.25 mg/d group (-47%, -71% to -22%; Figure 2). However, the PRE score for heart failure in the responder group predicted lower risk for the 0.75mg/d dose (-9%, -29% to +11%) than with the 1.25mg/d dose (+5%, - 19% to +29%; Figure 3). For non-responders, the PRE score predicted no renal 61

63 Prediction of the effect of atrasentan based on short term changes in multiple risk markers benefit for both doses: +4% (-22% to +31%) and +4% (-31% to +39%; Figure 2) for 0.75 mg/d and 1.25 mg/d respectively. For heart failure the estimates were +12% (- 10% to +33%) and +10% (-17% to +36%; Figure 3), respectively. PRE score predictions were similar after imputation of missing risk marker values at baseline and week 12 in the RADAR/JAPAN trials (Supplementary table S1). We finally assessed whether we could identify baseline determinants of a favorable PRE score. In univariate and multivariate analyses, none of the baseline risk markers showed an association with the PRE scores of each individual. Figure 2. Predicted risk change for renal outcomes for the total, responder and non-responder population based on changes in single risk markers and the integrated effect of all risk markers. Bars indicate percentage mean change in relative risk induced by treatment corrected for placebo + 95% confidence interval (CI). 62

64 Chapter 4 Discussion Using a multiple parameter response efficacy (PRE) score and the short-term response data to atrasentan, we predicted that long-term treatment with atrasentan on top of ACEi or ARB therapy would reduce renal risk without a significant increase in heart failure risk. However, confidence intervals for the heart failure outcome were wide, therefore leaving a certain degree of uncertainty. The albuminuria response is by far the most important contributor to the potential renoprotective effect. When the calculations were restricted to a responder subset defined as a >30% decrease in albuminuria, the PRE score showed a significant further reduction in predicted renal risk with no apparent dose difference. Similar to the overall population, the risk of heart failure among albuminuria responders was smaller with the 0.75 compared to the 1.25 mg/d dose of atrasentan, suggesting that renoprotective effects without significant risk of heart failure may be achieved with atrasentan 0.75 mg/d in this population. Proper dose selection is a key design element for any clinical trial and particularly for drugs with a narrow therapeutic index such as ERAs. The maximum recommended therapeutic dose is determined from pharmaceutical dose-response trials and represents the dose of the pharmaceutical agent with the optimal benefitrisk ratio. Knowledge about the dose-response relation of a drug is important because it provides information on the dose beyond which no additional benefit is expected or side effects become unacceptable (18). Past experience teaches us that selecting the optimal dose in clinical studies for ERAs is problematic. A phase 2 study with the ERA avosentan showed dose-dependent reductions in albuminuria with a maximum anti-albuminuric dose of 10 mg/day (7). Avosentan at doses beyond 10 mg/day did not further decrease albuminuria and caused a dose-dependent increase in body weight and fluid retention. Despite these findings avosentan at doses of 25 mg/day and 50 mg/day was tested in a hard outcome trial, that was terminated early due to excess heart failure and mortality in the avosentan treatment arms (11). This example illustrates the importance of carefully conducted doseranging studies in early stages of drug development. If validated in a full-scale trial, the PRE score can aid in the selection of the optimal dose and thereby improve clinical trial design, because it takes all known risk markers into account and 63

65 Prediction of the effect of atrasentan based on short term changes in multiple risk markers integrates them into a composite risk score. Our analysis shows that among albuminuria responders, atrasentan 0.75 mg/d and 1.25 mg/d have a similar predicted protective effect on renal outcomes. We predicted a slightly higher risk with the 1.25 vs mg/day dose of atrasentan on heart failure but the wide confidence intervals around the predicted effect preclude definitive conclusions regarding dose selection. In combination with results of additional pharmacokinetic and pharmacodynamic analyses, atrasentan 0.75 mg/d is selected for use in a long-term hard outcome trial. Figure 3. Predicted risk change for heart failure for the total, responder and non-responder population based on changes in single risk markers and the integrated effect of all risk markers. Bars indicate percentage mean change in relative risk induced by treatment corrected for placebo + 95% confidence interval (CI). 64

66 Chapter 4 Another important aspect to consider when designing a new clinical trial is the response to treatment and enrichment of the trial population with treatment responders. Our results indicate that the reduction in albuminuria was the most important contributor to the renoprotective prediction of atrasentan. In fact, the current study predicted that patients with >30% albuminuria reduction (responders) would have a clear protective benefit whereas those with less albuminuria reduction (non-responders) showed no predicted benefit. These findings suggest that randomizing the non-responders in a hard outcome trial would not only increase the total number of patients needed to be enrolled in such a trial, but would also expose these patients to long-term atrasentan treatment, potentially without any likelihood of benefit. A hard outcome study enriching the clinical trial population with albuminuria responders, by exposing all patients to atrasentan during an enrichment period prior to randomization could enhance the likelihood of detecting a renoprotective treatment effect. However, this requires validation in a prospective randomized controlled trial. Another advantage of an enrichment period is that patients who experience side effects, such as fluid retention, can be identified during this period and can be excluded from randomization. Selection of patients for a clinical trial based on the response to the drug thus maximizes the beneficial effect and minimizes exposure of the intervention to patients in whom it may be harmful (19). This approach mimics daily practice since in daily clinical care the dose and type of drugs are adjusted based on the response of the patients to the drug. However due to the stringent selection criteria of the enriched randomized population, the generalizability of the results to a broader population with type 2 diabetes and nephropathy is limited when conducting an enrichment trial. Such an enrichment design is applied in the Study Of Diabetic Nephropathy With Atrasentan (SONAR, Clinical Trial identifier: NCT ) hard outcome trial with atrasentan. SONAR will enroll patients with similar characteristics to those in RADAR/JAPAN (i.e. type 2 diabetes and nephropathy). Eligible patients will proceed to a 6-week enrichment period. The aim of this enrichment period is to determine albuminuria response as well as safety. After completion of the 6-week enrichment period, patients with a response in albuminuria (>30% reduction) and without unacceptable rise in body weight (<3kg) or BNP (<300pg/ml) will be randomly assigned to long-term treatment with atrasentan or placebo. The SONAR trial will provide a more clear answer as to whether 65

67 Prediction of the effect of atrasentan based on short term changes in multiple risk markers atrasentan confers renoprotection and whether the PRE score algorithm was actually accurate in predicting the clinical benefits of ERA. The reduction in albuminuria was an important driver of the predicted renal risk reduction. Prior trials with dual RAAS blockade showed a reduction in albuminuria and blood pressure but did not observe a renal risk reduction. However, in these trials off-target effects such as hyperkalemia, hypotension and a decrease in hemoglobin were observed which may offset the beneficial effect of albuminuria reduction. The PRE score integrates these multiple effects and translates them into long-term risk change. In RADAR/JAPAN, we noted that body weight and hemoglobin are the main contributors to increased renal and cardiovascular risk, whereas other risk markers had negligible effects. We previously applied the PRE score to a trial in which patients with type 2 diabetes and nephropathy were treated with aliskiren, and predicted that aliskiren treatment in the ALTITUDE trial would not result in the expected cardio-renal risk reduction calculated based on albuminuria reduction alone, due to off-target effects (hyperkalemia, hypotension) that negatively influenced the ultimate cardio-renal outcome (13). This was later confirmed by the early termination of the ALTITUDE trial. We note that based on this study no inferences can be made as to whether albuminuria is a valid target for renoprotective therapies. The SONAR trial with atrasentan, in which patients are randomized based on their albuminuria response, will provide more insight into this question. The PRE score may not only have implications for drug development or drug regulation but also for predicting the ultimate treatment effect on clinical outcomes in individual patient care. In current practice drug efficacy is monitored based on single risk markers, such as blood pressure for an antihypertensive drug. However, the PRE score may offer the physician and patient a better tool to estimate the overall predicted drug effect (15). In a recent study we showed that, on an individual level, the PRE score indeed provides a better prediction of who will benefit from RAAS treatment compared to using single markers alone (20). We could not predict the calculated PRE scores based on baseline values of risk markers. Similarly, none of the risk markers at baseline was able to predict the albuminuria-lowering or weight-increasing effect of atrasentan. Therefore, it is not possible to predict before exposure who will benefit from treatment using traditional recorded physical and clinical chemistry parameters. The development and implementation of novel tools, such as genomics, proteomics, or metabolomics, may 66

68 Chapter 4 Figure 4. Simulated UACR and weight changes and the effect on renal and heart failure outcomes, respectively. The shaded area reflects the 95% CI for the simulated UACR/weight responses. The black dot indicates changes observed in the total RADAR/JAPAN population + 95% CI. The dark and light grey dots represent changes observed in the albuminuria responder and non-responder population respectively. Predicted risk was calculated with risk marker changes from the total population + simulated values for either albuminuria or body weight. lead in the near future to more detailed phenotyping and may provide new insights and knowledge about individual determinants of treatment response. Some aspects of our model should be considered. First, we included all measured cardiorenal risk markers in the model to capture all potential measured effects of atrasentan. We note that some of the included risk markers may not be causally related to renal or heart failure outcomes, despite their association with outcome. However, even in the case that some of the included risk markers are not causally related to outcome, they may still be representative of the underlying 67

69 Prediction of the effect of atrasentan based on short term changes in multiple risk markers disease process and, as such, may accurately predict outcomes (21). We acknowledge that our model cannot make inferences as to whether the included risk markers are valid targets for therapy. A prospective trial targeting the multiple risk markers included in the PRE score is required to demonstrate this. Second, the model assumes that the predictive ability of short-term risk marker changes are independent of the drug or intervention. As such the model is potentially applicable to different interventions as long as all relevant risk marker changes are measured and included in the model, and risk marker-outcome relationships are not modified by treatment. This study has limitations. Firstly, the follow-up duration in the RADAR/JAPAN trials was 12 weeks. The PRE score analysis assumes that initial week-12 changes in risk markers are sustained during long-term follow-up. However, long-term stability of risk marker changes depends on various factors including treatment compliance, use of co-medication, and progression of disease. Secondly, we recognize that the sample size of the RADAR/JAPAN trials was relatively small which is reflected by the large confidence intervals in predicted treatment effects. Thirdly, we used the changes in body weight as a proxy for fluid retention. However, it is well known that changes in body weight are variable and we did not use standardized techniques (e.g. same procedures and weighing scale in all patients) to measure body weight. This has introduced random variability and limited our ability to precisely predict the long-term effect on heart failure. Lastly, we cannot exclude that there were other relevant risk markers not measured in RADAR/JAPAN and therefore not accounted for in the PRE score. In conclusion, based on short-term changes in risk markers, both atrasentan 0.75 mg/d and 1.25 mg/d are expected to decrease renal risk and slightly increase heart failure risk, the latter to a lesser extent with the low dose. Albuminuria responders to atrasentan (>30% reduction) are the major contributors to the predicted renal risk reductions. The predicted ratio of the renal risk reduction versus heart failure risk increase favors the atrasentan 0.75 mg/d dose. The ongoing hard outcome trial SONAR selects only patients with type 2 diabetes and nephropathy that respond (>30% albuminuria reduction) to atrasentan 0.75 mg/d, and will provide a more clear answer as to whether the PRE score predictions are accurate. 68

70 Chapter 4 Acknowledgements This work was performed as part of the Escher project (project nr. T6-503) within the framework of the Dutch Top Institute Pharma. Supplement Supplemental Figure S1. Overview of the risk marker changes for the responder and non-responder subset. For comparison, changes in risk markers in the overall placebo population are shown as well. Abbreviations: Plc, Placebo; Resp, Responder; Non-resp, Non-responder, SBP, systolic blood pressure, Hb, hemoglobin. 69

71 Prediction of the effect of atrasentan based on short term changes in multiple risk markers Supplementary Table S1. Renal endpoint Atrasentan 0.75mg/d Atrasentan 1.25mg/d Total population -26% (-48 to -4) -28% (-51 to -5) UACR Responders -51% (-73 to -28) -45% (-68 to -22) UACR non-responders +1% (-22 to +23) +1% (-33 to +36) Heart failure endpoint Atrasentan 0.75mg/d Atrasentan 1.25mg/d Total population 0% (-16 to +17) +8% (-12 to +28) UACR Responders -8% (-26 to +10) +6% (-16 to 28) UACR non-responders +8% (-11 to +27) +12% (-16 to +39) Predictions for the RADAR/JAPAN population (N=211) with imputed risk marker values in case of missing data. PRE scores are calculated for the total population, the responder subset and nonresponder subset. Results are shown in mean (95% CI). 70

72 Chapter 4 References 1. Gregg EW, Li Y, Wang J, et al. Changes in diabetes-related complications in the united states, N Engl J Med 2014; 370(16): Heerspink HJ, de Zeeuw D. The kidney in type II diabetes therapy. Rev Diabet Stud 2011; 8(3): Kohan DE, Pollock DM. Endothelin antagonists for diabetic and non-diabetic chronic kidney disease. Br J Clin Pharmacol 2013; 76(4): Kohan DE, Pritchett Y, Molitch M, et al. Addition of atrasentan to renin-angiotensin system blockade reduces albuminuria in diabetic nephropathy. J Am Soc Nephrol 2011; 22(4): Kohan DE, Barton M. Endothelin and endothelin antagonists in chronic kidney disease. Kidney Int 2014; 86(5): Chandrashekar K, Juncos LA. Endothelin antagonists in diabetic nephropathy: Back to basics. J Am Soc Nephrol 2014; 25(5): Wenzel RR, Littke T, Kuranoff S, et al, for the SPP301 (Avosentan) Endothelin Antagonist Evaluation in Diabetic Nephropathy Study Investigators. Avosentan reduces albumin excretion in diabetics with macroalbuminuria. J Am Soc Nephrol 2009; 20(3): Weber MA, Black H, Bakris G, et al. A selective endothelin-receptor antagonist to reduce blood pressure in patients with treatment-resistant hypertension: A randomised, double-blind, placebocontrolled trial. The Lancet 2009; 374(9699): Dhaun N, Melville V, Blackwell S, et al. Endothelin-A receptor antagonism modifies cardiovascular risk factors in CKD. J Am Soc Nephrol 2013; 24(1): de Zeeuw D, Coll B, Andress D, et al. The endothelin antagonist atrasentan lowers residual albuminuria in patients with type 2 diabetic nephropathy. J Am Soc Nephrol 2014; 25(5): Mann JFE, Green D, Jamerson K, et al, for the ASCEND Study Group. Avosentan for overt diabetic nephropathy. J Am Soc Nephrol 2010; 21(3): Kalra PR, Moon JCC, Coats AJS. Do results of the ENABLE (endothelin antagonist bosentan for lowering cardiac events in heart failure) study spell the end for non-selective endothelin antagonism in heart failure? Int J Cardiol 2002; 85(2-3):

