BMJ Open. Cardiovascular medicine. General diabetes < DIABETES & ENDOCRINOLOGY, EPIDEMIOLOGY, PREVENTIVE MEDICINE, PRIMARY CARE

Transcription

1 Development and validation of risk prediction equations to estimate future risk of heart failure in patients with diabetes: prospective cohort study. Journal: Manuscript ID: bmjopen Article Type: Research Date Submitted by the Author: -Apr-0 Complete List of Authors: Hippisley-Cox, Julia; University of Nottingham, ; ClinRisk Ltd, Coupland, Carol; University of Nottingham, Division of Primary Care <b>primary Subject Heading</b>: Secondary Subject Heading: Keywords: Cardiovascular medicine Cardiovascular medicine, Diabetes and endocrinology, Epidemiology, Evidence based practice, General practice / Family practice General diabetes < DIABETES & ENDOCRINOLOGY, EPIDEMIOLOGY, PREVENTIVE MEDICINE, PRIMARY CARE : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

2 Page of Development and validation of risk prediction equations to estimate future risk of heart failure in patients with diabetes: prospective cohort study. Authors Paper Presenting Original Research Julia Hippisley-Cox Professor of Clinical Epidemiology & General Practice. MD, FRCP, FRCGP, MBChB Carol Coupland Institutions Associate Professor in Medical Statistics PhD, BSc Division of Primary Care, th floor, Tower Building, University Park, Nottingham, NG RD, UK Author for correspondence Julia Hippisley-Cox Julia.hippisley-cox@nottingham.ac.uk Carol.coupland@nottingham.ac.uk Telephone: + (0) Fax: + (0) 0 Word count: words - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

3 Page of ABSTRACT Objective To develop and externally validate risk prediction equations to estimate the 0 year risk of heart failure in patients with diabetes aged - years. Design Cohort study using routinely collected data from general practices in England between and 0 contributing to the QResearch and CPRD databases. Setting We used QResearch practices to develop the equations. We validated it in different QResearch practices and CPRD practices. Participants,0 patients in the derivation cohort;,0 in the QResearch validation cohort and,0 in the CPRD validation cohort. Measurement Incident diagnosis of heart failure recorded on the patients linked electronic GP, mortality or hospital record. Risk factors included age, BMI, systolic blood pressure, cholesterol/hdl ratio, HBAC, material deprivation, ethnicity, smoking, diabetes duration, type of diabetes, atrial fibrillation, cardiovascular disease chronic renal disease and family history of premature coronary heart disease. Methods We used Cox proportional hazards models to derive separate risk equations in men and women for evaluation at 0 years. Measures of calibration, discrimination and sensitivity were determined in two external validation cohorts. Results We identified,0 cases of heart failure in the derivation cohort;, in the QResearch validation cohort and, in the CPRD cohort. The equations included: age, BMI, systolic blood pressure, cholesterol/hdl ratio, - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

4 Page of HBAC, material deprivation, ethnicity, smoking, duration and type of diabetes, atrial fibrillation, cardiovascular disease and chronic renal disease. The equations had good performance in CPRD for women (R of.%; D statistic of.; ROC statistic of 0.) and men (.%,.; 0. respectively). Conclusions We have developed and externally validated risk prediction equations to quantify absolute risk of heart failure in men and women with diabetes. They can be used to identify patients at high risk of heart failure for prevention or assessment. Strengths and limitations of this study The algorithms provide valid measures of absolute risk of heart failure in patients with type or type diabetes as shown by the performance in a separate validation cohort. Key strengths include size, duration of follow up, representativeness, and lack of selection, recall and respondent bias. The study has good face validity since it has been conducted in the setting where the majority of patients in the UK are assessed, treated and followed up. The study used linked hospital and mortality and is therefore likely to have picked up the majority of heart failure diagnoses. The algorithms do not include some biomarkers for heart failure such as BNP as these are not routinely recorded in electronic health records. - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

5 Page of Funding This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors. Web calculator Here is a simple web calculator to calculate the absolute risk of heart failure among patients with diabetes. This will be publically available alongside the paper. It also has the open source software for download. URL Username reviewer Password toflashamber It also has the open source software for download which can be viewed here : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

