Customize your JAMA Network experience by selecting one or more topics from the list below.
Curhan GC, Knight EL, Rosner B, Hankinson SE, Stampfer MJ. Lifetime Nonnarcotic Analgesic Use and Decline in Renal Function in Women. Arch Intern Med. 2004;164(14):1519–1524. doi:10.1001/archinte.164.14.1519
Analgesics are commonly used and may impair kidney function. However, limited prospective information is available on the long-term effects of aspirin, other nonsteroidal anti-inflammatory drugs (NSAIDs), and acetaminophen on renal function.
A total of 1697 women participating in the Nurses' Health Study provided information on a mailed questionnaire in 1999 about lifetime use of acetaminophen, aspirin, and NSAIDs and provided blood samples in 1989 and 2000. The main outcome was change in estimated glomerular filtration rate (GFR) in 11 years. Multivariate logistic regression was used to determine the odds of developing the outcome according to lifetime analgesic intake.
The mean ± SD estimated GFR decreased from 88 ± 17 to 79 ± 17 mL/min per 1.73 m2. There were no substantial differences in the unadjusted or estimated GFR levels among the categories of lifetime intake for the 3 analgesic groups at baseline or after 11 years. Acetaminophen use was associated with an increased risk of a GFR decline of at least 30 mL/min per 1.73 m2 (P trend = .01) and a GFR decline of 30% or greater (P trend<.001), but aspirin and NSAID use were not. Compared with women consuming less than 100 g of acetaminophen, multivariate-adjusted odds ratio (95% confidence intervals) for a decline in GFR of at least 30 mL/min per 1.73 m2 for women consuming more than 3000 g was 2.04 (1.28-3.24).
Higher lifetime use of aspirin and NSAIDs is not associated with renal function decline, but high acetaminophen use may increase the risk of loss of renal function. The absolute risk of renal function decline due to even high lifetime analgesic intake seems to be modest.
Analgesic use is common in the general population, particularly among women.1 Analgesics may impair kidney function by altering prostaglandin production,2 causing interstitial nephritis,3 or potentially through a direct toxic effect on the tubular epithelium. However, limited and inconsistent information is available on the long-term effects of aspirin, other nonsteroidal anti-inflammatory drugs (NSAIDs), and acetaminophen on renal function.
A recent large case-control study4 (N
= 1898) found a greater than 2-fold increased risk of newly diagnosed chronic renal insufficiency for regular users of acetaminophen or aspirin but not for NSAID users. Three previous cohort studies examined the association between analgesic use and change in renal function. The prospective Swiss watch factory study5 (N = 623) found an increased risk with phenacetin use (banned in the United States in 1983), but aspirin use was not associated with an increased plasma creatinine level. A prospective study6 from Belgium (N = 400) found a higher risk of decreased creatinine clearance among abusers of combination analgesic preparations, but they did not examine individual analgesics. A study7 of US male physicians (N = 11 032) assessed creatinine concentration at one point in time and found no association between the previous use of aspirin, NSAIDs, or acetaminophen and elevated creatinine concentration or reduced estimated creatinine clearance. In a follow-up substudy of 4494 subjects, these investigators found no association between analgesic use and change in renal function over a 14-year period.8
The optimal measure of renal function for large epidemiologic studies has not been resolved.9 Although plasma creatinine level alone is not adequate, estimates derived from plasma creatinine and other variables seem to approximate more closely the glomerular filtration rate (GFR). Examining the change in renal function over time is more informative than a single measurement.
To examine these issues further, we studied the association between lifetime use of acetaminophen, aspirin, and NSAIDs and change in renal function in 11 years in 1697 women participating in the Nurses' Health Study.
The Nurses' Health Study was established in 1976 when 121 700 female registered nurses aged 30 to 55 years from 11 US states completed and returned a mailed questionnaire.10 The participants provided information on their medical history, lifestyle practices, and other exposures of interest. Every 2 years, follow-up questionnaires were mailed to obtain updated information and to identify newly diagnosed diseases. On the 1990, 1992, and 1998 biennial questionnaires, we asked about the frequency of use of acetaminophen, aspirin, and NSAIDs.