73 Prediction of the effect of atrasentan based on short term changes in multiple risk markers 13. Smink P, Hoekman J, Grobbee D, et al. A prediction of the renal and cardiovascular efficacy of aliskiren in ALTITUDE using short-term changes in multiple risk markers. Eur J Prev Cardiol 2013; 21(4): Smink PA, Miao Y, Eijkemans MJC, et al. The importance of short-term off-target effects in estimating the long-term renal and cardiovascular protection of angiotensin receptor blockers. Clin Pharmacol Ther 2014; 95(2): Heerspink HJL, Grobbee DE, de Zeeuw D. A novel approach for establishing cardiovascular drug efficacy. Nat Rev Drug Discov 2014; 13(12): Lambers Heerspink HJ, Gansevoort RT, Brenner BM, et al. Comparison of different measures of urinary protein excretion for prediction of renal events. J Am Soc Nephrol 2010; 21(8): Hoekman J, Lambers Heerspink HJ, Viberti G, Green D, Mann JFE, de Zeeuw D. Predictors of congestive heart failure after treatment with an endothelin receptor antagonist. Clin J Am Soc Nephrol 2014; 9(3): Heerspink HL, de Zeeuw D. Pharmacology: Defining the optimal dose of a new drug: A crucial decision. Nat Rev Nephrol 2009; 5(9): Temple R. Enrichment of clinical study populations. Clin Pharmacol Ther 2010; 88(6): Schievink B, de Zeeuw D, Parving H-H, et al. The renal protective effect of angiotensin receptor blockers depends on intra-individual response variation in multiple risk markers. Br J Clin Pharmacol doi: /bcp Moons K, Kengne A, Grobbee D, et al. Risk prediction models: II. External validation, model updating, and impact assessment. Heart. 2012; 98:

74 Chapter 5 The renal protective effect of Angiotensin Receptor Blockers depends on intra-individual response variation in multiple risk markers Bauke Schievink Dick de Zeeuw Hans-Henrik Parving Peter Rossing Hiddo Lambers Heerspink British Journal of Clinical Pharmacology, 2015.

75 Intra individual response variation determines renal protective effect of ARBs Abstract Aims: Angiotensin Receptor Blockers (ARBs) are renoprotective and targeted to blood pressure. However, ARBs have multiple other (off-target) effects which may affect renal outcome. It is unknown whether on-target and off-target effects are congruent within individuals. If not, this variation in short-term effects may have important implications for the prediction of individual long-term renal outcomes. Our aim was to assess intra-individual variability in multiple parameters in response to ARBs in type 2 diabetes. Methods: Change in systolic blood pressure (SBP), albuminuria, potassium, hemoglobin, cholesterol and uric acid after 6 months of losartan treatment were assessed in the RENAAL database. Improvement in predictive performance of renal outcomes (ESRD or doubling serum creatinine) for each individual using ARBinduced changes in all risk markers was assessed by relative integrative discrimination index (RIDI). Results: SBP response showed high variability (mean -5.7 mmhg; 5 th to 95 th percentile to mmhg) between individuals. Changes in off-target parameters also showed high variability between individuals. No congruency was observed between responses to losartan in multiple parameters within individuals. Using individual responses in all risk markers significantly improved renal risk prediction (RIDI 30.4%; p<0.01) compared to using only SBP changes. Results were successfully replicated in two independent trials with irbesartan; IDNT and IRMA-2. Conclusions: In this post-hoc analysis we showed that ARBs have multiple offtarget effects which vary between and within individuals. Combining all ARB-induced responses beyond SBP provides a more accurate prediction of who will benefit from ARB therapy. Prospective trials are required to validate these findings. 74

76 Chapter 5 Introduction Intervention in the renin-angiotensin-aldosterone system (RAASi) is a mainstay of renoprotective therapy in diabetic (and non-diabetic) renal disease [1-3]. Although RAASi affords renoprotection on a group level, large individual variability in renoprotective effect exists which is mirrored by large individual variability in response in blood pressure, the primary target of Angiotensin Receptor Blockers (ARBs). The response in blood pressure serves as a proxy or surrogate for the longterm effect on cardiovascular and renal outcomes. However, blood pressure is not the only risk marker that is influenced by RAASi. RAASi has a broad spectrum of effects on other renal risk markers including decreasing albuminuria, hemoglobin, uric acid, cholesterol, and increasing potassium [4-8]. Changes in these renal risk markers have implications for patients clinical prognosis. Accordingly, RAASi induced changes in all these risk markers could influence the ultimate effect of RAAS blockade on renal outcomes, either positively or negatively. It has been assumed that the between-patient variability in blood pressure in response to RAAS blockade is paralleled by a similar between-patient variability in other risk markers. If true, measuring multiple risk markers within an individual would not be meaningful as the blood pressure response represents the variability in response to other markers as well. However, several studies show that in a proportion of patients a reduction in blood pressure is not accompanied by a reduction in albuminuria or vice versa [9-11]. It is unknown, however, whether the response to ARBs in multiple risk markers are paralleled or dissociated from the antihypertensive response within a patient, and whether this explains the inaccuracy of predicting long term renal protection by just looking at the target parameter blood pressure. We questioned whether a composite of multiple short-term risk marker changes, including on-target and off-target changes, would lead to a more accurate prediction of long-term renal protection. We thus investigated the variability in response in multiple risk markers within an individual and assessed the congruency in response in multiple risk markers within an individual. Secondly, we examined whether inclusion of changes in multiple risk markers of each individual improves prediction of renal outcome. Data of already finished large ARB intervention trials in 75

77 Intra individual response variation determines renal protective effect of ARBs type 2 diabetic patients with elevated albuminuria (RENAAL, IDNT and IRMA-2) were used for this analysis. Methods Data sources and patient population We used the individual patient data from the RENAAL, IDNT and IRMA-2 trials. The detailed design, rationale, and study outcome for these trials have been previously published [12-14]. All trials investigated the efficacy of an ARB (losartan in RENAAL and irbesartan in IDNT and IRMA-2) on renal outcomes in subjects with type 2 diabetes and nephropathy. Inclusion criteria for the RENAAL and IDNT trials were similar aside from a few minor differences. Patients with type 2 diabetes, hypertension and age between years were eligible for these trials. In RENAAL and IDNT, serum creatinine levels ranged between 1.0 mg/dl and 3.0 mg/dl and all subjects had proteinuria, defined as 24-hour urinary protein excretion of >900 mg in the IDNT trial whereas for RENAAL a urinary albumin to creatinine ratio (UACR) of >300 mg/g or a 24 hour urinary protein excretion >500 mg/day was required. In the IRMA-2 trial, eligible patients had type 2 diabetes and microalbuminuria, defined as urinary albumin excretion between 20 and 200 µg/min, and serum creatinine no more than 1.5 mg/dl in males or 1.1 mg/dl in females. Exclusion criteria for all three trials were type 1 diabetes or non-diabetic renal disease. Patients randomized to active study treatment received losartan 100 mg/day in RENAAL, irbesartan 300 mg/day in IDNT and irbesartan 150 mg/day or 300 mg/day in IRMA-2 to achieve a blood pressure target of at least 135/85 mmhg. If the blood pressure target was not achieved the dosage of other antihypertensive drugs were increased or additional antihypertensive agents (but not RAASi) were added to achieve the target blood pressure. The primary endpoint in RENAAL and IDNT used for this analysis was the time to a sustained doubling of baseline serum creatinine or end stage renal disease. In IDNT, a sustained serum creatinine 6.0 mg/dl was used as an additional component in the primary endpoint. All outcomes were adjudicated by an independent blinded endpoint committee using rigorous outcome definitions. Clinical renal endpoints were not recorded in the IRMA-2 trial. All patients signed informed 76

78 Chapter 5 consent before enrollment, and the local Institutional Review Board of each participating center approved the RENAAL, IDNT, and IRMA-2 trial. Responses in risk markers This post-hoc analysis focuses on the response in multiple markers including systolic blood pressure (SBP), albuminuria, serum potassium, hemoglobin, total cholesterol, and uric acid. These markers were selected since prior studies have shown that RAASi can affect these risk marker levels when compared to placebo [4-8]. Uric acid was not included in the analysis for the IDNT and IRMA-2 trials, as it was previously shown that irbesartan does not affect this risk marker [15]. All blood pressure measurements in all trials were taken after a period of at least 5 minutes in sitting position. Three consecutive blood pressure measurements were recorded in the same arm. The mean value of the three systolic and diastolic blood pressure readings was calculated for each study visit. In a subset of patients in the IRMA-2 trial, 24-hour ambulatory blood pressure monitoring was performed as well. Other risk markers (albuminuria, potassium, hemoglobin, cholesterol, uric acid) were measured in a central laboratory in each trial. Albuminuria was measured in first morning void urine collections in RENAAL and IRMA-2 for measurement of the albumin to creatinine ratio. In IDNT, 24-hour urine collections were performed for measurement of the albumin to creatinine ratio. Throughout this chapter albumin to creatinine ratio is designated as albuminuria. Response in each parameter was defined as the change between the month- 6 and baseline value. Responders were defined as patients with a risk marker change in the hypothesized direction. Hence, a 6-month decrease was used to define responders for SBP, albuminuria, hemoglobin, total cholesterol and uric acid, and a 6-month increase was used to define serum potassium responders. Albuminuria response at month 6 for each patient was calculated as (1 log ratio of month 6 to baseline albuminuria) multiplied by 100%. On the basis of previous analyses, the month 6 value was chosen because most parameters were measured at month 6, the treatment effects were considered fully present, and few events occurred during the first 6 months [12,13]. 77

79 Intra individual response variation determines renal protective effect of ARBs Integration of responses in multiple risk markers The effect on all risk markers was combined to calculate an integrated risk marker effect of ARB treatment. To this end, we used a previously described and validated algorithm referred to as the multiple Parameter Risk Efficacy (PRE) score [15,16]. In short, a multivariable Cox proportional hazards model was used to estimate the coefficients and hazard ratios associated with each risk marker for the first recorded renal event. The regression coefficients for each risk marker were then taken and used as weights for the risk algorithm. The risk algorithm was applied to the risk markers observed in the RENAAL and IDNT trial at baseline and month 6 in order to calculate 3-year renal risk at both time points. The percentage difference in risk between the two time points represents the individual PRE score [15,17]. Statistical analysis Changes in risk marker levels between baseline and month 6 were reported as mean with 5 th to 95 th percentile. Variables that were non-normally distributed were logtransformed and reported as geometric mean change. Treatment response was calculated by subtracting month-6 measurements from baseline values. Patients with a reduction at month 6 from baseline in either SBP, albuminuria, hemoglobin, cholesterol, uric acid, or increase in potassium were classified as responders for the respective risk marker. For each risk marker we created a responder population. This resulted in six responder populations. Radar plots were subsequently constructed to determine whether responses in one of the six markers were congruent with responses in other markers. The mean response in all risk markers in the overall population was plotted in each radar plot together with the responses in the responder population. We created radar plots for each of the six responder populations. Number of responders in each risk marker were counted and compared by Chi square test with a post-hoc Bonferroni correction for multiple testing. Correlations between risk markers in individual patients were calculated with Pearson correlations. Mean response for risk marker changes in each responder group were compared with the mean response in the total population by t test or Wilcoxon signed rank test with a post-hoc Bonferroni correction for multiple testing, where appropriate. We subsequently assessed whether changes in single risk markers or multiple risk markers improved renal risk prediction. These analyses were conducted in RENAAL 78

80 Chapter 5 and IDNT trials since in the IRMA-2 trial no clinical endpoints were recorded. We used responses in SBP, albuminuria, and individual PRE scores, that represents the integration of 6-month responses in multiple risk markers, in Cox regression analysis. Cox models were adjusted for baseline values of age, gender, egfr, albuminuria and hemoglobin to take into account differences between patients in renal risk at baseline. We adjusted for these risk markers as they were previously shown to independently predict renal risk [18]. The improvement in predictive performance was assessed by C statistic and relative integrated discrimination improvement (RIDI). The RIDI measures the percentage of increased discrimination when comparing prediction models [19]. A p value of <0.05 was selected as statistically significant. Statistical analysis was conducted with R version (R Foundation for Statistical Computing, Vienna, Austria). Results Baseline characteristics A total of 531 (71%) of patients assigned to ARB treatment had complete risk marker measurements at baseline and month 6. Baseline characteristics of these patients are presented in Table 1. The included population did not differ from the overall losartan assigned population. (Supplement Table S1). Between-patient variability in response to losartan A large variability in responses between individuals in systolic blood pressure (mean [5 th to 95 th percentile]: -5.7mmHg [-36.5 to +24.0]), albuminuria (-31% [-78 to +121]), serum potassium (0.22mEq/L [-0.55 to +1.00]), hemoglobin, (-0.6g/dL [-2.5 to +1.35]), total cholesterol (-10.1,mg/dL [-89.5 to +59.0]), and uric acid (0.02mg/dL [ to +2.10]) was observed (Figure 1). Within-patient variability in multiple markers in response to losartan To determine whether responses in systolic blood pressure were paralleled by responses in other risk markers within an individual we assessed responses in all risk markers in the overall population and subsequently in subjects with a reduction in systolic blood pressure. 79