6 Page of Introduction Cardiovascular disease is a major cause of morbidity and decreased life expectancy in patients with diabetes. Clinical guidelines recommend cardiovascular risk assessment in patients with diabetes[ ] to assist with prevention. Cardiovascular risk assessment tools, including Framingham[] and QRISK[], are used both for communicating risk to patients with diabetes and to guide public health decisions[ ]. Considerable attention, in clinical trials, guidelines and observational studies[], is focused on predicting and preventing cardiovascular disease (particularly coronary heart disease and stroke) reflecting perceptions that it is the main disease burden for diabetes[]. Less emphasis has been given to the development of risk assessment methods to predict heart failure among people with diabetes although it is recognised to be a substantial cause of morbidity and mortality requiring different investigations and treatments[-]. It is associated with significant morbidity and health care costs[]. Heart failure has also been associated as an adverse event with selected anti-hyperglycaemic agents used to treat heart failure []. Once heart failure is present in individuals with diabetes mellitus, there is a 0- fold increase in mortality and a year survival of only.%, a prognosis worse than metastatic breast cancer[0]. We have developed and validated various risk prediction equations including QRISK[] for cardiovascular disease, and QStroke[] for stroke, designed to quantify an individual s absolute risk of these clinically important outcomes based on risk factors recorded in linked electronic health records. These risk equations are used internationally in patients with and without diabetes to assist in decisions regarding further investigation, treatment or referral. Whilst there have been several risk scores to predict incident heart failure such as the - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

7 Page of Framingham[] and ARIC scores[], these are not based on contemporaneous ethnically diverse populations, only cover limited age ranges (for example - years) or time periods[] and include very few patients with diabetes[]. The objective of the present study is to derive and externally validate new risk prediction equations to quantify 0 year absolute risk of heart failure in patients with diabetes using linked electronic health records. The purpose of the equations is to better quantify absolute risk of heart failure in order to prompt (a) closer management of modifiable risk factors (b) earlier diagnosis of heart failure in at risk patients and (c) to provide better information for patients and doctors to inform treatment decisions about anti-hyperglycaemic agents which might also inadvertently increase risk of heart failure. Methods. Study design and data source We undertook a cohort study to derive and validate the risk equations in a large population of primary care patients using the large UK QResearch primary care database (version, QResearch is a continually updated patient level pseudonymised database with event level data extending back to. QResearch includes clinical and demographic data from over,000 general practices covering a population of > 0 million patients, collected in the course of routine healthcare by general practitioners and associated staff. The primary care data includes demographic information, diagnoses, prescriptions, referrals, laboratory results and clinical values. Diagnoses are recorded using the Read code classification. QResearch has been used for a wide range of clinical research - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

8 Page of including the development and validation of risk prediction models[ ]. The primary care data is linked at individual patient level to Hospital Episode Statistics (HES), and mortality records from the Office for National Statistics (ONS). HES provides details of all National Health Service (NHS) inpatient admissions since including primary and secondary causes coded using the ICD-0 classifications. ONS provides details of all deaths in England with primary and underlying causes, also coded using the ICD-0 classification. Patient records are linked using a project specific pseudonymised NHS number which is valid and complete for.% of primary care patients,.% for ONS mortality records and % for hospital admissions records. We included all QResearch practices in England who had been using their Egton Medical Information Systems (EMIS) computer system for at least a year. We randomly allocated three quarters of these practices to the derivation dataset and the remaining quarter to a validation dataset. In both datasets we identified open cohorts of patients aged - years registered with eligible practices between 0 Jan and st July 0. We then selected patients with diabetes if they had a Read code for diabetes or more than one prescription for insulin or oral hypoglycaemics. We classified patients as having type diabetes if they had been diagnosed under the age of and prescribed insulin[]. We excluded patients without a postcode related deprivation score and patients with an existing diagnosis of heart failure. We determined an entry date to the cohort for each patient, which was the latest of the following dates: th birthday, date of registration with the practice plus one year, date on which the practice computer system was installed plus one year, date of diagnosis of diabetes and the beginning of the study period (0 January ). Patients were - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

9 Page of censored at the earliest date of the diagnosis of heart failure, death, de-registration with the practice, last upload of computerised data, or the study end date ( st August 0). We undertook a second external validation using general practices in England contributing to the Clinical Research Practice Datalink (CPRD). CPRD is a similar database to QResearch except that it is derived from practices using a different clinical computer system. We used the subset of CPRD practices with linked to ONS mortality and hospital admissions. We used the same definitions for selecting a validation cohort as for QResearch except that study end date was st August 0, the latest date for which linked data were available.. Outcome We classified patients as having incident heart failure if there was a record of the relevant diagnosis either in their primary care record, their linked hospital record or mortality record. We used Read codes to identify recorded clinical diagnoses of heart failure from the primary care record (G%,Gyy,GyyA,f,g,h,i). We used ICD-0 clinical codes (I0, I0, I, I0) to identify incident cases of heart failure from hospital and mortality records. We used the earliest recorded date of heart failure on any of the three data sources as the index date for the diagnosis of heart failure.. Predictor variables We examined the following predictor variables based on established risk factors for cardiovascular disease and heart failure[] []. We focused on variables likely to be recorded as coded data in the patient s electronic record. For continuous variables, we used the closest values recorded prior to the baseline date or within the months after the cohort entry date. All other predictor variables were based on the latest information - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

