Acetaminophen, aspirin and progression of advanced chronic kidney disease

Transcription

1 1908 M. Evans et al. Nephrol Dial Transplant (2009) 24: doi: /ndt/gfn745 Advance Access publication 20 January Rao M, Guo D, Perianayagam MC et al. Plasma interleukin-6 predicts cardiovascular mortality in hemodialysis patients. AmJKidneyDis 2005; 45: Zimmermann J, Herrlinger S, Pruy A et al. Inflammation enhances cardiovascular risk and mortality in hemodialysis patients. Kidney Int 1999; 55: Zoccali C, Tripepi G, Mallamaci F. Predictors of cardiovascular death in ESRD. Semin Nephrol 2005; 25: Stenvinkel P. Inflammation in end-stage renal disease a fire that burns within. Contrib Nephrol 2005; 149: Ridker PM, Rifai N, Rose L et al. Comparison of C-reactive protein and low-density lipoprotein cholesterol levels in the prediction of first cardiovascular events. NEnglJMed2002; 347: Whicher J, Biasucci L, Rifai N. Inflammation, the acute phase response and atherosclerosis. Clin Chem Lab Med 1999; 37: Bolton CH, Downs LG, Victory JG et al. Endothelial dysfunction in chronic renal failure: roles of lipoprotein oxidation and proinflammatory cytokines. Nephrol Dial Transplant 2001; 16: Pereira BJ, Shapiro L, King AJ et al. Plasma levels of IL-1 beta, TNF alpha and their specific inhibitors in undialyzed chronic renal failure, CAPD and hemodialysis patients. Kidney Int 1994; 45: Knight EL, Rimm EB, Pai JK et al. Kidney dysfunction, inflammation, and coronary events: a prospective study. J Am Soc Nephrol 2004;15: Tonelli M, Sacks F, Pfeffer M et al. Cholesterol and Recurrent Events (CARE) Trial Investigators. Biomarkers of inflammation and progression of chronic kidney disease. Kidney Int 2005; 68: Martinsons A, Rudzite V, Cernevskis H et al. The influence of L-tryptophan peroral load on glomerular filtration rate in chronic glomerulonephritis and chronic renal failure. Adv Exp Med Biol 2003; 527: Braun D, Longman RS, Albert ML. A two-step induction of indoleamine 2,3 dioxygenase (IDO) activity during dendritic-cell maturation. Blood 2005; 106: Received for publication: ; Accepted in revised form: Acetaminophen, aspirin and progression of advanced chronic kidney disease Marie Evans 1, Carl Michael Fored 2, Rino Bellocco 3,4, Garrett Fitzmaurice 5, Jon P. Fryzek 6,7, Joseph K. McLaughlin 6,7,OlofNyrén 3,7 and Carl-Gustaf Elinder 1 1 Nephrology Unit, Department of Clinical Sciences Intervention and Technology, Karolinska Institutet and University Hospital, Stockholm, Sweden, 2 Clinical Epidemiology Unit, Department of Medicine, Karolinska Institutet, Stockholm, Sweden, 3 Department of Medical Epidemiology and Biostatistics, Karolinska Institutet, Stockholm, Sweden, 4 Department of Statistics, University of Milano-Bicocca, Milan, Italy, 5 Department of Psychiatry, Harvard Medical School, Boston, MA, USA, 6 International Epidemiology Institute, Rockville, MD and 7 Department of Medicine, Vanderbuilt University Medical Center, Nashville, TN, USA Correspondence and offprint requests to: Marie Evans; marie.evans@ ki.se Abstract Background. Although many studies have investigated the possible association between analgesic use (acetaminophen and aspirin) and the development of chronic kidney disease (CKD), the effect of analgesics on the progression of established CKD of any cause has not yet been investigated. Methods. In this population-based Swedish cohort study, we investigated the decline over 5 7 years in estimated glomerular filtration rate (egfr) among 801 patients with incident, advanced CKD (serum creatinine >3.4 mg/dl for men, >2.8 mg/dl for women for the first time) and with different analgesic exposures. Lifetime analgesic use and current regular use were ascertained through in-person interviews at inclusion while data on analgesic use during the follow-up was abstracted from the medical records at the end of the study period. A linear regression slope, based on their egfr values during the follow-up, provided a summary of within-individual change. In the final multivariate analyses, a linear mixed effects model was implemented to assess the relation of analgesic use and change in egfr over time. Results. The progression rate for regular users of acetaminophen was slower than that for non-regular users (regular users progressed 0.93 ml/min/1.73 m 2 per year slower than non-regular users; 95% CI 0.03, 1.8). For regular users of aspirin, the progression rate was significantly slower than that for non-regular users (regular users progressed 0.80 ml/min/1.73 m 2 per year slower than nonregular users; 95% CI 0.1, 1.5). Different levels of lifetime cumulative dose of acetaminophen and aspirin did not significantly affect the progression rate. Conclusion. We suggest that single substance acetaminophen and aspirin may be safe to use by patients C The Author [2009]. Published by Oxford University Press on behalf of ERA-EDTA. All rights reserved. For Permissions, please journals.permissions@oxfordjournals.org

2 Analgesics and progression of CKD 1909 with diagnosed advanced CKD stage 4 5 without an adverse effect on the progression rate of the disease. Keywords: acetaminophen; analgesics; disease progression; epidemiology; population-based with elevated S-Cr, and a network of clinical nephrologists throughout the country confirmed that the cause was CKD [11]. A confirmatory S-Cr test was done about 3 months after the first, if there was any doubt whether a patient had CKD or not. Patients with pre- or post-renal causes of S-Cr elevation, or a renal transplant, were not included. Introduction A link between analgesic abuse and chronic kidney disease (CKD) has long been postulated [1]. In particular, the association between phenacetin over-consumption and renal papillary necrosis combined with chronic interstitial nephritis (analgesic nephropathy) is well established [2,3]. By the end of the 1970s, phenacetin was banned from the markets of most industrialized countries. However, concern has been raised regarding the possible association between other commonly used analgesic drugs and