81 Intra individual response variation determines renal protective effect of ARBs Figure 1. Discordance between ARB-induced responses in multiple parameters within individual patients. A: Overview of the variation in risk marker response in the total population (inter-individual variability). The thick short line in the boxplots indicates the median and the dot the mean change. Box and whiskers represent interquartile range and 5th to 95th percentile, respectively. B: Radar plot showing the overall response (bold black outer line with numbers indicating mean values) for each risk marker in RENAAL. Dashed grey lines indicate 95% confidence interval. C: Radar plots showing discordance of the different responses within individuals. For example, the red lines and numbers indicate the risk marker responses in blood pressure responders. The black outer line represents the response in the overall population as explained in B. The overlap between each of the responder populations (colored line) versus the overall population (black line) indicates that responses within an individual are discordant. N underneath radar plots indicate number of responders in each figure. Abbreviations: SBP, systolic blood pressure; ACR, albuminuria; K, potassium; Hb, hemoglobin; chol, cholesterol; UA, uric acid. 80

82 Chapter 5 Table 1. Baseline characteristics of the included patients. RENAAL (N=531) IDNT (N=376) IRMA-2 (N=255) Age (years) 60.0 (7.1) 59.1 (7.1) 57.9 (7.9) Males, N (%) 328 (62) 250 (66) 171 (67) Weight, kg 82.4 (21.2) 88.8 (18.1) 85.2 (14.5) Race, N (%) White 241 (45) 284 (76) 253 (99) Black 87 (16) 45 (12) 0 (0) Hispanic 102 (19) 17 (4) 0 (0) Asian 97 (18) 12 (3) 0 (0) Other 4 (1) 18 (5) 2 (1) SBP, mmhg (18.7) (19.0) (14.0) ACR, mg/g 1155 [ ] 1420 [ ] 55 [34-96] K, meq/l 4.59 (0.48) 4.65 (0.53) 4.75 (0.52) Hb, g/dl 12.6 (1.8) 13.0 (2.0) 14.4 (1.2) Cholesterol, mg/dl (54.4) (51.4) (50.0) Uric acid, mg/dl 6.73 (1.76) 6.78 (1.81) NA HbA1c, % 8.44 (1.56) 7.97 (1.67) 7.18 (1.73) egfr, ml/min/1.73m (11.8) 47.4 (16.7) 72.3 (14.3) In mean (SD) unless otherwise indicated. ACR is calculated as median + interquartile range. Baseline measurements for uric acid were not available in the IRMA-2 trial. The radar plot in Figure 1 shows that the magnitude of responses in all risk markers were similar in the overall population and in the subgroup of patients with a reduction in systolic blood pressure, suggesting that responses are discordant. When responder populations were defined by responses in off-target parameters, the magnitude of responses in the remaining parameters were similar to the overall population (Figure 1), except for the albuminuria response in uric acid responders. The number and proportion of patients with a response to losartan in each risk marker are shown in Table 2. In the overall population a response in SBP was observed in 61% of subjects, albuminuria in 72%, potassium in 66%, hemoglobin in 72%, cholesterol in 61%, and uric acid in 47%. These percentages were not statistically different in sub-group populations defined by a response in SBP or other off-target biomarkers (Table 2). The correlation between responses in individual risk markers within an individual is shown in Table 3. There was no correlation between responses in different parameters within an individual, with the highest correlation observed between hemoglobin and cholesterol (r=0.30). 81

83 Intra individual response variation determines renal protective effect of ARBs Table 2. Proportion of patients with response in overall population and various responder populations. Response in Total population (N=531) SBP responders (N=323) ACR responders (N=382) K responders (N=349) Hb responders (N=383) Cholesterol responders (N=325) Uric Acid Responders (N=251) SBP 323 (61) 323 (100) 256 (67) 208 (60) 237 (61.9) 202 (62) 134 (53) ACR 382 (72) 256 (79) 382 (100) 262 (75) 280 (73.1) 241 (74) 157 (63) K 349 (66) 208 (64) 262 (69) 349 (100) 255 (67) 217 (67) 159 (63) Hb 383 (72) 237 (73) 280 (73) 255 (73) 383 (100) 259 (80) 176 (70) Cholesterol 325 (61) 202 (63) 241 (63) 217 (62) 259 (68) 325 (100) 150 (60) Uric acid 251 (47) 134 (42) 157 (41) 159 (46) 176 (46) 150 (46) 251 (100) Number of patients (percentage) with a response in each risk marker. Results are displayed for the total population, and for subsets of responders per risk marker. Chi square tests with a post-hoc Bonferroni correction revealed no statistically significant differences. Prediction of renal outcome During a median follow-up of 2.6 years, 151 (28.4%) patients treated with losartan reached a composite event of doubling of serum creatinine or ESRD. Changes in multiple biomarkers for each individual were integrated and represented by the PRE score. PRE scores of individual patients were associated with renal outcome independent of baseline renal risk markers (HR 3.18 (95% confidence interval 2.32 to 4.37; p<0.01) per unit increment in PRE score. Relative to using changes in SBP to monitor the efficacy of losartan, using the PRE score significantly improved renal risk prediction (RIDI 30.4%; p<0.01; Table 4). The C statistic of the PRE score for the renal outcome was 0.840, significantly higher (p<0.01) than using changes in SBP alone (C statistic: 0.796). Validation In order to validate the results, we assessed the variation in response in multiple risk markers to ARBs in two other datasets from completed clinical trials. First, in the IDNT trial, in which patients were treated with the ARB irbesartan, we observed similar response patterns when compared to RENAAL (Supplemental Figure S1, Table S2 and Table S3). As in RENAAL, a response in systolic blood pressure with irbesartan was not associated with other risk markers within individual patients. Additionally, we observed no correlations between responses within patients in different parameters suggesting that responses in multiple parameters within a patient are discordant. Using the PRE score improved renal risk prediction by 30.5% (p=0.02 compared to using only SBP, Table 4), with a C statistic of Secondly, 82

84 Chapter 5 we validated our results in the IRMA-2 trial in which patients with microalbuminuria were treated with irbesartan (Supplemental Figure S2, Table S4 and Table S5). Results were again similar as in RENAAL with a lack of congruency between changes in multiple risk markers within individuals. In a subset of patients in the IRMA-2 trial systolic blood pressure was measured by 24-hour ambulatory blood pressure monitoring. Again, we observed no correlations in response, similar to the overall population (Supplemental Figure S3, Table S6). Table 3. Pearson correlation coefficients between responses in individual risk markers in individual patients. SBP ACR K Hb Cholesterol Uric acid SBP ACR K Hb Cholesterol Uric Acid Discussion This study shows that the individual variability in renoprotection of RAASi is not only determined by response variation of the target parameter blood pressure, but also by the variation of multiple off-target effects. Since variations in on-target and off-target effects were discordant within individual patients, a composite response score that takes changes in all risk markers into account is needed to optimize the predictive power of treatment-induced short-term risk marker changes for long-term renal outcomes. These results were replicated in two independent datasets. The variation in response to RAASi has been an area of research interest for several decades [20,21]. A large variability in the renal response to RAAS intervention has been observed ranging from no effect to a complete arrest of renal function decline [22,23]. A meta-analysis concluded from trial level data that the degree of blood pressure lowering is associated with the degree of renal protection [24]. It is therefore logical that antihypertensive drugs are titrated to target blood pressure, expecting that this is an accurate indicator of kidney protection. However, ARBs have been shown to affect other renal risk markers than blood pressure, and 83

85 Intra individual response variation determines renal protective effect of ARBs changes in these risk markers also affect the ultimate renal outcome. In addition to the variable effects of ARBs on multiple risk markers between patients, we also showed that the response in multiple risk markers within individual patients is highly variable. Earlier studies with RAASi already revealed a discordance between blood pressure and albuminuria response to ARBs within individual patients, with both risk markers being independently associated with renal outcome [9-11]. In line, previous studies also showed that the dose-response for blood pressure is different from the dose-response for albuminuria, confirming that responses to ARBs in these parameters are discordant [25-27]. Our study builds upon these previous studies and shows for the first time that blood pressure response is also not congruent with responses in other risk markers. Importantly, none of the other (off-target) risk markers revealed a correlation in response with other risk markers. Therefore, monitoring blood pressure in case of ARB therapy is not sufficient to predict the ultimate renal outcome. Table 4. RIDI and C statistic RIDI C statistic RENAAL SBP change ref (ref) albuminuria change 19.2% (p=0.03) (p=0.01) PRE score 30.4% (p<0.01) (p<0.01) IDNT SBP change ref (ref) albuminuria change 23.3% (p=0.04) (p=0.19) PRE score 30.5% (p=0.02) (p=0.10) Predictive performance of using change in albuminuria or PRE score in a risk prediction model (corrected for baseline age, gender, egfr, albuminuria and hemoglobin), compared to using SBP. Results are shown for RENAAL and IDNT since clinical renal outcomes were not collected in IRMA-2. What are potential implications for clinical practice? Blood pressure control is beyond doubt critical to achieve long-term renoprotection. Clinical practice guidelines therefore recommend to measure blood pressure regularly after start of antihypertensive medication to monitor the effectiveness of the instituted therapy. However, only measuring blood pressure does not capture the potential response in other markers. Because changes in other risk markers are also associated with longterm renal outcome, either contributing or offsetting the effect predicted by blood 84

86 Chapter 5 pressure response alone, our results suggest that it is necessary to monitor the effect on all known risk markers and integrate these effects to be able to accurately predict the ultimate treatment effect of ARBs in individual patients. This could also imply that further dose increase needs to be explored even if the SBP for an individual patient is controlled: the maximal effect on SBP does not necessarily equal the maximal renoprotective effect. We developed an individual risk algorithm that incorporates drug responses in multiple parameters. In previous studies we showed that this algorithm can be used to predict the effect of ARBs on a population level. In this study we showed that it also accurately predicts the ARB response on an individual level. Our results imply that long-term renoprotection is still possible in the absence of a blood pressure response as long as the composite of the response in other risk markers is favorable for renoprotection. Vice versa, long-term renal damage can occur even if blood pressure is decreased, but changes in other risk markers sum up to a degree of damage that exceeds the renoprotection induced by blood pressure reduction. The underlying biological mechanisms that govern variation in response within individuals are not yet elucidated but several possible explanations exist. Firstly, there is the possibility of differences in systemic versus renal tissue-specific RAAS activity. In this respect it has been shown that the blood pressure response depends to a large extent on extra-renal RAAS inhibition whereas it may be possible that the response in other risk markers, such as albuminuria, hemoglobin, potassium, depends on intra-renal RAAS inhibition [28]. Secondly, the susceptibility of an individual in terms of sodium/potassium balance, albuminuria, or blood pressure changes following changes in RAAS activity may be different as a result of differences in genetic make-up, dietary consumption, or their combination [29,30]. For example, genetic differences in CYP2C9, the enzyme metabolizing the pro-drug losartan to its active metabolite, may result in variation in drug exposure between individuals [31]. Thirdly, as patients with diabetes and nephropathy use multiple drugs, and concomitant drug use was present in the analyzed clinical trials, drugdrug interaction leading to different responses cannot be excluded. Patients may also respond differently to drugs due to differences in comorbidities, such as renal artery stenosis. Finally, it is possible that the lack of correlation in responses is due to measurement error. However, all risk markers were measured in central laboratories using strict guidelines and criteria and blood pressure was measured 85

87 Intra individual response variation determines renal protective effect of ARBs according to standardized protocol guidelines. Additionally, the finding that even 24- hour blood pressure response did not produce correlations with responses in other markers in a subset of the IRMA-2 trial makes the possibility of misclassification less likely. We showed that adding ARB-induced changes to a renal risk model markedly improved renal risk prediction. Much research is focused on developing new biomarker-based models for predicting diabetic disease progression and renal outcomes with generally modest additional value [32-34]. In this respect it is noteworthy that prediction of renal endpoints markedly improved by considering response to ARBs in readily available clinical parameters. Thus, using the multiple parameter response efficacy (PRE) score is a pragmatic, cheap, and effective tool to improve renal risk prediction. Our study has limitations. First, the trials included in this study were designed to assess the effects of ARBs on renal disease progression and were not designed to investigate the variability in response. The results of this study are therefore only hypothesis generating. Prospective studies are required to confirm our findings and are currently ongoing (IMPROVE study; Dutch trial register NTR 4439). Although blood pressure was measured according predefined protocol guidelines, 24-hr blood pressure monitoring results were only available in a small subset of patients. In addition, changes in other risk markers were based on changes between two predetermined time points without a confirmatory measurement. We can therefore not exclude that part of the 'response' variability is due to random day-to-day variability and/or measurement variability. Second, the predictions of the PRE score are based on the assumption that the drug effect at six months persists over time. It may be possible however that during prolonged follow-up risk markers of some patients may regress to baseline values. This may have led to an underestimation of the predictive performance of the PRE score. In conclusion, our study shows that ARBs have variable effects on multiple risk markers (between-patient variability) and these effects vary within patients. An individual risk model that takes variability in treatment response to all known risk markers into account provides a more accurate prediction of who will benefit from ARB therapy than using blood pressure or any other single marker alone. This suggests that in clinical practice all relevant risk markers should be monitored and 86

88 Chapter 5 integrated to fully appreciate the ARB treatment effect on renal outcomes. Further studies are required to prospectively validate these findings. Acknowledgements This work was performed as part of the Escher project (project nr. T6-503) within the framework of the Dutch Top Institute Pharma. We would like to thank Skander Mulder from the University of Groningen, Groningen, The Netherlands for his help with creating the radar plots. 87