10 Page of recorded in the primary care record before entry to the cohort. We didn t include or natriuretic peptide levels since this is not routinely measured or recorded in electronic health records. Age at entry to cohort (continuous) Type of diabetes: type or type Number of years since diagnosis of diabetes (< year; -; -; -0; years) Smoking status (non-smoker; ex-smoker; light(- cigarettes/day); moderate(0- /day); heavy (0+/day) Ethnic group (White/not recorded, Indian, Pakistani, Bangladeshi, Other Asian, Black Caribbean, Black African, Chinese, Other) [] Townsend deprivation score (continuous) [] where a higher score indicates greater levels of material deprivation. Family history of premature coronary heart disease (in a first degree relative aged less than 0) [] Chronic renal disease[] Cardiovascular disease[] Atrial fibrillation[] Treated hypertension[] Rheumatoid arthritis[] Body mass index kg/m (continuous)[] Cholesterol/HDL level (continuous) [] Systolic blood pressure (continuous)[] Glycosylated haemoglobin HBAC mmol/mol (continuous)[ ] - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

11 Page 0 of Derivation of the models We developed the risk prediction equations in the derivation cohort using established methods [0]. We used fractional polynomials[] on complete cases to model non-linear risk relationships with continuous variables if appropriate (age, body mass index, systolic blood pressure, cholesterol/hdl ratio, HBAC). We used multiple imputation to replace missing values for continuous values and smoking status and used these values in our main analyses[ ]. We carried out 0 imputations. We used Cox s proportional hazards models to estimate the coefficients for each risk factor for men and women separately. We used Rubin s rules to combine the regression coefficients across the imputed datasets[]. We fitted full models initially then retained variables if they had a hazard ratio of < 0.0 or >.0 (for binary variables) and were statistically significant at the 0.0 level. We examined interactions between predictor variables and age and included these where they were significant, plausible and improved model fit. We used regression coefficients for each variable from the final model as weights which we combined with the baseline survivor function evaluated up to years to derive risk equations over a period of years of follow-up[]. This enabled us to derive risk estimates for each year of follow-up, with a specific focus on 0 year risk estimates. We estimated the baseline survivor function based on zero values of centred continuous variables, with all binary predictor values set to zero. Model fit was assessed by measuring the AIC and BIC values for each imputed set of data.. Validation of the models We used multiple imputation in the two validation cohorts to replace missing values for continuous variables and smoking status. We carried out 0 imputations. We applied the risk equations for men and women obtained from the derivation cohort to the validation : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

12 Page of cohorts and calculated measures of discrimination. We calculated R values (explained variation in time to diagnosis of heart failure[]), D statistics[] (a measure of discrimination where higher values indicate better discrimination) and the area under the receiver operating characteristic curve (ROC statistic) at 0 years and combined the model performance measures across datasets using Rubin s rules. We assessed calibration (comparing the mean predicted risks at 0 years with the observed risk by tenth of predicted risk). The observed risks were obtained using Kaplan-Meier estimates evaluated at 0 years. We applied the equations to the validation cohorts to define thresholds for the 0% of patients at highest estimated risk of heart failure at 0 years. We used all the available data on each database to maximise the power and also generalisability of the results. We used STATA (version.) for all analyses. The TRIPOD statement was adhered to in reporting[]. Results. Overall study population Overall, 0 QResearch practices in England met our practice inclusion criteria, of which were randomly assigned to the derivation dataset with the remaining practices assigned to the validation cohort. We identified, patients aged - years with diabetes in the derivation cohort. We excluded patients (0.%) without a recorded Townsend deprivation score and, (.%) with a diagnosis of heart failure at baseline, leaving,0 for analysis. - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

13 Page of We identified, patients aged - years with diabetes in the QResearch validation cohort. We excluded patients (0.%) without a recorded Townsend deprivation score and, (.%) with a diagnosis of heart failure at baseline leaving,0 for validation. We identified 0,00 patients aged - years with diabetes in the CPRD validation cohort from the practices with linked hospital and mortality data. We excluded, patients (.%) with a diagnosis of heart failure at baseline leaving,0 for validation.. Baseline characteristics Table shows baseline characteristics of,0 patients with diabetes in the derivation cohort free from heart failure at study entry. Of these,, (.%) had type diabetes, and,0 (.%) had type diabetes,.% had treated hypertension,.% had cardiovascular disease,.0% had chronic renal disease and.% had atrial fibrillation. Just over half (.%) been diagnosed with diabetes for less than a year at cohort entry,.0% had been diagnosed for - years,.% for - years,.% for -0 years and.% for or more years before cohort entry. Smoking was recorded in.% of patients, ethnicity in.%, body mass index in 0.%, systolic blood pressure in.0%, HBAC in 0.% and cholesterol/hdl in %..% of QResearch patients had complete data for all variables. Baseline characteristics for patients in the QResearch validation cohort were similar to corresponding values in the derivation cohort. Baseline characteristics of the CPRD validation cohort were also similar except recording of ethnicity (.0%) and HBAC (.%) was substantially lower in CPRD..% of CPRD patients had complete data for all variables. - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

14 Page of Primary outcome of heart failure Table shows the number of incident cases of heart failure during follow-up and the directly age standardized incidence rates in each cohort. There were,0 cases of heart failure in the QResearch derivation cohort,, in the QResearch validation cohort and, in the CPRD validation cohort.. Predictor variables Table shows the adjusted hazard ratios for variables in the final models for men and women in the derivation cohort which included: age, body mass index, systolic blood pressure, cholesterol/hdl, HBAC, deprivation, duration and type of diabetes, smoking status, ethnicity, atrial fibrillation, cardiovascular disease and chronic renal disease. Increasing duration of diabetes was associated with increased heart failure risk despite adjustment for current age and other risk factors. Non-white ethnic groups tended to have a lower risk of heart failure compared with those whose ethnic group was either white or not recorded. There was a dose-response relationship for smoking with heavy smokers having the highest risk of heart failure. In women, type diabetes had a % higher risk of heart failure compared with type ; atrial fibrillation a % higher risk; cardiovascular disease a % increased risk and chronic renal disease a % increased risk. The pattern was similar for men. Figure shows the adjusted hazard ratios using fractional polynomial terms for continuous variables. Increasing age, body mass index, systolic blood pressure and HBAC were all associated with increasing hazard ratios of heart failure. The full equation for the model is published as open source software at (username reviewer password toflashamber). - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

15 Page of Validation.. Discrimination Table shows the performance of each equation in each validation cohort for women and men separately. In the CPRD validation cohort, in women the algorithm explained.% of the variation in time to diagnosis of heart failure (R ) and discrimination was good with a D statistic of. and ROC value of 0.. Corresponding values for men were.%,. and 0.. The results of the validation in the QResearch validation cohort were very similar... Calibration Figure shows the mean predicted risks and observed risks of heart failure at 0 years by tenth of predicted risk applying the equations to all men and women in the QResearch and CPRD validation cohorts. There was close correspondence between the mean predicted risks and the observed risks within each model tenth in women and men indicating that the equations were well calibrated across both validation cohorts... Performance at threshold for the 0% at highest risk The sensitivity, specificity and observed risk for the 0% of women at highest predicted risk of heart failure on the QResearch validation cohort (i.e. cut-off for 0 year risk of.%) were.%,.%, and.0% respectively. The corresponding figures for men (using a 0 year risk threshold of.%) were 0.%,.%, and.%. The corresponding results for women on CPRD were (.%,.%,.%) and for men were (.%,.%,.%). Figure shows a clinical example of the implementation of the equation as a web calculator. For international users the postcode field can be left blank. - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

16 Page of Discussion. Key findings We have developed and externally validated risk prediction equations to quantify absolute 0 year risk of heart failure in men and women with either type or type diabetes. To our knowledge, this is the first equation to predict 0 year risk of heart failure in men and women with diabetes based on a very large representative ethnically diverse contemporaneous population. The equations include variables: age, BMI, systolic blood pressure, cholesterol/hdl ratio, HBAC, material deprivation, ethnicity, smoking, duration and type of diabetes, atrial fibrillation, cardiovascular disease and chronic renal disease. The equations were well calibrated and had good discrimination with ROC values exceeding 0. in both validation cohorts. The purpose of the equation is to better quantify absolute risk of heart failure in order to prompt (a) closer management of modifiable risk factors (b) earlier diagnosis and treatment of heart failure in at risk patients and (c) to provide better information for patients and doctors to inform the treatment decisions about antihyperglycaemic agents which might also inadvertently increase risk of heart failure.. Comparisons with the literature We included established risk factors in our equation and demonstrated hazard ratios similar in both magnitude and direction to those reported for heart failure elsewhere[] [] which increases the clinical face validity of the equations. Cardiovascular disease was associated with a -fold increased risk of heart failure as in the study of, patients with diabetes by Nichol et al[]. Chronic renal disease, increasing body mass index, increasing systolic blood pressure and HBAC level[] and a longer duration of diabetes were - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

17 Page of associated with higher risk of heart failure[]. Compared with the ARIC study based on, US patients aged - years with heart failure events[], we have included a much larger, more contemporaneous ethnically diverse population of patients with either type or type diabetes spanning a wider age range. We have included additional predictors such as atrial fibrillation, chronic renal disease and HBAC known to be associated with risk of heart failure. Our ROC values were similar to those from ARIC[] and higher than that reported for the published Framingham heart failure score[].. Methodological considerations The methods used to derive and validate these models are the same as for a range of other clinical risk prediction tools derived from the QResearch database[ -]. The strengths and limitations of the approach have been discussed in detail[ 0] including information on multiple imputation of missing data. In summary, key strengths include size, duration of follow up, representativeness, and lack of selection, recall and respondent bias. UK general practices have good levels of accuracy and completeness in recording clinical diagnoses and prescribed medications []. Our database has linked hospital and mortality records for nearly all patients and is therefore likely to have picked up the majority of cases of diagnosed heart failure thereby minimising ascertainment bias. We undertook two validations, one using a separate set of practices and patients contributing to QResearch and the other using a fully external set of practices contributing to CPRD. The results of both validations were extremely similar which is consistent with previous validation studies showing comparable performance using different populations[0 ]. Whilst we have derived and validated the score using UK datasets, the score can be used internationally by leaving the postcode deprivation field blank (in which case a mean value - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

18 Page of would be substituted). Best practice would be to undertake a validation using a relevant population before local use to ensure good calibration in the applicable population. Limitations of our study include the lack of formal adjudication of diagnoses, and potential for bias due to missing data. However we think our ascertainment of diagnosed heart failure is likely to be high given the combination of the three linked data sources although we may have missed clinically silent heart failure. We didn t include or natriuretic peptide levels since this is not routinely measured or recorded in electronic health records. We have not provided definite comment on what threshold of absolute risk should be used to define a high risk group as that would require (a) the balance of risks and benefits for individuals and (b) cost-effectiveness analyses which are outside the scope of this study. Conclusion We have developed and validated a new risk prediction equation to quantify the absolute risk of heart failure in patients with type or type diabetes. It can be used to identify patients with diabetes at high risk of heart failure for further assessment and proactive treatment. Other information.. Acknowledgements We acknowledge the contribution of EMIS practices who contribute to the QResearch and EMIS for expertise in establishing, developing and supporting the database. - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

19 Page of Approvals: The project was reviewed in accordance with the QResearch agreement with Trent Multi- Centre Ethics Committee [reference 0//0]. The project was reviewed by the independent scientific committee of the Clinical Research Practice Datalink [reference _0]... Contributorship JHC initiated the study, undertook the literature review, data extraction, data manipulation and primary data analysis and wrote the first draft of the paper. CC contributed to the design, analysis, interpretation and drafting of the paper... Competing Interests All authors have completed the Unified Competing Interest form at (available on request from the corresponding author) and declare: JHC is professor of clinical epidemiology at the University of Nottingham and codirector of QResearch a not-for-profit organisation which is a joint partnership between the University of Nottingham and Egton Medical Information Systems (leading commercial supplier of IT for 0% of general practices in the UK). JHC is also a paid director of ClinRisk Ltd which produces open and closed source software to ensure the reliable and updatable implementation of clinical risk equations within clinical computer systems to help improve patient care. CC is associate professor of Medical Statistics at the University of Nottingham and a paid consultant statistician for ClinRisk Ltd. This work and any views expressed within it are solely those of the co-authors and not of any affiliated bodies or organisations. - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

20 Page of Copyright The Corresponding Author has the right to grant on behalf of all authors and does grant on behalf of all authors, an exclusive licence (or non-exclusive for government employees) on a worldwide basis to the BMJ Publishing Group Ltd, and its Licensees to permit this article (if accepted) to be published in BMJ editions and any other BMJPGL products and to exploit all subsidiary rights, as set out in our licence (bmj.com/advice/copyright.shtml)... Data Sharing The equations presented in this paper will be released as Open Source Software under the GNU lesser GPL v. The open source software allows use without charge under the terms of the GNU lesser public license version. Closed source software can be licensed at a fee. - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

21 Page 0 of References. National Clinical Guideline Centre. Lipid modification: cardiovascular risk assessment and the modification of blood lipids for the primary and secondary prevention of cardiovascular disease. London, 0:.. Goff DC, Lloyd-Jones DM, Bennett G, et al. 0 ACC/AHA Guideline on the Assessment of Cardiovascular Risk: A Report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. Circulation 0 doi: 0./0.cir [published Online First: Epub Date].. Anderson KM, Odell PM, Wilson PWF, Kannel WB. Cardiovascular disease risk profiles. American Heart Journal ;:-. Hippisley-Cox J, Coupland C, Vinogradova Y, Robson J, Brindle P. Performance of the QRISK cardiovascular risk prediction algorithm in an independent UK sample of patients from general practice: a validation study. Heart 00;:- doi: 0./hrt.00.0[published Online First: Epub Date].. van der Leeuw J, van Dieren S, Beulens JWJ, et al. The validation of cardiovascular risk scores for patients with type diabetes mellitus. Heart 0;0():- doi: 0./heartjnl-0-00[published Online First: Epub Date].. Shah AD, Langenberg C, Rapsomaniki E, et al. Type diabetes and incidence of cardiovascular diseases: a cohort study in million people. The Lancet Diabetes & Endocrinology;():0- doi: 0.0/S-()0-0[published Online First: Epub Date].. Health and Social Care Information Centre. National Diabetes Audit 0-0 Report Complications and Mortality, 0:.. Yang X, Ma RC, So WY, et al. Development and validation of a risk score for hospitalization for heart failure in patients with Type diabetes mellitus. Cardiovascular diabetology 00;: doi: 0./-0--[published Online First: Epub Date].. Home PD, Pocock SJ, Beck-Nielsen H, et al. Rosiglitazone evaluated for cardiovascular outcomes in oral agent combination therapy for type diabetes (RECORD): a multicentre, randomised, open-label trial. Lancet 00;():- doi: 0.0/S00-(0)0-[published Online First: Epub Date]. 0. Khan SS, Butler J, Gheorghiade M. Management of comorbid diabetes mellitus and worsening heart failure. JAMA 0;():-0 doi: 0.00/jama.0.[published Online First: Epub Date].. Hippisley-Cox J, Coupland C, Brindle P. Derivation and validation of QStroke score for predicting risk of ischaemic stroke in primary care and comparison with other risk scores: a prospective open cohort study. BMJ 0;:f doi: 0./bmj.f[published Online First: Epub Date].. Kannel WB, D'Agostino RB, Silbershatz H, Belanger AJ, Wilson PF, Levy D. Profile for estimating risk of heart failure. Archives of Internal Medicine ;():-0 doi: 0.00/archinte...[published Online First: Epub Date].. Agarwal SK, Chambless LE, Ballantyne CM, et al. Prediction of Incident Heart Failure in General Practice: The Atherosclerosis Risk in Communities (ARIC) Study. Circulation: Heart Failure 0;():- doi: 0./circheartfailure..[published Online First: Epub Date].. Hippisley-Cox J, Coupland C, Vinogradova Y, et al. Predicting cardiovascular risk in England and Wales: prospective derivation and validation of QRISK. BMJ 00:bmj.0.. doi: 0./bmj.0..[published Online First: Epub Date].. Hippisley-Cox J, Pringle M. Prevalence, Care and Outcomes for patients with diet controlled diabetes in general practice: cross sectional survey. Lancet 00;: : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

22 Page of Clarke PM, Gray AM, Briggs A, et al. A model to estimate the lifetime health outcomes of patients with type diabetes: the United Kingdom Prospective Diabetes Study (UKPDS) Outcomes Model (UKPDS no. ). Diabetologia 00;(0):- doi: 0.00/s z[published Online First: Epub Date].. Davis TME, Coleman RL, Holman RR, Group U. Ethnicity and long-term vascular outcomes in Type diabetes: a prospective observational study (UKPDS ). Diabetic Medicine 0;():00-0 doi: 0./dme.[published Online First: Epub Date].. Nichols GA, Gullion CM, Koro CE, Ephross SA, Brown JB. The Incidence of Congestive Heart Failure in Type Diabetes: An update. Diabetes Care 00;():- doi: 0./diacare...[published Online First: Epub Date].. Stratton IM, Adler AI, Neil HA, et al. Association of glycaemia with macrovascular and microvascular complications of type diabetes (UKPDS ): prospective observational study. BMJ 000;():0-0. Collins GS, Altman DG. Predicting the 0 year risk of cardiovascular disease in the United Kingdom: independent and external validation of an updated version of QRISK. BMJ 0;:e doi: 0./bmj.e[published Online First: Epub Date].. Royston P, Ambler G, Sauerbrei W. The use of fractional polynomials to model continuous risk variables in epidemiology. Int J Epidemiol ;:-. Steyerberg EW, van Veen M. Imputation is beneficial for handling missing data in predictive models. J Epidemiol Community Health 00;0:. Moons KGM, Donders RART, Stijnen T, Harrell FJ. Using the outcome for imputation of missing predictor values was preferred. J Epidemiol Community Health 00;:0. Rubin DB. Multiple Imputation for Non-response in Surveys. New York: John Wiley,.. Hosmer D, Lemeshow S. Applied Logistic Regressopm. New York: John Wiley & Sons, Inc.,.. Royston P. Explained variation for survival models. Stata J 00;:-. Royston P, Sauerbrei W. A new measure of prognostic separation in survival data. Stat Med 00;:-. Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): The TRIPOD StatementThe TRIPOD Statement. Annals of Internal Medicine 0;():- doi: 0./M-0[published Online First: Epub Date].. Hippisley-Cox J, Coupland C. Predicting risk of emergency admission to hospital using primary care data: derivation and validation of QAdmissions score. 0;():e00 doi: 0./bmjopen-0-00[published Online First: Epub Date]. 0. Hippisley-Cox J, Coupland C. Development and validation of risk prediction algorithms to estimate future risk of common cancers in men and women: prospective cohort study. 0;() doi: 0./bmjopen-0-00[published Online First: Epub Date].. Hippisley-Cox J, Coupland C. Derivation and validation of updated QFracture algorithm to predict risk of osteoporotic fracture in primary care in the United Kingdom: prospective open cohort study. BMJ 0;(may ):e-e doi: 0./bmj.e[published Online First: Epub Date].. Majeed A. Sources, uses, strengths and limitations of data collected in primary care in England. Health Statistics Quarterly 00():-. Collins GS, Mallett S, Altman DG. Predicting risk of osteoporotic and hip fracture in the United Kingdom: prospective independent and external validation of QFractureScores. BMJ 0;:d - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

23 Page of Table Baseline characteristics of patients with diabetes aged - years and without heart failure at baseline in the QResearch derivation cohort and both validation cohorts. Values are numbers (percentages) unless stated otherwise. QResearch Derivation cohort QResearch Validation cohort CPRD Validation cohort total patients without heart failure 0 (00) 0 (00) 0 (00) women (.) 0 (.) (.) men 0 (.) (.) (.) Type diabetes (.) (.) (.) type diabetes 0 (.) 0 (.) 0 (.) Years since diagnosis of diabetes < year (.) (.) 0 (.) - years (.0) 0 (.) 0 (.) - years 0 (.) (.) 0 (.) -0 years (.) 0 (.) (.) years 0 (.) 0 (.) (.) mean age (SD) 0.0 (.) 0. (.).0 (.) mean Townsend deprivation score (SD) 0. (.) 0. (.) -0. (.) Ethnicity Ethnicity recorded (.) 00 (.) 0 (.0) Ethnicity not recorded 0 (.) (.) 0 (.0) White/not recorded ± 0 (.) (.) (.) Indian (.) (.) (.) Pakistani 0 (.) (.) (0.) Bangladeshi (.) (.) 0 (0.) Other Asian 0 (.) (.) 0 (.0) Black Caribbean (.) (.0) (0.) Black African (.) 0 (.) (0.) Chinese (0.) 0 (0.) 0 (0.) Other (.0) (.) (.) smoking status Smoking status recorded (.) 0 (.) (.) non smoker 0 (0.) (.) (.) Ex-smoker 0 (.) 0 (.) (0.) light smoker (-/day) (0.) (0.0) 00 (.) moderate smoker (0-/day) (.) (.) 0 (.0) heavy smoker (0+/day) 00 (.) 0 (.) 0 (.) Smoker amount not recorded 0 (0.0) 0 (0.0) (.) Family history and co-morbidity family history premature CHD (.) 0 (.) 0 (.) rheumatoid arthritis 0 (.) 0 (.) (.) atrial fibrillation (.) (.) (.) cardiovascular disease (.) (.) 00 (.) treated hypertension (.) 0 (.) 0 (0.) - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

24 Page of chronic renal disease 0 (.0) (.0) 0 (0.) clinical values HBAC recorded 0 (0.) 0 (0.) 0 (.) mean HBAC (SD). (.). (.). (.) BMI recorded 0 (0.) (0.) 0 (.) mean BMI (SD) 0. (.) 0. (.) 0. (.) cholesterol /HDL ratio recorded 0 (.) (.) (0.0) mean cholesterol/hdl ratio (SD). (.). (.). (.) systolic blood pressure recorded (.0) (.) 0 (.) mean systolic blood pressure (SD). (.). (.) 0. (.) increasinglevels of Townsend scores indicating increasing levels of material deprivation. ± includes patients who were either recorded as a white ethnicity or those without any ethnicity recorded - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

25 Page of Table Numbers of incident cases of heart failure during follow-up and age standardised incidence rates per,000 person years in men and women with diabetes aged - years in the derivation cohort and validation cohorts HF cases QResearch derivation cohort rate per,000 person years (% CI) HF cases QResearch validation cohort rate per,000 person years (% CI) Total HF cases CPRD validation cohort rate per,000 person years (% CI) women 0,0.0 (. to. ),. (. to.),.0 (. to. ) Men,0. (. to.),. (. to.),. (. to. ) HF= heart failure Notes: Rates were directly age standardised to the overall age distribution of patients aged to within the QResearch derivation cohort in -year age bands. on 0 March 0 by guest. Protected by copyright. - : first published as 0./bmjopen on September 0. Downloaded from

26 Page of Table Adjusted hazard ratios with % confidence intervals for heart failure in men and women in the derivation cohort. For fractional polynomial terms see footnotes and figure. adjusted hazard ratio (% CI) women men age.0 (.0 to.0).0 (.0 to.0) Cholesterol/HDL ratio.0 (.0 to.0).0 (.0 to.0) Townsend deprivation score,. (. to.). (. to.0) duration of diabetes at baseline < year - years. (. to. ). (. to. ) - years. (. to. ). (. to. ) -0 years. (. to. ). (. to. ) years. (. to.0 ). (. to. ) smoking status non-smoker Ex-smoker.0 (.0 to. ).0 (0. to.0 ) light smoker (-/day). (. to. ). (. to. ) moderate smoker (0-/day). (. to. ). (. to. ) heavy smoker (0+/day). (. to. ). (. to. ) ethnicity white/not recorded Indian.0 (0. to. ) 0. (0. to.0 ) Pakistani.0 (0. to. ) 0. (0.0 to.0 ) Bangladeshi 0. (0. to. ). (.00 to.0 ) Other Asian 0. (0. to. ) 0. (0. to 0. ) Black Caribbean 0. (0. to 0. ) 0. (0. to 0. ) Black African 0. (0. to. ) 0. (0. to 0. ) Chinese 0. (0. to. ) 0. (0. to 0.0 ) Other ethnic group 0. (0. to 0. ) 0. (0. to 0. ) co-morbidity type diabetes. (.0 to. ). (.0 to. ) atrial fibrillation. (. to.0 ).0 (. to. ) cardiovascular disease. (. to.0 ). (. to. ) chronic renal disease. (.0 to. ). (. to.0 ) Notes heart failure model in women included body mass index ( FP terms --.), systolic blood pressure ( FP terms -.0), hbac ( FP terms --) heart failure model in men included body mass index ( FP terms -0), systolic blood pressure ( FP terms 0.), hbac ( FP terms - -) - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

27 Page of adjusted hazard ratio is per unit increase the Townsend deprivation score ranges between -(most affluent) and + (most deprived). adjusted hazard ratio compared with type diabetes adjusted hazard ratio compared with patients without the condition at baseline : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

28 Page of Table Performance of the equations in men and women in QResearch validation cohort and CPRD validation cohort statistic QResearch validation cohort Mean (% CI) CPRD validation cohort Mean (% CI) women D statistic. (.0 to. ). (. to. ) R. (.0 to. ). (. to. ) ROC 0.0 (0. to 0. ) 0. (0. to 0. ) men D statistic. (. to. ). (. to. ) R 0.0 (. to. ). (. to 0.00 ) ROC 0. (0. to 0. ) 0. (0. to 0. ) Notes on understanding validation statistics: Discrimination is the ability of the risk prediction model to differentiate between patients who experience a admission event during the study and those who do not. This measure is quantified by calculating the area under the receiver operating characteristic curve (ROC) statistic; where a value of 0. represents chance and represents perfect discrimination. The D statistic is also a measure of discrimination which is specific to censored survival data. As with the ROC, higher values indicate better discrimination. R measures explained variation and higher values indicate more variation is explained. - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

29 Page of Figure show graphs of the adjusted hazard ratios for the fractional polynomial terms for age, body mass index, HBAC and systolic blood pressure in the derivation cohort. Figure shows the mean predicted risks and observed risks of heart failure at 0 years by tenth of predicted risk applying the equations to all men and women in the QResearch and CPRD validation cohorts. Figure shows web calculator applied to an example patient : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

30 Page of xmm ( x DPI) - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

31 Page 0 of xmm ( x DPI) - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

32 Page of Figure shows web calculator applied to an example patient xmm ( x DPI) - : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

33 Page of Section/topic Title and abstract Item Development or validation? Title D;V Checklist item Identify the study as developing and/or validating a multivariable prediction model, the target population, and the outcome to be predicted Abstract D;V Introduction Background and objectives Methods Source of data a b a b D;V D;V D;V D;V Participants a D;V Provide a summary of objectives, study design, setting, participants, sample size, predictors, outcome, statistical analysis, results, and conclusions Explain the medical context (including whether diagnostic or prognostic) and rationale for developing or validating the multivariable prediction model, including references to existing models Specify the objectives, including whether the study describes the development or validation of the model, or both Describe the study design or source of data (for example, randomised trial, cohort, or registry data), separately for the development and validation data sets, if applicable Specify the key study dates, including start of accrual; end of accrual; and, if applicable, end of follow-up Specify key elements of the study setting (for example, primary care, secondary care, general population) including number and location of centres Julia Hipisley-Cox, Page : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.

34 Page of Section/topic Outcome Predictors Page Item b c Development or validation? D;V D;V Checklist item Describe eligibility criteria for participants Give details of treatments received, if relevant a b a b D;V D;V D;V D;V Sample size D;V Missing data D;V Statistical analysis methods 0a 0b 0c 0d 0e D D V D;V V Clearly define the outcome that is predicted by the prediction model, including how and when assessed Report any actions to blind assessment of the outcome to be predicted Clearly define all predictors used in developing the multivariable prediction model, including how and when they were measured Report any actions to blind assessment of predictors for the outcome and other predictors Explain how the study size was arrived at. Describe how missing data were handled (for example, completecase analysis, single imputation, multiple imputation) with details of any imputation method Describe how predictors were handled in the analyses Specify type of model, all modelbuilding procedures (including any predictor selection), and method for internal validation For validation, describe how the predictions were calculated Specify all measures used to assess model performance and, if relevant, to compare multiple models Describe any model updating (for example, recalibration) arising - n/a n/a n/a Page : first published as 0./bmjopen on September 0. Downloaded from on 0 March 0 by guest. Protected by copyright.