CKD, particularly acetaminophen, a metabolite of phenacetin. Acetaminophen has been shown to affect renal function in laboratory animals [4] and is presently the most frequently used analgesic drug in many countries. In Sweden, annual sales are 41.4 Defined Daily Doses (DDD) per 1000 inhabitants [5]. Several epidemiological studies have investigated the association between the use of analgesic drugs, including acetaminophen and aspirin, and the development of CKD, with mixed and inconclusive results. A number of case control studies have shown positive associations between CKD and either acetaminophen [6 8], aspirin [9,10] or both [11], while others found no such association [12]. A recent follow-up study of healthy men found no evidence of deteriorating renal function with any examined analgesic drug [13]. In contrast, a study of healthy women found a moderate decline in renal function among individuals with a high lifetime use of acetaminophen, but not aspirin [14]. Few studies [15 17] have been designed to investigate the association between analgesic consumption and decline in renal function among patients with established kidney disease. These studies were small, and in two of them [15,16] confounding by phenacetin could not be excluded. We initiated a population-based cohort study in Sweden to explore the relation between the use of single substance analgesic drugs and progression of CKD. Subjects and methods Study subjects The study was approved by the regional ethics committee and all patients gave informed consent. The population-based cohort consists of patients diagnosed with incident advanced CKD between 20 May 1996 and 31 May 1998, as part of a nationwide case control study of risk factors for CKD in Sweden [11]. Patients eligible for the cohort study were all native Swedes aged years, whose serum creatinine (S-Cr) was found to exceed 3.4 mg/dl (300 µmol/l) for men, or 2.8 mg/dl (250 µmol/l) for women, for the first time and permanently during the inclusion period. The S-Cr tests were a part of routine clinical work among doctors at hospitals and primary care units. The inclusion date was the date when the first elevated test was taken. The limit was chosen to select patients early enough for them not to be on renal replacement therapy (RRT), but late enough to ensure that CKD was the likely cause of the S-Cr elevation. Virtually all clinical laboratories (n = 68) in Sweden identified patients Exposure assessment At inclusion, all patients underwent computer-aided face-to-face interviews by professional interviewers who were trained to interview in a standardized manner. The patients were shown pictures of current and historic packages of 78 major brands of non-narcotic analgesics to help them recall earlier use. They were asked to give a complete history of their lifetime use of analgesics and information on current use. Information on use of acetaminophen and aspirin during the follow-up period was abstracted from the medical records at the end of follow-up. Covariates The patients answered detailed questions about other exposures of interest (smoking, alcohol use, work-related exposures and level of education). They were further asked about current and previous use of other drugs, such as angiotensin converting enzyme (ACE) inhibitors. For body mass index (BMI) we used self-reported weight and height at inclusion. We collected additional information on co-morbidity status from the patients medical records at inclusion, using a standardized form. Abstracted information included congestive heart failure, coronary heart disease (angina or coronary arterial by-pass surgery), previous myocardial infarction, peripheral arteriosclerosis (claudication, by-pass surgery or amputation), previous cerebrovascular disease (including transitory ischemic disease), haemiplegia, dementia, diabetes (type 1 or type 2) with or without multiple complications, pulmonary disease, connective tissue disease (rheumatoid arthritis, vasculitis, systemic lupus erythematosus), mild and moderately severe liver disease, ulcer disease, metastatic cancer, solid tumour disease, leukaemia and acquired immune deficiency syndrome (AIDS). Outcome We used up to six routine S-Cr measurements taken at the discretion of the managing clinician to follow the progression of renal disease. This information, along with other blood chemistry data and pertinent renal variables, was collected from the medical records. Estimated glomerular filtration rate (egfr) was calculated from S-Cr using the abbreviated Modification of Diet in Renal Disease formula (egfr = 186 [S- Cr] [Age] [0.742 if patient is female] [1.212 if patient is black]) [18,19]. Information about the start date of RRT was obtained from the Swedish Registry of Renal Replacement Therapy (SRAU) [20], an essentially complete nationwide register based on the unique personal identification number assigned to all Swedish residents at birth or immigration. Similarly, death dates were ascertained from the Swedish National Death Registry. Follow-up The patients were followed from first inclusion (the date of the first S-Cr over the predetermined level) in until start of RRT (dialysis or kidney transplantation), death or end of follow-up (1 June 2003), whichever came first. In the analysis of renal survival (the combined endpoint of time to RRT or death before RRT), we extended the follow-up period to 31 December Statistical analysis Patients lifetime cumulative doses of acetaminophen and aspirin until the date of inclusion were categorized as never used, 99 gram (g), g and g and 3000 g. Patients were also classified into regular versus non-regular user (non-users and sporadic users) at inclusion based upon the self-reported information. Regular use was defined as use at least twice a week for 2 months prior to inclusion into the cohort. The patients prescribed low-dose aspirin, either self-reported or indicated in the medical record, were considered regular users. A patient whose lifetime cumulative dose was fewer than 20 tablets was categorized as never used [11]. During the follow-up period, a patient was classified as a user, yes/no, if he or she was prescribed the analgesic drug for at least 3 months. We assigned each co-morbid condition a weighted value to form the Charlson Co-morbidity Index (CCI) [21], a frequently used and validated index in