89 Intra individual response variation determines renal protective effect of ARBs Supplement Supplemental Table S1. Comparison of all patients randomized to losartan versus complete case subset derived from the RENAAL study. Complete cases (N=531) All patients (N=751) P value Age (years) 60.0 (7.1) 60.0 (7.4) 0.99 Males, N (%) 328 (62) 462 (62) 0.97 Weight, kg 82.4 (21.2) 82.6 (19.2) 0.85 Race, N (%) White 241 (45) 358 (48) - Black 87 (16) 125 (17) - Hispanic 102 (19) 140 (19) - Asian 97 (18) 117 (16) - Other 4 (1) 11 (1) - SBP, mmhg (18.7) (18.7) 0.75 ACR, mg/g 1155 [ ] 1173 [ ] 0.33 K, meq/l 4.6 (0.5) 4.6 (0.5) 0.96 Hb, g/dl 12.5 (1.8) 12.5 (1.8) 0.81 Cholesterol, mg/dl (54.4) (55.6) 0.41 Uric acid, mg/dl 6.7 (1.8) 6.7 (1.7) 0.72 HbA1c,% 8.4 (1.6) 8.5 (1.8) 0.31 egfr, ml/min/1.73m (11.8) 39.6 (11.9) 0.57 Supplemental Table S2. Proportion of patients with response in overall population and responder populations in IDNT. Response in Total population (N=376) SBP responders (N=307) ACR responders (N=267) K responders (N=202) Hb responders (N=242) Cholesterol responders (N=221) SBP 307 (82) 307 (100) 234 (88) 164 (81) 193 (80) 186 (84) ACR 267 (71) 234 (76) 267 (100) 143 (71) 178 (74) 162 (73) K 202 (54) 164 (53) 143 (54) 202 (100) 128 (53) 115 (52) Hb 242 (64) 193 (63) 178 (67) 128 (63) 242 (100) 161 (73) Cholesterol 221 (59) 186 (61) 162 (61) 115 (57) 161 (67) 221 (100) Number of patients (percentage) with a response in each risk marker. Results are displayed for the total population, and for subsets of responders per risk marker. Chi square tests with a post-hoc Bonferroni correction revealed no statistically significant differences. 88

90 Chapter 5 Supplemental Table S3. Pearson s correlation coefficients between responses in individual risk markers in individual patients for IDNT. SBP ACR K Hb Cholesterol SBP ACR K Hb Cholesterol Supplemental Table S4. Proportion of patients with response in overall population and responder populations in IRMA-2. Response in Total population (N=255) SBP responders (N=208) ACR responders (N=188) K responders (N=109) Hb responders (N=141) Cholesterol responders (N=146) SBP 208 (82) 208 (100) 157 (84) 87 (80) 122 (87) 121 (83) ACR 188 (74) 157 (75) 188 (100) 85 (78) 111 (79) 108 (74) K 109 (43) 87 (42) 85 (45) 109 (100) 50 (35) 60 (41) Hb 141 (55) 122 (59) 111 (59) 50 (46) 141 (100) 94 (64) Cholesterol 146 (57) 121 (58) 108 (57) 60 (55) 94 (67) 146 (100) Number of patients (percentage) with a response in each risk marker. Results are displayed for the total population, and for subsets of responders per risk marker. Chi square tests with a post-hoc Bonferroni correction revealed no statistically significant differences. Supplemental Table S5. Pearson s correlation coefficients between responses in individual risk markers in individual patients for IRMA-2. SBP ACR K Hb Cholesterol SBP ,15 ACR K ,02 1 0,17 0,07 Hb -0,02 0,17 0,17 1 0,28 Cholesterol 0,15 0,00 0,07 0,

91 Intra individual response variation determines renal protective effect of ARBs Supplemental Figure S1. Variation in risk marker response and responder populations in IDNT, similar to Figure 1. Supplemental Figure S2. Variation in risk marker response and responder populations in IRMA-2, similar to Figure 1. 90

92 Chapter 5 Supplemental Figure S3. Variation in SBP response for total and responder population in IRMA-2 subset with 24h SBP measurements. Supplemental Table S6. Total population SBP responders Pearson correlation Response in: SBP 45 (71) 45 (100) 1 ACR 47 (75) 36 (80) 0.05 K 26 (41) 19 (42) Hb 37 (59) 29 (64) Cholesterol 26 (41) 20 (44) Proportion of patients with response in overall population and responder populations, and Pearson correlation coefficients between SBP and other responses in IRMA-2 patients with 24h SBP measurements. 91

93 Intra individual response variation determines renal protective effect of ARBs References 1. Volpe M, Cosentino F, Tocci G, Palano F, Paneni F. Antihypertensive therapy in diabetes: The legacy effect and RAAS blockade. Curr Hypertens Rep 2011; 13(4): Steckelings UM, Rompe F, Kaschina E, Unger T. The evolving story of the RAAS in hypertension, diabetes and CV disease? moving from macrovascular to microvascular targets. Fundam Clin Pharmacol 2009; 23(6): Heerspink HJ, de Zeeuw D. The kidney in type II diabetes therapy. Rev Diabet Stud 2011; 8(3): Smink PA, Bakker SJL, Laverman GD, Berl T, Cooper ME, de Zeeuw D, Lambers Heerspink HJ. An initial reduction in serum uric acid during angiotensin receptor blocker treatment is associated with cardiovascular protection: A post-hoc analysis of the RENAAL and IDNT trials. J Hypertens 2012; 30(5): Miao Y, Dobre D, Lambers Heerspink HJ, Brenner BM, Cooper ME, Parving H, Shahinfar S, Grobbee D, Zeeuw D. Increased serum potassium affects renal outcomes: A post hoc analysis of the reduction of endpoints in NIDDM with the angiotensin II antagonist losartan (RENAAL) trial. Diabetologia 2011; 54(1): Mohanram A, Zhang Z, Shahinfar S, Lyle PA, Toto RD. The effect of losartan on hemoglobin concentration and renal outcome in diabetic nephropathy of type 2 diabetes. Kidney Int 2007; 73(5): de Zeeuw D, Remuzzi G, Parving H, Keane WF, Zhang Z, Shahinfar S, Snapinn S, Cooper ME, Mitch WE, Brenner BM. Proteinuria, a target for renoprotection in patients with type 2 diabetic nephropathy: Lessons from RENAAL. Kidney Int 2004; 65(6): Tershakovec AM, Keane WF, Zhang Z, Lyle PA, Appel GB, McGill JB, Parving H, Cooper ME, Shahinfar S, Brenner BM. Effect of LDL cholesterol and treatment with losartan on end-stage renal disease in the RENAAL study. Diabetes Care 2008; 31(3): Holtkamp FA, de Zeeuw D, de Graeff PA, Laverman GD, Berl T, Remuzzi G, Packham D, Lewis JB, Parving H, Lambers Heerspink HJ. Albuminuria and blood pressure, independent targets for cardioprotective therapy in patients with diabetes and nephropathy: A post hoc analysis of the combined RENAAL and IDNT trials. Eur Heart J 2011; 32(12): Eijkelkamp WBA, Zhang Z, Remuzzi G, Parving H, Cooper ME, Keane WF, Shahinfar S, Gleim GW, Weir MR, Brenner BM, de Zeeuw D. Albuminuria is a target for renoprotective therapy 92

94 Chapter 5 independent from blood pressure in patients with type 2 diabetic nephropathy: Post hoc analysis from the reduction of endpoints in NIDDM with the angiotensin II antagonist losartan (RENAAL) trial. JASN 2007; 18(5): Laverman GD, Andersen S, Rossing P, Navis G, de Zeeuw D, Parving H. Renoprotection with and without blood pressure reduction. Kidney Int 2005; 67: S54-S Brenner BM, Cooper ME, de Zeeuw D, Keane WF, Mitch WE, Parving H, Remuzzi G, Snapinn SM, Zhang Z, Shahinfar S. Effects of losartan on renal and cardiovascular outcomes in patients with type 2 diabetes and nephropathy. N Engl J Med 2001; 345(12): Lewis EJ, Hunsicker LG, Clarke WR, Berl T, Pohl MA, Lewis JB, Ritz E, Atkins RC, Rohde R, Raz I. Renoprotective effect of the angiotensin-receptor antagonist irbesartan in patients with nephropathy due to type 2 diabetes. N Engl J Med 2001; 345(12): Parving H, Lehnert H, Bröchner-Mortensen J, Gomis R, Andersen S, Arner P. The effect of irbesartan on the development of diabetic nephropathy in patients with type 2 diabetes. N Engl J Med 2001; 345(12): Smink PA, Miao Y, Eijkemans MJC, Bakker SJL, Raz I, Parving H, Hoekman J, Grobbee DE, de Zeeuw D, Lambers Heerspink,H.J. The importance of short-term off-target effects in estimating the long-term renal and cardiovascular protection of angiotensin receptor blockers. Clin Pharmacol Ther 2014; 95(2): Smink P, Hoekman J, Grobbee D, Eijkemans M, Parving HH, Persson F, Ibsen H, Lindholm L, Wachtell K, de Zeeuw D, Heerspink HL. A prediction of the renal and cardiovascular efficacy of aliskiren in ALTITUDE using short-term changes in multiple risk markers. Eur J Prev Cardiol 2014; 21(4): Heerspink HJL, Grobbee DE, de Zeeuw D. A novel approach for establishing cardiovascular drug efficacy. Nat Rev Drug Discov 2014; 13(12): Keane WF, Brenner BM, De Zeeuw D, Grunfeld J, Mcgill J, Mitch WE, Ribeiro AB, Shahinfar S, Simpson RL, Snapinn SM, Toto R. The risk of developing end-stage renal disease in patients with type 2 diabetes and nephropathy: The RENAAL study. Kidney Int 2003; 63(4): Pencina MJ, D'Agostino RB, D'Agostino RB, Vasan RS. Evaluating the added predictive ability of a new marker: From area under the ROC curve to reclassification and beyond. Stat Med 2008; 27(2): Heeg JE, de Jong P,E., van der Hem G,K., de Zeeuw D. Efficacy and variability of the antiproteinuric effect of ACE inhibition by lisinopril. Kidney Int 1989; 36(2):

95 Intra individual response variation determines renal protective effect of ARBs 21. Fornage M, Amos CI, Kardia S, Sing CF, Turner ST, Boerwinkle E. Variation in the region of the angiotensin-converting enzyme gene influences interindividual differences in blood pressure levels in young white males. Circulation 1998; 97(18): Rossing P, Hommel E, Smidt UM, Parving H-. Reduction in albuminuria predicts a beneficial effect on diminishing the progression of human diabetic nephropathy during antihypertensive treatment. Diabetologia 1994; 37(5): Bos H, Andersen S, Rossing P, de Zeeuw D, Parving H, de Jong P,E., Navis G. Role of patient factors in therapy resistance to antiproteinuric intervention in nondiabetic and diabetic nephropathy. Kidney Int 2000; 57: S32-S Bakris GL, Williams M, Dworkin L, Elliott WJ, Epstein M, Toto R, Tuttle K, Douglas J, Hsueh W, Sowers J. Preserving renal function in adults with hypertension and diabetes: A consensus approach. AJKD 2000; 36(3): Burgess E, Muirhead N, de Cotret PR, Chiu A, Pichette V, Tobe S, the SMART (Supra Maximal Atacand Renal Trial) Investigators. Supramaximal dose of candesartan in proteinuric renal disease. JASN 2009; 20(4): Rossing K, Schjoedt KJ, Jensen BR, Boomsma F, Parving H. Enhanced renoprotective effects of ultrahigh doses of irbesartan in patients with type 2 diabetes and microalbuminuria. Kidney Int 2005; 68(3): Hollenberg NK, Parving H, Viberti G, Remuzzi G, Ritter S, Zelenkofske S, Kandra A, Daley WL, Rocha R. Albuminuria response to very high-dose valsartan in type 2 diabetes mellitus. J Hypertens. 2007; 25(9): Crowley SD, Gurley SB, Oliverio MI, Pazmino AK, Griffiths R, Flannery PJ, Spurney RF, Kim H, Smithies O, Le TH, Coffman TM. Distinct roles for the kidney and systemic tissues in blood pressure regulation by the renin-angiotensin system. J Clin Invest 2005; 115(4): Vogt L, Waanders F, Boomsma F, de Zeeuw D, Navis G. Effects of dietary sodium and hydrochlorothiazide on the antiproteinuric efficacy of losartan. Journal of the American Society of Nephrology 2008; 19(5): Lely AT, Heerspink HJL, Zuurman M, Visser FW, Kocks MJA, Boomsma F, Navis G. Response to angiotensin-converting enzyme inhibition is selectively blunted by high sodium in angiotensinconverting enzyme DD genotype: Evidence for gene-environment interaction in healthy volunteers. J Hypertens. 2010; 28(12):

96 Chapter Yasar Ü, Forslund-Bergengren C, Tybring G, Dorado P, LLerena A, Sjöqvist F, Eliasson E, Dahl M. Pharmacokinetics of losartan and its metabolite E-3174 in relation to the CYP2C9 genotype. Clin Pharmacol Ther 2002; 71(1): Schena FP. Biomarkers and personalized therapy in chronic kidney diseases. Expert Opin Investig Drugs : 2014; 23(8): Kolberg JA, Jørgensen T, Gerwien RW, Hamren S, McKenna MP, Moler E, Rowe MW, Urdea MS, Xu XM, Hansen T, Pedersen O, Borch-Johnsen K. Development of a type 2 diabetes risk model from a panel of serum biomarkers from the Inter99 cohort. Diabetes Care 2009; 32(7): Wu H, Yu Z, Qi Q, Li H, Sun Q, Lin X. Joint analysis of multiple biomarkers for identifying type 2 diabetes in middle-aged and older chinese: A cross-sectional study. BMJ Open 2011; 1(1): e

97

98 Chapter 6 The use of surrogate endpoints in regulating medicines for cardio-renal disease: opinions of stakeholders Bauke Schievink Hiddo Lambers Heerspink Hubert Leufkens Dick de Zeeuw Jarno Hoekman PLoS One, 2014.