We limited the initial study population to the 32 826 participants who had provided a blood sample in 1989. Women with a history of cancer (except nonmelanoma skin cancer) or cardiovascular disease (myocardial infarction, angina, stroke, or transient ischemic attack) were excluded from the blood sample collection in 1989. The characteristics of the women who provided blood samples were similar to those of the total cohort in terms of age, weight, hypertension, diabetes mellitus, and hypercholesterolemia, but the former group was slightly less likely to currently smoke. To enrich our final study population with participants who were likely to have had high lifetime use of analgesics, we identified from among the 32 826 women those who reported using acetaminophen, aspirin, or NSAIDs 15 or more days per month on both the 1992 and the 1998 biennial questionnaires. To identify women who were likely to have low lifetime intake, we also included all women who reported no use of any of the 3 analgesics on the 1990, 1992, and 1998 biennial questionnaires. There were 4238 eligible women who provided a blood sample in 1989 and who fulfilled the criteria for frequency of analgesic use. A supplementary questionnaire (see the following subsection) was mailed in 1999 to collect detailed information on current and lifetime use of analgesics from these 4238 women. The supplementary questionnaire was returned by 3876 women (91%). A second blood sample was collected in 2000 from participants who originally provided a specimen in 1989. From those who returned the supplementary questionnaire and provided a second blood sample (n = 3123), we selected all women with lifetime consumption of at least 1501 tablets of one of the analgesics and a random sample of women with intake of less than 1501 tablets. Owing to cost constraints, plasma creatinine concentration was not measured in all participants. For the 1769 women selected, we measured plasma creatinine levels in the blood samples collected in 1989 and 2000. After excluding women with creatinine levels less than 0.5 mg/dL (<44 µmol/L) (n = 61) and women with a body mass index less than 17 (calculated as weight in kilograms divided by the square of height in meters) or a missing body mass index measurement (n
= 11), there were 1697 women included in the analysis.
We mailed a supplementary questionnaire in 1999 to collect detailed information on the current use of each of the 3 analgesic classes, including frequency in days per months, tablets per day, tablet dosage, brand, and indication for current use. The questionnaire also asked about total consumption in 2 periods: the past 10 years and before 1990. The total number of tablets taken in those 2 periods was collected in 11 categories: none, 1 to 100, 101 to 500, 501 to 1000, 1001 to 1500, 1501 to 3000, 3001 to 5000, 5001 to 10 000, 10 001 to 15 000, 15 001 to 20 000, and 20 001 or more. For the main analysis, we used the combined total from the 2 periods by adding the midpoints of the categories. We converted number of tablets to lifetime intake, in grams, by multiplying the total number of tablets (the midpoint of each category) by the most common dosage of each analgesic (aspirin and acetaminophen, 325 mg; and NSAIDs, 200 mg). For example, if a participant reported consuming 3001 to 5000 tablets of aspirin, the estimated lifetime intake would be calculated as follows: 4000 tablets × 325 mg per tablet
= 1300 g of aspirin.
Blood samples were collected by study participants in 1989 and again in 2000. The samples were collected locally and returned to us by overnight mail in ice packs. On receipt, the plasma was separated and stored in liquid nitrogen freezers (−130°C or colder). In 2001, we thawed all of the samples at the same time and sent them to the laboratory at Boston Children's Hospital for measurement of plasma creatinine levels. Both samples from each woman were run in the same laboratory batch. Creatinine concentration was measured using an autoanalyzer (Hitachi Ltd, Tokyo, Japan) and the modified kinetic Jaffe reaction. The overall coefficient of variation of the masked quality control samples was 10%.
Two different measures of renal function were used (creatinine concentration expressed in milligrams per deciliter [micromoles per liter]): (1) the simplified version of the Modification of Diet in Renal Disease formula for estimating the GFR11: [186 × creatinine − 1.154 × age − 0.203 × 0.742 × 1.210 (if black)], expressed in milliliters per minute per 1.73 m2, and (2) creatinine clearance estimated using a modified Cockcroft-Gault formula: [146 − age] × [(weight kg × 0.287) + (height2 × 9.74)]/[60 × creatinine], expressed in milliliters per minute.12 The results of the Modification of Diet in Renal Disease GFR and creatinine clearance were similar; for brevity, we provide in the tables only the results of the Modification of Diet in Renal Disease GFR.
Weight was reported on each biennial questionnaire. A participant was considered to have hypertension or diabetes mellitus if these had been reported on any biennial questionnaire up to and including the 1990 questionnaire. Blood pressure was self-reported on the 1990 questionnaire, with 9 categories for systolic (ranging from <105 to ≥175 mm Hg) and 7 categories for diastolic (ranging from <65 to ≥105 mm Hg). Smoking status was based on the response to the 1990 questionnaire.
To compare our results with those of the recent Swedish case-control study,4 we chose to use the same lifetime consumption categories (0, 1-99, 100-499, 500-2999, and ≥3000 g). We combined the 2 lowest categories because we had too few women who had never taken any of the analgesics and the odds ratio (OR) in the Swedish study4 for the 1- to 99-g group was close to 1.0. Two outcomes were examined: an estimated GFR decrease of 30 mL/min per 1.73 m2 or greater and an estimated GFR decrease of 30% or greater. Multivariate logistic regression was used to calculate ORs and 95% confidence intervals (95% CIs) after controlling simultaneously for potential confounders. Variables considered in the models were age (continuous), weight (continuous), history of hypertension, categories of systolic and diastolic blood pressure, history of diabetes mellitus, current smoker, and categories (including "missing") of lifetime consumption for all 3 analgesics. We also examined whether the results changed after adjusting for the development of hypertension or diabetes mellitus during the study. Multivariate tests for trend for the analgesics were assessed by using the midpoint of the lifetime consumption category and by modeling the continuous variable. All analyses were performed using a statistical software program (SAS version 6.1; SAS Institute Inc, Cary, NC).
The characteristics of the cohort according to categories of lifetime analgesic intake are given in Table 1. In each class of analgesic, individuals in the lowest category of lifetime intake had lower mean weight, systolic and diastolic blood pressure, and frequency of hypertension and diabetes mellitus compared with the other nonmissing categories.
The means of the different assessments of renal function at baseline and at 11 years of follow-up according to lifetime analgesic consumption are given in Table 2. Overall, the mean ± SD plasma creatinine level increased from 0.75 ± 0.14 to 0.81 ± 0.17 mg/dL (from 66 ± 12 to 72 ± 15 µmol/L) and the mean ± SD estimated GFR decreased from 88 ± 17 to 79 ± 17 mL/min per 1.73 m2. Only 7 women (0.4%) had a baseline plasma creatinine value greater than or equal to 1.3 mg/dL (≥115 µmol/L). There were no substantial differences in the unadjusted mean creatinine level or estimated GFR among the categories of lifetime intake for the 3 analgesic groups at baseline or at 11 years. In addition, the unadjusted change in creatinine level and estimated GFR did not vary substantially by analgesic intake in 11 years.
There were 168 women (10%) for whom the estimated GFR decreased by at least 30 mL/min per 1.73 m2 during the 11-year study, and the multivariate ORs are given in Table 3. Acetaminophen use was significantly associated with increased risk (P trend = .01), but aspirin (P trend = .71) and NSAID (P trend = .88) use were not. The unadjusted proportion of women who had consumed less than 100 g of acetaminophen and had at least a 30–mL/min per 1.73 m2 decline in GFR was 7% compared with 13% of women who had consumed 3000 g or more. Compared with women who had consumed less than 100 g of acetaminophen, the odds of at least a 30–mL/min per 1.73 m2 GFR decline were significantly increased in women who had consumed 100 to 499 g (OR, 1.80; 95% CI, 1.02-3.17), 500 to 2999 g (OR, 2.23; 95% CI, 1.36-3.63), and 3000 g or more (OR, 2.04; 95% CI, 1.28-3.24). When change in estimated creatinine clearance was used, the magnitudes of the estimates were attenuated and no longer statistically significant (P trend = .06).
There were 177 women for whom the estimated GFR decreased by at least 30% in 11 years, and the multivariate ORs are given in Table 4. Acetaminophen use was significantly associated with risk of GFR decline (P trend<.001), but aspirin (P trend<.20) and NSAID (P trend<.56) use were not associated with risk. The unadjusted proportion of women who had consumed less than 100 g of acetaminophen and had a 30% or greater decline in estimated GFR was 8% compared with 15% of women who had consumed 3000 g or more. The odds of at least a 30% decline in GFR increased with increasing lifetime intake. Compared with women who had consumed less than 100 g of acetaminophen, the odds were significantly increased in women who had consumed 100 to 499 g (OR, 1.40; 95% CI, 0.79-2.49), 500 to 2999 g (OR, 1.64; 95% CI, 1.00-2.69), and 3000 g or more (OR, 2.19; 95% CI, 1.40-3.35). When change in estimated creatinine clearance was used, the magnitudes of the estimates were attenuated (P trend = .04).
We found no significant interaction between aspirin and acetaminophen use, even for women who had consumed 500 g or more of acetaminophen and 500 g or more of aspirin.
The results of the multivariate analyses did not materially change after excluding women with a self-reported history of abnormal kidney function or those taking diuretics or angiotensin-converting enzyme inhibitors at the time of the second blood sample collection. In addition, the findings did not materially change after adjusting for (1) development of hypertension or diabetes mellitus during the study or (2) frequency of use of the individual analgesics in 2000. Furthermore, we explored whether the findings differed by baseline estimated GFR. We used as the outcome a decline in GFR of 25% or more because there were too few cases in the subgroups when we used a change of 30% or more. The results were similar whether the baseline GFR was greater than or equal to 90 or less than 90 mL/min per 1.73 m2.
In 11 years, we observed no association between lifetime use of aspirin or NSAIDs and risk of renal function decline, even in women who had consumed 3000 g or more of the drug. In contrast, women who had consumed 100 g or more of acetaminophen in their lifetime seemed to have a statistically significantly greater risk of losing an important proportion of their renal function compared with women who had consumed less than 100 g. However, most individuals who had consumed 3000 g or more did not have clinically important renal dysfunction. In addition, the magnitude of the association depended on the formula used to estimate renal function. There was no apparent interaction between acetaminophen and aspirin.
This study differs from previous studies in several important ways. First, renal function was assessed at 2 points in time 11 years apart, allowing examination of the impact of analgesic use on change in renal function. Second, we selected participants based on their reported lifetime use of analgesics rather than on the outcome of interest, that is, renal function. This allowed us to oversample individuals with a broad range of lifetime intakes for each of the 3 analgesic classes. Third, we examined clinically meaningful absolute and percentage changes in the estimated GFR. Fourth, because it is well known that plasma creatinine level alone does not estimate the true GFR very well, we used formulas that have been derived to improve this estimate. Finally, because participants were selected from a general population, the results apply to most analgesic users who do not already have known established renal dysfunction.
Our findings support some previous studies of acetaminophen and risk of renal dysfunction. In a study involving 716 cases and 361 controls, Perneger et al13 found a 2-fold higher risk of end-stage renal disease in individuals who had consumed more than 1000 tablets of acetaminophen. However, that study assessed analgesic use after the cases had already begun dialysis, thus introducing the possibility of recall bias or confounding by indication. In addition, it is not clear whether the results could be generalized to individuals who did not have renal dysfunction.
A recent case-control study4 in Sweden of newly diagnosed renal dysfunction (defined as a plasma creatinine level ≥3.4 mg/dL [≥300.6 µmol/L] in men and ≥2.8 mg/dL [≥247.5 µmol/L] in women) identified 918 cases. The investigators found that individuals who consumed 500 g or more of acetaminophen during their lifetime had an increased risk (OR, 3.3; 95% CI, 2.0-5.5) compared with never users.
In contrast, a study7 of 11 032 male physicians aged 54 to 98 years found no increased risk of renal dysfunction (defined as a plasma creatinine level ≥1.5 mg/dL [≥133 µmol/L]) with acetaminophen use, even with intakes greater than 7000 tablets (approximately 2000 g). In a subsequent study, analgesic use was not associated with change in renal function over a 14-year period.8
The mechanism by which acetaminophen might cause renal function decline is unclear. After administration of acetaminophen, which itself is the principal metabolite of phenacetin, potentially toxic metabolites, particularly p-aminophenol, are concentrated in the renal papilla.14 Nephrotoxicity may be the result of oxidized metabolites binding covalently to sulfhydryl-containing tissue macromolecules and depleting stores of reduced glutathione, leading to cell death.
Our aspirin results are consistent with those of the Swiss prospective study5 and the male physician study,7 which found no association between aspirin use and adverse renal outcomes. Our results differ from those of the Swedish case-control study,4 which found an OR of 1.9 (95% CI, 1.3-2.9) for individuals who had consumed 500 g or more of aspirin in their lifetime. The explanation for the disparate findings is unclear but may be related to differences in study design.
Published results of the association between NSAID use and risk of renal dysfunction have been inconsistent. These drugs have well-documented acute effects on the kidney, including decreased GFR; diminished ability to excrete sodium, potassium, and water; and interstitial nephritis with glomerular changes.3 Although some case-control studies13,15 found an increased risk, our findings are consistent with the recent studies4,7 that found that NSAID use is not associated with risk of chronic renal dysfunction.
We did not study combination preparations, which are uncommon in the United States. A prospective cohort study6 from Belgium found that in 7 years, analgesic abusers were at higher risk of developing renal dysfunction (defined as being below the third percentile rank) than controls. As most abusers consumed preparations containing more than 1 ingredient, the investigators could not comment on individual analgesics.
The limitations of this study deserve comment. Similar to other published studies, we could not validate self-reported analgesic use because of the lack of independent records for these medications, which are available without a prescription. However, it is unlikely that there was substantial recall bias as 98% of nurse participants did not think that they had abnormal kidney function at the time of the assessment of analgesic use, and only 0.4% had a plasma creatinine value of 1.3 mg/dL or greater (≥115 µmol/L) at baseline. We did not collect information on the time of use for lifetime number of tablets before the first blood sample collection. Thus, some individuals may have had high use a decade or two before the first collection but then not have used the analgesic since that time. Therefore, we could not evaluate the interval between consumption of a certain cumulative amount of an analgesic and the first detection of an impact on renal function. Similarly, 11 years of follow-up may be too short to assess the long-term impact of analgesic use on the development of chronic renal dysfunction. We had too few women with substantially reduced renal function to assess whether preexisting renal disease is required for an adverse impact of analgesics, as has been suggested elsewhere.4 We also did not have information on the indication for use in the past, and it is possible, although unlikely, that the reason for taking the analgesic may be associated with renal dysfunction. Common indications for frequent use of analgesics in younger women include headache, musculoskeletal pain, and menstrual cramps; however, none of these conditions have been reported to be associated with risk of developing renal dysfunction, so confounding by indication is not likely to explain our findings. COX-2 inhibitors were not introduced in the United States until 1999; hence, long-term data on their impact on renal function are not yet available.
The choice of the appropriate measure of renal function for epidemiologic studies is difficult.9 Creatinine concentration alone may underestimate GFR, especially in older individuals with normal plasma creatinine values.16,17 Therefore, we used 2 formulas that were developed to more accurately measure renal function. The modified Cockcroft-Gault formula was developed and validated in women with normal creatinine clearance.12 The Modification of Diet in Renal Disease formula has been recommended by the National Kidney Foundation to estimate GFR,11 but the validation study included few individuals with GFR greater than or equal to 90 mL/min per 1.73 m2, so it is unclear how well it predicts true GFR in individuals with normal renal function. It is somewhat reassuring that both estimates of renal function provided similar findings, but the magnitude and strength of the association were lower when the modified Cockcroft-Gault formula was used. Thus, the choice of formula affects the interpretation of the results.
Most women who met our outcome criteria (using any of the definitions) would not necessarily be considered to have clinically important renal dysfunction. Nonetheless, if the observations hold true, over time the individuals who had higher lifetime intake of acetaminophen would be more likely to develop clinically relevant renal dysfunction.
In conclusion, higher lifetime use of acetaminophen was associated with a greater decline in renal function at 11-year follow-up, whereas lifetime use of aspirin and NSAIDs was not. Most individuals with very high lifetime intake will not experience an adverse effect on their rate of renal function decline. Each of these analgesics has other potential adverse effects that should be considered when deciding whether an analgesic is needed and which one to use. Future studies should address whether these analgesics affect the rate of decline of renal function in individuals with established renal disease.
Correspondence: Gary C. Curhan, MD, ScD, Channing Laboratory, Brigham and Women's Hospital, 181 Longwood Ave, Boston, MA 02115 (Gary.Curhan@channing.harvard.edu).
Accepted for publication September 4, 2003.
This work was supported by grants DK52866 and CA87969 from the National Institutes of Health, Bethesda, Md.
Create a personal account or sign in to: