Article Text

Extended report
Chronic NSAID use and long-term decline of renal function in a prospective rheumatoid arthritis cohort study
  1. Burkhard Möller1,
  2. Menno Pruijm2,
  3. Sabine Adler1,
  4. Almut Scherer3,
  5. Peter M Villiger1,
  6. Axel Finckh4,
  7. on behalf of the Swiss Clinical Quality Management in Rheumatic Diseases (SCQM) Foundation, CH-8048 Zurich, Switzerland
  1. 1Department of Rheumatology and Clinical Immunology, Inselspital, University Hospital Bern, Switzerland
  2. 2Department of Nephrology, University Hospital of Lausanne (CHUV), Lausanne, Switzerland
  3. 3SCQM Office, Zurich, Switzerland
  4. 4Division of Rheumatology and Division of Clinical Epidemiology, University Hospital Geneva (HCUGE), Geneva, Switzerland
  1. Correspondence to Dr Burkhard Möller, Department of Rheumatology and Clinical Immunology, Inselspital-University Hospital, Bern 3010, Switzerland; Burkhard.moeller{at}


Objectives Non-steroidal anti-inflammatory drugs (NSAIDs) may cause kidney damage. This study assessed the impact of prolonged NSAID exposure on renal function in a large rheumatoid arthritis (RA) patient cohort.

Methods Renal function was prospectively followed between 1996 and 2007 in 4101 RA patients with multilevel mixed models for longitudinal data over a mean period of 3.2 years. Among the 2739 ‘NSAID users’ were 1290 patients treated with cyclooxygenase type 2 selective NSAIDs, while 1362 subjects were ‘NSAID naive’. Primary outcome was the estimated glomerular filtration rate according to the Cockroft–Gault formula (eGFRCG), and secondary the Modification of Diet in Renal Disease and Chronic Kidney Disease Epidemiology Collaboration formula equations and serum creatinine concentrations. In sensitivity analyses, NSAID dosing effects were compared for patients with NSAID registration in ≤/>50%, ≤/>80% or ≤/>90% of assessments.

Findings In patients with baseline eGFRCG >30 mL/min, eGFRCG evolved without significant differences over time between ‘NSAID users’ (mean change in eGFRCG −0.87 mL/min/year, 95% CI −1.15 to −0.59) and ‘NSAID naive’ (−0.67 mL/min/year, 95% CI −1.26 to −0.09, p=0.63). In a multivariate Cox regression analysis adjusted for significant confounders age, sex, body mass index, arterial hypertension, heart disease and for other insignificant factors, NSAIDs were an independent predictor for accelerated renal function decline only in patients with advanced baseline renal impairment (eGFRCG <30 mL/min). Analyses with secondary outcomes and sensitivity analyses confirmed these results.

Conclusions NSAIDs had no negative impact on renal function estimates but in patients with advanced renal impairment.

  • NSAIDs
  • Rheumatoid Arthritis
  • Outcomes research

Statistics from


Exposure to non-steroidal anti-inflammatory drugs (NSAIDs) may lead to reversible acute kidney injury or end stage renal disease (ESRD) due to vasoconstriction, tubular necrosis and acute interstitial nephritis, regardless of an individual's history of renal disease.1 ,2 Although the cumulative lifetime dose of non-selective NSAIDs is reportedly associated with decline in renal function, little is known about the evolution of renal function under prolonged NSAID therapy.3 While glomerular filtration rates (GFRs) may decrease within days after the initiation of NSAID therapy,4 clinically apparent events are poorly predictable and may occur years later.5 ,6 The gastrointestinal toxicity of NSAIDs could largely be overcome by concomitant use of proton pump inhibitors and the development of type 2 cyclooxygenase-selective NSAIDs (coxibs),7 but non-selective NSAIDs and coxibs carry similar statistical risks of cardiovascular and renal complications as non-selective NSAIDs.8 Given the broad definition of NSAID contraindication in today’s practice guidelines, it is obvious that NSAIDs are frequently used in a way that breaches current safety recommendations.9–11

Rheumatoid arthritis (RA) patients commonly require more extensive NSAID use than others.12 They are at increased risk of impaired renal function and renal failure.3 ,13 RA patients should be regularly monitored for renal and other organ toxicities in relation to their disease-modifying antirheumatic (DMARD) therapy.14 In this study, we aimed to quantify the long-term decline in renal function in centrally registered prospective serial assessments of different GFR estimates in relation to concomitant NSAID prescription and other renal risk factors.