3 1910 M. Evans et al. follow-up studies [22], including studies of patients with end-stage renal disease [23]. The CCI was categorized into predefined levels where the lowest category had a value of 2, which represents no diseases other than renal disease. Age at inclusion was categorized into predefined groups at inclusion (<45 years, 45 64, 65). The self-reported level of education was categorized into predefined groups as shown in Table 1. These limits correspond to the different levels in the Swedish school system (elementary school, high school and graduate school). For BMI we used two categories: <25 kg/m 2 and 25 kg/m 2 at inclusion. Alcohol use at inclusion was categorized into non-users, and two groups above and below the median value (g/week) among users. Smoking categorization was defined similarly based on total pack-years at the time of inclusion. The primary renal disease diagnosis (categorized into diabetic nephropathy, nephrosclerosis, glomerulonephritis and other diseases) were assigned by the clinical nephrologists at inclusion, based on routine clinical evaluation. For categorization of mean arterial pressure at inclusion we used approximate quartiles. Proteinuria was categorized as high or low, according to the first registered value at inclusion. High was defined as a value >1500 mg total proteinuria/24 h or >1000 mg total albuminuria/24 h or a dipstick quantitative value exceeding one. Summary statistics (mean, median, 25th and 75th percentiles, standard deviation and interquantile range) described the distribution of egfr at the start of follow-up. Results were stratified by the main exposures (acetaminophen and aspirin). Proportions (categorical variables) and means (continuous variables) were produced and stratified by both regular use and lifetime dose of acetaminophen and aspirin. Parametric (t-test and chi-square) and non-parametric tests (Kruskal-Wallis) assessed statistical differences between groups. We aimed to have the same number of measurements for all patients, although time of follow-up varied widely. To achieve that, we divided the follow-up time for each participant (time from inclusion date to either death, start of RRT or 1 June 2003) into five equal intervals, thereby generating six dates for every patient. For each date, we searched in the medical record and noted the S-Cr closest to that date. Thus, we recorded up to six different S-Cr values for each patient with varying inter-measurement time periods as the result of the differences in follow-up time. We first described the individual-specific trend by fitting a linear regression model for each individual estimating the beta coefficient (slope), which represents the rate of change in egfr per year. Thus, the relationship between egfr and time was assumed to be linear, i.e. the decline of egfr was assumed to occur at a constant rate. The linear assumption is the most common way to describe renal deterioration in the scientific literature [24,25]. The estimated beta coefficients distribution was summarized using both mean and median, stratified by treatment and all other variables of interest. Patients with followup shorter than 1 year had their progression rate multiplied and presented by year to be comparable with the others in the univariate analysis (Table 3). In the further analyses, both the correlation among repeated measurements and the unbalanced data were handled by using a linear mixed effects model, with both fixed and random intercept and slope. For each exposure of interest (regular use versus no regular use at inclusion and lifetime cumulative dose), the fitted model included the exposure and the linear time effects, sex and age at baseline. The study hypothesis was tested by adding appropriate interaction terms between the exposure and time of follow-up. The multivariate model included as potential confounders: alcohol use, smoking, education, cardiovascular disease, CCI score, ACE inhibitor or angiotensin II receptor blocker (ARB) use at inclusion, primary renal disease, BMI at inclusion, mean arterial pressure at inclusion, proteinuria and use of prophylactic low-dose aspirin at inclusion. However, only statistically significant or biologically relevant variables were left in the final model. We had a large number of missing data (missing = 206) on proteinuria, but the coefficient did not change significantly after introduction of that variable. The goodness-of-fit of the final model was assessed through visual inspection of the observed and fitted trajectories and examination of the estimated standardized residuals. All the fitted models reproduced the observed trends, and the residual plots did not reveal any substantial deviation from the underlying assumptions. In the analysis of renal survival, we used a Cox proportional hazard model, including the same confounding variables as in the mixed effects model. In addition, we included co-morbidity and primary renal disease. Patients were censored at the date of RRT onset or the date of death, whichever occurred first. Data from all 920 patients were analysed in the survival analysis. All analyses were performed using STATA 9.0. Results Out of 1189 eligible patients, 69 died before contact could be established, 83 could not be interviewed due to severe illness and 117 refused to participate. Of the remaining 920 patients eligible for the follow-up, we excluded 62 with a time of follow-up shorter than 14 days and 57 for whom we were unable to find the medical record or a second S- Cr measurement, leaving 801 (68%) patients for inclusion in our analysis. The excluded patients (n = 119) did not differ significantly with respect to age, sex, primary renal disease or acetaminophen and aspirin use compared with the patients included in the analyses. The mean follow-up was 768 days with a range of days (7.0 years). The 801 patients generated 4291 S- Cr observations. Most of the patients (68%) had a complete record with six measurements whereas 5% had only two. The inter-measurement variation was substantial and the data were unbalanced due to the differences in follow-up time. Table 1 shows the baseline characteristics of the cohort according to acetaminophen use. Compared with nonregular users of acetaminophen, regular users at inclusion had a significantly higher CCI score, lower alcohol intake, lower level of education, more cardiovascular disease and diabetes, and a larger proportion were women and regular users of aspirin. Table 2 displays the same characteristics, according to aspirin use. Regular users of aspirin at inclusion were characterized by higher CCI, lower level of education, more cardiovascular disease, diabetes and nephrosclerosis and higher age and level of smoking compared with patients who did not use aspirin regularly. The mean number of S-Cr measurements was slightly lower and the mean time of follow-up was slightly shorter among regular users of acetaminophen and aspirin compared with non-regular users (Tables 1 and 2). A scatter plot of all 4291 egfr data points against time of follow-up among the 801 patients was created. To visualize some examples of individual trajectories, we interconnected the data points for one randomly selected patient from each decile of egfr at baseline (Figure 1). At inclusion, the mean egfr was 16.5 ml/min/1.73 m 2 (range , standard deviation 3.5). The mean egfr declined over time by 9.0 ml/min/1.73 m 2 per year according to the linear regression coefficient while the median decline was 5.1 ml/min/1.73 m 2 per year. Patients with faster progression rates reached the study endpoint (RRT or death) faster and consequently had shorter time of follow-up. The distribution of the patients rates of change (ml/min/1.73 m 2 per year) is shown in Figure 2. In Table 3 the median rates of egfr decline (progression rates) calculated by simple linear regression are presented according to various subgroups of patients. The rate of decline was inversely related to patients age, with the youngest patients having the most rapid decline, and the rate was faster in men than in women. The rate also varied with underlying renal disease; the fastest progression rate was

4 Table 1. Characteristics of 801 patients with chronic kidney disease in a population-based cohort in Sweden according to acetaminophen use Lifetime cumulative dose acetaminophen c Test Acetaminophen use at inclusion d Test Never used 99 gram (g) g g 3000 g P-value b No regular use Regular use P-value b Characteristics (n = 287) (n = 258) (n = 135) (n = 62) (n = 52) (n = 669) (n = 127) Glomerular filtration rate at inclusion (mean) a Time of follow-up in days (mean) Number of measurements (mean) Age, years (mean) < Sex (%) <0.01 <0.01 Male/Female 75.6/ / / / / / /48.8 Level of education (%) < years years years Body mass index at inclusion (%) <25 kg/m 2 / 25 kg/m / / / / / / /44.1 Smoking (%) Ever non-smokers pack-years >20 pack-years Alcohol use at inclusion (%) 0.25 <0.01 No drinkers g/week >32.6 g/week Mean arterial pressure at baseline (%) (n = 745) <100 mmhg mmhg mmhg >118 mmhg Co-morbidity index at inclusion (%) <0.01 <0.01 Charlson Index Charlson Index Charlson Index > Primary renal disease (%) Diabetes Glomerulonephritis Other Nephrosclerosis Cardiovascular disease at inclusion (%) No/Yes 58.2/ / / / / / /51.2 Low-dose aspirin use at inclusion (%) No/yes 71.8/ / / / / / /30.7 ACE-inhibitor use at inclusion (%) No/yes 71.4/ / / / / / /28.4 Proteinuria at baseline (%) (n = 595) Low/high 31.8/ / / / / / /66.3 a Expressed in ml/min estimated from MDRD abbreviated formula. To convert ml/min to ml/s, multiply by b P-values for comparisons between subjects of different lifetime cumulative dose (non-users as reference group), and regular versus non-regular users (non-regular users as reference group) were calculated with chi-square statistics. P-values for comparisons between means were calculated using the Kruskal-Wallis test (non-parametric). c Missing = 7. d Missing = 5. Analgesics and progression of CKD 1911

5 Table 2. Characteristics of 801 patients with chronic kidney disease in a population-based cohort in Sweden according to aspirin use Lifetime cumulative dose aspirin c Test Aspirin use at inclusion Test Never users 99 gram (g) g g 3000 g P-value b No regular use Regular use P-value b Characteristic (n = 253) (n = 245) (n = 187) (n = 68) (n = 41) (n = 538) (n = 263) Glomerular filtration rate at inclusion (mean) a Time of follow-up in days (mean) Number of measurements (mean) Age, years (mean) <0.01 Sex (%) Male 64.8/ / / / / / /36.1 Level of education (%) 0.24 < years years years Body mass index at inclusion (%) <25 kg/m 2 / 25 kg/m / / / / / / /50.6 Smoking (%) 0.13 <0.01 Non-smokers pack-years >20 pack-years Alcohol use at inclusion (%) No drinkers g/week >32.6 g/week Mean arterial pressure at baseline (%) <110 mmhg mmhg mmhg >118 mmhg Co-morbidity index at inclusion (%) 0.03 <0.01 Charlson Index Charlson Index Charlson Index > Primary renal disease (%) 0.78 <0.01 Diabetes Glomerulonephritis Other Nephrosclerosis Cardiovascular disease at inclusion (%) <0.01 <0.01 No/yes 65.2/ / / / / / /71.1 Low-dose aspirin use at inclusion (%) <0.01 <0.01 No/yes 99.6/ / / / / /0 14.1/85.9 ACE-inhibitor use at inclusion (%) No/yes 71.9/ / / / / / /30.4 Proteinuria at baseline (%) Low/high 30.0/ / / / / / /68.9 a Expressed in ml/min estimated from MDRD abbreviated formula. To convert ml/min to ml/s, multiply by b P-values for comparisons between subjects of different lifetime cumulative dose (non-users as reference group) and regular versus non-regular users (non-regular users as reference group) were calculated with chi-square statistics. P-values for comparisons between means were calculated using the Kruskal-Wallis test (non-parametric). c Missing = M. Evans et al.