Materials and methods

We estimated the sample size necessary to detect the progression of kidney dysfunction with the use of NSAIDs using population-based figures of GFR decline associated with diabetes mellitus.15 Assuming a statistical power of 90%, a type I error probability of 0.05, a two-sided probability test and a ratio between the groups of approximately 2:1 NSAID users versus non-users, 1113 patients with sequential GFR assessments would be required.

Study design and setting

This longitudinal cohort study is nested within the Swiss clinical quality management (SCQM) database, which is described elsewhere.16 In total, 4101 patients with RA diagnosis in guidance of the 1987 classification criteria RA17 and at least two assessments of renal function were included. The mean follow-up time was 3.2 years. Registration in the SCQM database is restricted to patients treated by board-certified rheumatologists in order to track the effectiveness and safety of antirheumatic therapies. Approximately two-thirds of Swiss rheumatologists contributed patients to the SCQM database. About 60% of the study participants came from private practices, while the others were followed in clinical or university centres. Ethical approval for the cohort was obtained from the Swiss Academy of Medical Sciences review board, and all participants provided written informed consent to participate in the cohort. Based on the assumed general prevalence of RA, the participants represent approximately 9% of the Swiss RA population.18 ,19 In case of missing follow-up data 3 months after the proposed annual visit schedule, the treating rheumatologists were reminded and reasons for missing follow-up were collected by another phone call.

The cohort started in 1996 and is ongoing, but this analysis ended in 2007 with the stop in reporting of creatinine and other laboratory safety parameters. Patients were considered as lost to follow-up if no visit was available for more than 15 months before close of the study database in 12 October 2007.

Outcome measures

Traditionally, the Cockroft–Gault formula has been used throughout the world to estimate GFR in order to adjust the dose of prescribed drugs to renal function. As the debate about the best glomerular filtration estimate is not yet settled,20 ,21 we decided to use the classic eGFR estimate according to the Cockroft–Gault formula, eGFRCG,22 as our primary outcome. Secondary outcomes included the more recent Modification of Diet in Renal Disease (eGFRMDRD) formula23 and the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine equation (eGFRCKD),24 as well as absolute serum creatinine concentrations. EGFRCG was dichotomised to obtain a proportion of patients with a ‘clinically important decrease of clearance after one year’ defined by a ≥30% decrease in eGFRCG.25

Exposure variables

Participation in SCQM required to record both prescribed and requested over-the-counter NSAID use. NSAID exposure was operationally defined as ‘NSAID users’ and ‘NSAID naive’. ‘NSAID users’ had at least one NSAID registration during follow-up. In the absence of registration, it was assumed that no NSAID treatment had been received. The most frequently used NSAIDs are listed in supplementary appendix table S1. Acetaminophen and other non-opioid analgesics were handled as a separate drug exposure category.


We considered age, sex, renal and cardiovascular disease, hypertension and diabetes as confounders a priori and forced these variables in the model. The RA disease activity score,26 body mass index and antirheumatic RA therapies other than NSAIDs were included in the model only if they were found to be substantial confounders using the 10% change in estimate criteria.27 Comorbidities were registered at the discretion of the treating rheumatologists without further specification, while the treatment of these conditions was not recorded. In case of sporadically missing covariates (<5%), we used the population mean as a substitute. We defined renal function stages according to the eGFRCG, explored potential effect modification by baseline renal function stages 1–5 of CKD according to the definitions of the National Kidney Foundation Kidney Disease Outcomes Quality Initiative (KDOQI).28 Furthermore, eGFRCG>130 mL/min was used as the functional correlate of hyperfiltration.29

Statistical analysis

Baseline characteristics were analysed in ‘NSAID naive’ and ‘NSAID users’ and compared with Student t test for normally distributed continuous variables, with the Kruskal–Wallis test for non-normally distributed continuous variables or with Pearson's χ2 test for dichotomous variables. We tested the normality assumption of the primary outcome with the Shapiro–Wilk W test. The evolution of eGFRCG and secondary outcome variables was analysed using mixed models for longitudinal data. We chose the best-fitting and parsimonious covariance structure using series of likelihood ratio tests based on restricted maximum likelihood estimates. We examined whether time best fit the data as a linear trend or as a quadratic function. Finally, we selected the best-fitting mean response model for renal function outcome, controlling for other risk factors of renal deterioration. Crude and covariate-adjusted analyses were performed in STATA V.11 for Windows (StatCorp LP, Texas, USA).

Sensitivity analyses

We used alternative exposure definitions based on the frequency of NSAID reports (NSAIDs recorded in >50%, >80% or >90% of visits) and on current NSAID use versus current non-use. To identify effects of specific NSAIDs with established cardiovascular and renal effects,30 we also examined the impact of coxibs, and more specifically of Rofecoxib use, on the evolution of eGFRCG in separate analyses.


Baseline data

Patient characteristics at baseline are separately summarised for ‘NSAID users’ and ‘NSAID naive’ (table 1). In the crude analysis, median eGFRCG at inclusion was 2 mL/min higher in ‘NSAID users’ than in ‘NSAID naive’ (p=0.044). More ‘NSAID naive’ exhibited renal hyperfiltration at baseline (χ2=6.48, p=0.011).

Table 1

Patient characteristics, renal function and comorbidities at baseline

Primary outcome

In the total population, eGFRCG decreased by 0.83 (95% CI −1.08 to −0.57) mL/min per year. In ‘NSAID users’, crude eGFRCG decreased with a similar slope of the curve (mean −0.87 mL/min/year, 95% CI −1.15 to −0.59) as in ‘NSAID naive’ (mean −0.67 mL/min/year, 95% CI −1.26 to −0.09, p=0.63). Adjustments for potential confounding variables minimised the baseline differences in eGFRCG, but the decline in renal function remained almost identical (p=0.64) in the two groups (figure 1). Significant negative predictors for eGFRCG decline were age, baseline hyperfiltration, arterial hypertension, registered kidney and heart disease. In contrast, being female, having higher body mass index and using tumour necrosis factor-α (TNF)-antagonists had significant positive effects on eGFRCG outcome. Similar proportions of ‘NSAID naive’ (2.1%) and ‘NSAID users’ (3.0%) experienced a reduction in eGFRCG of >30% after 1 year (Pearson’s χ2=2.47, p=0.116).

Figure 1

Evolution of mean glomerular filtration lines according to the Cockroft–Gault formula, eGFRCG,22 is depicted from baseline for a period of four years for ‘non-steroidal anti-inflammatory drug (NSAID) users’ (line with open boxes) and for ‘NSAID naive’ (dashed line with black diamonds). Divergence of baseline data to table 1 in the depicted trajectories is caused by adjustments for baseline parameters age, sex, baseline chronic kidney disease stage, diabetes, arterial hypertension, heart disease and renal disease. The trend in renal function is identical in the two groups (p=0.63). Vertical lines represent the 95% CI of the means. They are overlapping at any time, but are only depicted in divergent directions for better visibility. The numbers in the table represent patients under observation at given time points.

Secondary outcome parameters and sensitivity analyses

All eGFRCG data were confirmed by the results of secondary outcome parameters (eGFRMDRD: likelihood ratio (LR) χ2=1.19, p=0.55; eGFRCKD: LR χ2=4.51, p=0.11; creatinine serum concentrations: LR χ2=1.50, p=0.47), with no significant differences between ‘NSAID users’ and ‘NSAID naive’. In the sensitivity analyses, the longitudinal models for cumulative dose effects of NSAIDs indicated similar eGFRCG deterioration in patients with NSAID use at >50% of the visits compared with the ‘≤50% users’ (p=0.08, figure 2), or when ‘>50% users’ were compared with ‘NSAID naive’ patients (p=0.87). Analogous observations were made for ‘>80% users’ and for ‘>90% users’ versus ‘NSAID users’ below the respective thresholds (p=0.77 or p=0.54), and in comparison with ‘NSAID naive’ patients (p=0.66 or p=0.52). These results were confirmed when the ‘maximal possible NSAID exposure time’ before a visit was used instead of the simple visit counts with NSAID registration. eGFRCG evolved even slightly better in patients ‘currently on NSAIDs’ (+0.1 mL/min/year, 95% CI −0.34 to 0.54) versus “not on NSAIDs”. Finally, neither coxibs as a class (LR χ2=2.50, p=0.287) nor Rofecoxib alone (LR χ2=2.50, p=0.68) demonstrated significantly different effects from the other NSAIDs.