6 Analgesics and progression of CKD 1913 Fig. 1. Scatterplot of 4291 estimated GFR during 7 years of follow-up in a population-based cohort of patients with advanced chronic kidney disease. Dots are interconnected for 10 randomly selected individuals. observed among patients with diabetic nephropathy, followed by patients classified as having glomerulonephritis. Not surprisingly, individuals with high blood pressure at inclusion showed faster progression than those with low, and patients with severe proteinuria progressed faster than those with milder proteinuria. However, there was no significant association between progression rate and CCI, presence or absence of cardiovascular disease, ACE-inhibitor or ARB use, BMI or lifestyle factors such as education, smoking or alcohol use. Association between acetaminophen use and decline of egfr The unadjusted progression rate based on simple linear regression among regular users of acetaminophen at the time of inclusion in the cohort was 5.2 ml/min/1.73 m 2 per year. Non-regular users of acetaminophen at inclusion progressed at 5.1 ml/min/1.73 m 2 per year (Table 3). There was no significant difference in the progression rate for patients with ever versus never use of acetaminophen. In Table 4, we present the results of the mixed effects model. The coefficient presented is the difference in progression rate (ml/min/1.73 m 2 per year) associated with analgesic use. A positive coefficient means a slower progression rate whereas a negative value indicates faster progression as compared to the reference. In these multivariate analyses (adjusting for age, sex, ACE or ARB use and mean arterial pressure), neither regular use of acetaminophen at time of inclusion nor lifetime consumption showed any significant association with egfr decline (Table 4). The progression rate for patients with the highest lifetime cumulative dose was 0.24 ml/min per year slower than for patients who had never used acetaminophen. Patients with regular acetaminophen use also showed a positive increment; the progression rate for regular users was 0.93 ml/min per year slower compared with non-regular users. Restricting the analysis to exclusive users of acetaminophen (by excluding 50 patients who were regular users of both aspirin and acetaminophen at inclusion) changed the coefficient to 0.53 ml/min per year [95% Confidence Interval (CI) 0.6, 1.7]. Among exclusive users of acetaminophen there Fig. 2. Distribution of rates of change in glomerular filtration rate (egfr, ml/min/1.73 m 2 per year), estimated by linear regression for each subject, among 801 Swedish patients with incident advanced chronic kidney disease. was a slightly faster progression rate among patients in the two highest groups [for patients with g, the regression coefficient was 0.74 ml/min (95% CI 2.2, 0.7), and for patients with 3000 g, the coefficient was 0.75 ml/min (95% CI 2.5, 1.0)], although it did not reach statistical significance. In other subgroup analyses of progression rate stratified by sex, cardiovascular disease and regular aspirin use, we observed no interaction between these variables and regular use or cumulative dose of acetaminophen (data not shown). The progression rate associated with acetaminophen regular use was somewhat slower for patients with nephrosclerosis compared to other primary renal diseases (data not shown). No group was associated with a significantly faster progression rate. Stratification on egfr at inclusion showed that patients with more preserved renal function at inclusion progressed faster compared to patients included late in their course of disease. However, this relationship did not influence the effect of acetaminophen use significantly. Patients who were regular users of acetaminophen at inclusion and continued their usage during the follow-up (n = 72) had significantly slower decline of egfr than that in patients who were not regular users at inclusion (Table 4, last column). Furthermore, there was no significant effect on decline of egfr for patients with different levels of lifetime acetaminophen use who used acetaminophen during the follow-up (n = 149). We repeated the analyses after excluding all patients with follow-up shorter than 3 months and less than three measurements of egfr (n = 113), without any significant change in the coefficients or interpretation of the data (data not shown). We also reanalysed the data excluding patients with egfr at inclusion <7 ml/min without any significant change in the results. Association between aspirin use and decline of egfr For regular aspirin users at inclusion, the unadjusted progression rate based on simple linear regression was slower than that among patients who were not regular users (Table 3). In the mixed effects multivariate model, regular users of aspirin progressed 0.8 ml/min per year slower than patients without such use (Table 4). The progression rate in the highest lifetime accumulated dose was 0.7 ml/min per

7 1914 M. Evans et al. Table 3. Median decline in glomerular filtration rate in a cohort of 801 patients with advanced chronic kidney disease Median progression rate in Groups (number) ml/min/1.73 m 2 per year a Significance test b Median rate of decline N = Time of follow-up <6 months (113) 23.3 < months 3 years (469) 6.6 >3 years (219) 1.8 Glomerular filtration rate at baseline ml/min (201) 4.1 < ml/min (200) ml/min (200) ml/min (200) 6.6 Age <0.01 <45 (155) (328) (318) 3.8 Sex <0.01 Male (518) 6.0 Female (283) 4.0 Primary renal disease <0.01 Diabetes (253) 6.5 Glomerulonephritis (190) 5.7 Nephrosclerosis (119) 4.6 Other (239) 4.7 Acetaminophen Ever used (507) Never used (287) 5.1 Regular use at inclusion (669) No regular use at inclusion (127) 5.1 Aspirin Ever used (541) Never used (253) 5.7 Regular use at inclusion (263) No regular use at inclusion (538) 5.3 Co-morbidity 0.21 Charlson index <3 (250) 5.3 Charlson index 3, 4 (296) 4.9 Charlson index >4 (250) 5.7 Education years (463) years (184) years (143) 5.5 Smoking 0.53 Never smokers (312) pack-years (259) 5.1 > 20 pack-years (218) 5.5 Alcohol 0.55 No use (199) g (292) 4.8 >32.6 g (301) 5.5 Body mass Index 0.68 <25 kg/m 2 (398) kg/m 2 (403) 5.1 Mean arterial pressure <0.01 <100 mmhg (154) mmhg (222) mmhg (177) 5.7 >118 mmhg (188) 6.4 ACE-inhibitor or ARB use Regular use at inclusion (258) No regular use at inclusion (543) 4.9 Proteinuria <0.01 Low (191) 3.0 High (404) 6.0 a Progression rate was calculated by linear regression from up to six repeated measurements of estimated glomerular filtration rate per individual. b P-values for differences among subgroups were calculated using quantile regression.

8 Analgesics and progression of CKD 1915 Table 4. Differences in estimated progression rates, (change in egfr in ml/min/1.73 m 2 per year) associated with analgesic use, in a cohort of patients with advanced chronic kidney disease Crude coefficient a Adjusted coefficient b Adjusted coefficient b for those with Acetaminophen (95% CI) (95% CI) use also during follow-up c (95% CI) (n) Lifetime cumulative dose Never used Ref. Ref. Ref. 99 g 0.17 ( 0.9, 0.6) 0.17 ( 0.9, 0.6) 0.81 ( 0.5, 2.2) (n = 48) g 0.71 ( 0.2, 1.6) 0.60 ( 0.3, 1.5) 2.34 (0.8, 3.9) (n = 38) g 0.26 ( 1 0,1.5) 0.65 ( 0.7, 2.0) 1.31 ( 0.3, 3.0) (n = 31) 3000 g 0.05 ( 1.3, 1.4) 0.24 ( 1.2, 1.7) 1.13 ( 0.6, 2.8) (n = 32) P-value trend 0.39 P-value trend 0.24 Use at inclusion into the cohort No regular use Ref. Ref. Regular use 0.57 ( 0.3, 1.4) 0.93 (0.03, 1.8) 1.7 (0.6, 2.8) (n = 72) Aspirin Lifetime cumulative dose Never used Ref. Ref. Ref. 99 g 0.56 ( 0.2, 1.3) 0.53 ( 0.3, 1.3) 1.12 ( 0.1, 2.3)(n = 70) g 0.63 ( 0.2, 1.5) 0.68 ( 0.2, 1.6) 0.92 ( 0.2, 2.1) (n = 82) g 0.64 ( 0.6, 1.8) 0.69 ( 0.6, 1.9) 1.22 ( 0.6, 3.1) (n = 25) 3000 g 0.71 ( 0.8, 2.2) 0.71 ( 0.8, 2.3) 1.95 ( 0.3, 4.2) (n = 14) P-value trend 0.14 P-value trend 0.12 P-value trend 0.01 Use at inclusion into the cohort No regular use Ref. Ref. Ref. Regular use 0.72 (0.1, 1.4) 0.80 (0.1, 1.5) 1.1 (0.4, 1.9) (n = 187) Aspirin regular use No low dose aspirin 0.1 ( 1.6, 1.5) 0.14 ( 1.8, 1.5) Low dose aspirin 0.84 (0.1, 1.5) 0.94 (0.2, 1.7) CI = Confidence Interval, g = gram. a In ml/min/1.73 m 2 per year, unadjusted. Negative values mean that the progression rate is faster, while positive values mean that the progression rate is slower. The values are expressed as the absolute difference in progression rate between the reference group and the study group per year (regression coefficient). If a group s coefficient is 1.0 that group progress 1.0 ml/min/year faster compared to the reference group. b In ml/min/1.73 m 2 per year, adjusted by sex, age, angiotensin converting enzyme inhibitor or angiotensin II receptor blocker use, and mean arterial pressure, at baseline. c Restricted analysis. year slower than for patients who had never used aspirin. Stratification by the presence or absence of cardiovascular disease at inclusion suggested that the apparent protective effect of regular aspirin use was more pronounced among patients with no cardiovascular disease [regression coefficient 0.7 ml/min per year (95% CI 0.4, 1.9) among patients with regular use of aspirin]. Among patients with cardiovascular disease, the regression coefficient among regular users of aspirin was 0.2 ml/min per year (95% CI 0.8, 1.2), although the progression rate was generally slower. When we stratified by sex, the protective effect of aspirin was present in both, but more pronounced among men [progression rate for regular users of aspirin was 1.1 ml/min slower per year (95% CI 0.2, 2.0) than for non-regular users] compared with women [progression rate for regular users of aspirin was 0.4 ml/min slower per year (95% CI 0.6, 1.4) than for non-regular users]. The protective effect of aspirin was present among all primary renal diseases, but more pronounced among patients with glomerulonephritis [patients with regular aspirin use progressed 2.0 ml/min per year slower compared to nonregular users (95% CI 0.5, 3.4)]. Restricting the analysis to exclusive users of aspirin (by excluding 50 patients who were both regular users of acetaminophen and aspirin) did not materially change these patterns. The differences in progression rate for different lifetime cumulative doses were not significant, although indicating slightly slower progression rates with higher lifetime cumulative doses of aspirin. There was a non-significantly slower progression rate for patients who had been never users of aspirin and used acetaminophen regularly (data not shown). Patients who continued to use aspirin regularly during the follow-up (n = 187) showed a slower progression rate compared to non-regular users. Additional sub-analyses did not reveal any substantial deviations from the presented results after exclusion of patients with less than 3 months of followup or less than three measurements or egfr at inclusion <7 ml/min. Association between analgesic use and time to RRT In the age- and sex-adjusted Cox proportional hazards model regular acetaminophen use was associated with a slightly elevated risk of RRT [hazard ratio (HR) 1.2; 95%CI 1.0, 1.4]. In the multivariate analysis male sex, diabetes nephropathy, high level of proteinuria and high mean arterial pressure at baseline were significantly associated with a shorter time to RRT. The HR for regular acetaminophen use in the multivariate model was 1.1 (95% CI 0.9, 1.4). The HR associated with different levels of lifetime acetaminophen use was very close to unity (data not shown). The age- and sex-adjusted HR for regular aspirin use was 1.0 (95% CI 0.9, 1.2). Regular aspirin use was associated with lower, although not statistically significant, risk for RRT (HR = 0.9; 95% CI 0.7, 1.1) in the multivariate model, while lifetime cumulative dose of aspirin was not associated with time to RRT. Discussion In this population-based follow-up study of unselected patients with CKD stage 4 5 (NKF K/DOQI guidelines) [26], we saw a 0.93 ml/min per year slower progression rate in patients who were regular users of acetaminophen

9 1916 M. Evans et al. compared with non-regular users at baseline, and a slower progression rate of borderline statistical significance among regular users of aspirin relative to patients who did not use aspirin regularly at baseline. We did not observe any association between total lifetime cumulative dose of acetaminophen or aspirin and progression of CKD. Many other known risk factors for progression in CKD, such as high level of proteinuria, high mean arterial pressure and male sex were confirmed. The strengths of our study include the prospectively collected exposure information on analgesic use at inclusion, thorough registration of lifetime cumulative analgesic use at inclusion, rich covariate information and the population-based nature of the investigation, designed to include all CKD-patients nationwide, newly diagnosed with S-Cr above a predetermined level. The use of up to six repeated measurements of S-Cr made the estimate of the progression rate fairly insensitive to random day-to-day variation, and the mixed effects model is effective in taking care of unbalanced datasets such as ours [27]. A weakness is that S-Cr was not measured in a standardized manner. The blood samples were drawn in routine clinical work and analysed at different laboratories. However, these differences, which contributed more to inter- than intra-individual variation, should not be of major concern because we made intraindividual comparisons of the S-Cr values to calculate the egfr decline, and most patients had their sequential tests analysed at the same laboratory. We chose to estimate the GFR by using the abbreviated MDRD equation, which offers estimates close to measured GFR among patients with advanced CKD [28]. The accuracy of measuring progression rate by egfr has been questioned. Recent analyses, however, show that a progression rate derived from serial egfr estimations is comparable with a progression rate based upon repeated I iothalamate clearances [29]. Sequential measurements of GFR with calculation of individual linear slopes are considered the best way to study decline in renal function among patients with advanced CKD [30]. Progression rates have been measured in other cohorts of patients with CKD as severe as in our study [31,32]. The average progression rate was similar in our cohort compared with most other cohorts of patients with CKD [24] in spite of the unique composition of our study cohort, representing unselected, incident cases in a defined geographical area during a defined period of time. Several of the patients in our cohort had low egfr at inclusion. This is a consequence of the case-selection procedure. Most patients were included close to the threshold level (S-Cr ), whereas some patients were first identified with severe kidney failure. Although the true onset of CKD is generally unknown because the start is insidious, we operationally defined incidence as first documentation of a permanent elevation of S-Cr above a predefined level. Thirteen percent of the patients were unaware of their renal disease at inclusion. Most other cohorts of CKD patients have included a mix of incident and prevalent cases. Because of the high mortality rate in this cohort (42% of the patients had died already by the end of 2002) [33], we were unable to contact patients again to ask them about further use of analgesics. Instead we chose to use information from their medical records. If patients bought analgesic drugs over-the-counter (OTC) we would have missed that during the follow-up, but we have reasons to believe that OTC use was limited. In Sweden, the maximum cost for prescribed drugs is about 180 per year. Costs above that limit are covered by the National Social Welfare system. Patients with advanced CKD exceed this limit easily, and in order to keep costs low they ask for a prescription for all drugs, including analgesics. In fact, only about 35% of the DDD of acetaminophen in Sweden are sold OTC [5]. In Sweden there is limited use of mixed types of analgesics of any kind. Today, there are no mixtures available OTC in combination with acetaminophen, and since the study started no combinations of two analgesics and caffeine have been on the market. Several case control studies [6 12,34] and a few cohort studies [13,14,35 37] have investigated the relation between analgesic use and renal function. Most of them were traditional aetiologic studies with focus on exposure before disease onset and with incidence of end-stage renal disease/rrt as the outcome [6 12]. Among these conventional aetiologic studies investigating acetaminophen and CKD risk, a majority reported a positive association [6,8,11,36], but two studies did not [10,12], and one study was positive only if acetaminophen was taken in combination with other non-phenacetin analgesics [7]. Two US cohort studies one of male physicians [13] and one of female nurses [14] were able to follow changes in renal function among apparently healthy individuals through two measurements of S-Cr years apart. The female nurses study revealed a dose-dependent 2-fold gradient in risk for a 30% fall in GFR with increasing lifetime cumulative dose of acetaminophen, while the study of male physicians showed no relationship at all [13]. As stratified analyses in our present study did not reveal any gender differences (data not shown), our findings are consistent with the results of the latter cohort study but not the former. To our knowledge, there are no large-scale prospective studies, similar to our study, that have investigated the possible impact of acetaminophen intake on the rate of progression in patients already diagnosed with CKD, although the followup of small groups of patients with analgesic nephropathy [16] or analgesic-associated nephropathy [15,17] has consistently demonstrated a greater loss of renal function in those with continued use of acetaminophen compared with those who stopped using acetaminophen. The majority of the studies that could meaningfully address the association between aspirin use and risk of CKD (not only analgesic nephropathy) were negative [6,8,12,36], but a few studies, including the population-based case control analysis that gave rise to our present cohort [11], found exclusive use of aspirin to be an independent risk factor [9 11], and yet another study [7] reported a positive association for aspirin in combination with other nonphenacetin analgesics like acetaminophen, but not for aspirin alone. Of the two US cohort studies that examined GFR at 11 and 14 years in apparently healthy men and women, respectively, neither found any evidence of detrimental effects of aspirin use. In fact, in the study of female nurses [14] the overall odds ratios for a >30% decrease in GFR were below unity (although not statistically