Figure 2

Evolution of mean glomerular filtration lines according to the Cockroft–Gault formula, eGFRCG,22 is depicted from baseline for a period of four years in ‘frequent non-steroidal anti-inflammatory drug (NSAID) users’ (NSAID registered at >50% of visits, line with open boxes) and a group of more ‘occasional NSAID users’ (NSAID registered at ≤50% of visits, dashed line with black diamonds). The evolution of renal function is more favourable in patients with more frequent NSAID use (p=0.08) despite statistical adjustments for confounding. Vertical lines represent the 95% CI of the means. They are overlapping at any time. The numbers in the table represent patients under observation at given time points.

Effect modification by baseline renal function

The slope of eGFRCG decline was in general steeper over time in the 17 patients with CKD stages 4 or 5 (9.98±2.12 mL/min/year) than in patients with CKD stages 1 and 2 (1.27±0.17 mL/min/year) or stage 3 (1.04±0.29 mL/min/year). We observed no effect modification of eGFRCG decrease with NSAID use in patients with CKD stages 1–3. Despite even higher absolute changes (mean 21.3 mL/min, 95% CI −32.2 to −10.3, p=0.001) for the total group of patients with hyperfiltration at baseline, the slope of their eGFRCG curves remained unaffected by exposure to NSAIDs (coefficient 4.24, 95% CI −5.31 to 13.8, p=0.384). In contrast, only in patients with baseline eGFRCG <30 mL/min, according to the CKD stages 4 or 5, eGFRCG decline was significantly faster while being on NSAID therapy (delta 10.19, 95% CI −20.14 to −0.23, p=0.045).

Effect modification by disease activity

Inflammatory activity of RA was measured by the disease activity score DAS28 at baseline and over time.26 This parameter had no impact on GFR estimates in any of the performed analyses.

Renal NSAID toxicity

Investigator-defined NSAID-related renal toxicity was reported in five subjects (see online supplementary appendix table S2).

Participants lost to follow-up

The majority of patients (n=2419, 60%) remained under observation until the database was formally closed for this analysis, but the voluntary and non-paid data entry was stopped before in 1.6% to 7.1% of the patients per year. These patients are further characterised in supplementary appendix table S3. Preferentially older and multimorbid patients with a worse state of RA left the database prematurely. Death was reported in 185 patients and was predominantly caused by cardiovascular disease or malignancies (mortality rate 14.3 cases per 1000 patient-years), but reasons for the loss to follow-up remained unclear in another 1497 cases.


The mean decline in renal function in this representative at-risk RA population with regard to NSAID use, age, comorbidities and mortality was overall not higher than in reports from healthy individuals.15 ,31 The evolution in eGFRCG differed only significantly to the disadvantage of patients with NSAIDs when baseline eGFRCG was <30 mL/min; no class-specific, substance-specific or cumulative dosing NSAID effects occurred on the kidney function over time.

Taking into account the many patients considered to have relevant kidney problems, the literature regarding prospectively collected eGFR data from patients undergoing NSAID therapy is surprisingly limited. Renal complications were rare in prospective population studies with low-risk constellations30 ,32 ,33 or probably a priori avoided by patient exclusion in randomised controlled trials and subsequent meta-analyses.30 In contrast, retrospective case–control studies were able to address the clinical endpoints but may have selected only the ‘worst cases’. Large to very large study populations were necessary after the withdrawal of phenacetin, the major noxious substance of chronic analgesic nephropathy, 34 to demonstrate clinically apparent kidney injury of the remaining NSAIDs in these studies.1 ,3 ,5 ,6 ,34–36

The low frequency of stochastically occurring clinical endpoints in the present study is within the range of published incident numbers between 1 in 20035 and 1 in 10 000 cases6 depending on study-specific risk constellations. With the large visit intervals in our study, we did not repeat published rapid declines in eGFR starting shortly after NSAID exposure in high-risk populations.4 In case patients on NSAIDs developing renal problems had stopped their treatment in reaction to deteriorated renal function between the visits without reporting, this interruption would have happened in time and without irreversible effects on the renal function outcome in study completers.

The registration rate of renal disease in the present was comparable to other RA cohorts37 ,38 for which eGFRs are unknown. However, prevalence of kidney disease appeared remarkably underestimated in the present cohort with regard to the calculated eGFRCG. In this study, the creatinine levels were locally collected, which allowed to prospectively follow the evolution of renal function surrogates in a representative patient population exposed to NSAIDs despite established renal risk factors.28 The creatinine measurement methods complied with national standardisation requirements, and all laboratories had successfully passed the mandatory annual round robin tests, but more details about the applied measurement method are unknown. The high fluctuation of eGFRCG over time in either direction appears to occur at random, explainable by the fact of maximum 30% allowed variation in measuring plasma creatinine in these quality controls (see online supplementary appendix figure S1). Variability was similarly high for the secondary outcomes, which may have other advantages over eGFRCG.39

Observational studies have inherent limitations. We do not know possible selection effects on patient referral to rheumatologists in relation to the patients’ comorbidities. Selection bias must be considered with regard to better preservation of renal function in current NSAID users and in patients with high NSAID registration rates. We attempted to adjust for any differences of exposure variables at baseline, which did not essentially affect the course of eGFRCG. Unmeasured confounding factors can be considered on the level of unintended effects.40 We had no data on the precise cumulative dose of NSAID intake, which may have introduced a significant type II error in case of only occasional use. Furthermore, data on the use of diuretics, antihypertensive drugs or antibiotics were missing. It is unlikely that patients enrolled in this study were less intensively treated for their registered comorbidities with these and other renal risk drugs in the ‘fee for service’-based Swiss healthcare system than in any other industrial country. Consequently, the same unreported renal risks in NSAID users and NSAID-naive patients would have biased any differences towards the null, but the favourable course of renal function in the total cohort makes suchlike significant effects of unreported renal risks on eGFR unlikely.15 ,31 Study attrition even in an expected range in contrast remains the main concern for the validity of the present data,41 although left censorship of unreported renal events may be put into the perspective of identical NSAID registration rates and similar evolution in renal function in the last documented interval in completers and in patients with premature end of observation (see online supplementary appendix table S3).

In summary, aggravated renal function decline in our study occurred only in patients with CKD 4 or 5 at baseline. Though we have limited data, we like to propose <30 mL/min as a preliminary cut-off for definitive renal risk situation on the basis of renal function and contraindication for NSAIDs. We found reassuring data regarding preserved renal function despite long-term NSAID use in the remaining majority of RA patients. According to this finding, it appears unjustified to generally deny an effective NSAID treatment in patients after having considered less critical means. Nevertheless, whenever using NSAIDs to the benefit of anguished patients with an eGFRCG ≥30 mL/min, ongoing efforts in vigilance are mandatory. This recommendation is especially true but not limited to established renal risk situations.


We are grateful to the rheumatologists and patients that participated in the SCQM and thereby made this study possible. A list of rheumatology offices and hospitals that are contributing to the SCQM registries can be found at We would also like to emphasise our gratitude to Dr Hans Schwarz, former president of SCQM, for his ongoing successful and generous efforts, and to Doris Wisler for her valuable support. Editing and proofreading for language and style were performed by San Francisco Edit, Mill Valley, CA.


View Abstract

Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Handling editor Tore K Kvien

  • Contributors BM planned this study, performed statistical analyses and is responsible for the overall content as guarantor. AS, MP and SA made contributions to the methods and discussion. PMV provided administrative support. AF gave methodological advice and supervised all statistical procedures. All authors approved the final manuscript.

  • Funding This study on SCQM data received no study-specific funding. The SCQM Foundation has received grants from the Swiss Health authorities (BAG), the Swiss Academy for Medical Sciences (SAMW), the JL Warnery Foundation and the Swiss Society of polyarthritis patients (SPV). The pharmaceutical companies Abbott, Merck (formerly Essex Switzerland), Pfizer (formerly Wyeth Switzerland), Roche, Bristol-Myers Squibb, Mepha, Novartis and Sanofi-Aventis are patrons of SCQM. SCQM patrons and sponsors had no role in study design, data collection, data analysis, data interpretation, writing of this manuscript or the decision to submit this manuscript.

  • Competing interests None.

  • Patient consent Obtained.

  • Ethics approval Swiss Academy of Medical Sciences review board.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.