10 Analgesics and progression of CKD 1917 significant) among aspirin users. Interestingly, the study of healthy male physicians [13] demonstrated a reduced risk for decline in renal function among participants who took aspirin and had no cardiovascular risk factors, in agreement with our own findings among stage 4 5 CKD patients. In the original population-based case control study [11], which ascertained the cases included in the present cohort, regular use of acetaminophen and aspirin both was associated with a 2.5-fold increase in the odds of developing CKD compared with non-use. However, in this study, when the cases with CKD stage 4 5 were followed forward in time and comparisons were made within this case group, there was no relation between the level of analgesic exposure at inclusion and rate of CKD progression. The reasons for the apparent inconsistency are not obvious. One interpretation could be that these drugs may act more as initiators in the early stages of the disease, and that continued use after the damage has already occurred will have no more than trivial effects. Then again, the conflicting results in the traditional aetiologic studies suggest that bias/confounding may have played a role. Prior investigators of pre-disease analgesic drug use and risk of CKD development have all been faced with the question of protopathic bias [38], which means that individuals with developing CKD, in contrast to healthy individuals, may have been more likely to use analgesic drugs as a consequence of early symptoms from their underlying renal disease. Although our previous case control analysis only included incident cases most with advanced CKD and lagged analyses provided no indication of protopathic bias, we cannot rule out the possibility that symptoms from early stages of the renal disease, or other diseases such as diabetes associated with CKD, may have contributed to analgesic use. This concern is even greater in some of the other case control studies that have end-stage renal disease, instead of CKD, as the outcome. Unless a very long latency is worked into the design of these studies, some degree of protopathic bias is inevitable. In addition, recall bias, a common bias in case control studies, may further distort the results. Another related concern is that phenacetin use in the distant past may have inflated the observed associations. In our previous case control study, this concern was somewhat allayed by the fact that Sweden was one of the first countries to ban phenacetin. Another interpretation is that our present results may be biased. If patients with more analgesic use were among those who could not be interviewed, due to incapacitating disease or early death, there would be an underrepresentation of heavy or regular analgesic users in the present cohort. However, among those who were interviewed but not included (n = 119, due to a follow-up <14 days or lack of a second S-Cr measurement), we could not detect any differences in exposures compared with the patients included into the analyses. Protopathic bias is unlikely to explain our present findings; in order to do that, patients with faster progression would have had to consume less analgesic drugs. Reverse causation after CKD diagnosis is another plausible explanation of our results. Doctors might have been aware of the possible association between analgesic use and risk of renal failure and reduced analgesic prescriptions to patients with severe disease and faster progression rate, compared with those with less severe disease. The exposure information will then be exaggerated among patients with faster progression rate. This would lead to underestimation of the dose progression relation, but is unlikely to invert it. Negative confounding by comorbidity seems less likely, particularly since we included a co-morbidity index in our model. This prospective population-based cohort study confirms many of the known risk factors for the progression of advanced CKD. The strengths of this prospective cohort approach, with valid information about the egfr level at entry, repeated subsequent measurements over the entire follow-up time, thus reducing misclassification of the outcome, rich covariate information and little evidence of exposure information bias, lead us to suggest despite some inconsistencies in relation to other studies that single substance acetaminophen and aspirin may be safely used by patients diagnosed with CKD stage 4 5 without an adverse effect on the progression rate of the disease. Acknowledgements. We thank the Swedish Renal Association and the Swedish Registry for Renal Replacement Therapy for their support. We are indebted to the Swedish nephrology clinics and units throughout Sweden for willingly supplying follow-up information. Conflict of interest statement. The study was supported by the International Epidemiology Institute (J.K.M. and J.P.F.), an independent biomedical research organization, which in turn received funding from McNeil Consumer products. McNeil was not involved in any aspect of the study design, data collection, interpretation, writing or reviewing of this paper. References 1. Spühler O, Zollinger HO. Die chronisch-interstitielle Nephritis. Z Klin Med 1953; 151: Grimlund K. Phenacetin and renal damage at a Swedish factory. Acta Med Scand 1963; 174: Dubach U, Rosner B, Pfister E. Epidemiologic study of abuse of analgesics containing phenacetin. Renal morbidity and mortality ( ). N Engl J Med 1983; 308: Nanra RS, Kincaid Smith P. Experimental evidence for nephrotoxicity of analgesics. In: Stewart JH (ed.). Analgesic and NSAID-Induced Kidney Disease. Oxford: Oxford University Press, 1993: pp Smärtstillande läkemedel 2006 Statistikenheten, 1 5 February 2007, 6. Sandler D, Smith J, Weinberg C et al. Analgesic use and chronic renal disease. N Engl J Med 1989; 320: Pommer W, Bronder E, Greiser E et al. Regular analgesic intake and the risk of end-stage renal failure. Am J Nephrol 1989; 9: Perneger TV, Whelton PK, Klag MJ. Risk of kidney failure associated with the use of acetaminophen, aspirin, and non-steroidal antiinflammatory drugs. N Engl J Med 1994; 331: Morlans M, Laporte JR, Vidal X et al. End-stage renal disease and non-narcotic analgesics: a case-control study. Br J Clin Pharmacol 1990; 30: Ibáñez L, Morlans M, Vidal X et al. Case-control study of regular analgesic and non-steroidal anti-inflammatory use and end-stage renal disease. Kidney Int 2005; 67: Fored CM, Ejerblad E, Lindblad P et al. Acetaminophen, aspirin, and chronic renal failure. N Engl J Med 2001; 345: Murray TG, Stolley PD, Anthony JC et al. Epidemiologic study of regular analgesic use and end-stage renal disease. Arch Intern Med 1983; 143: