Asgari MM, Chren M, Warton EM, Friedman GD, White E. Association Between Nonsteroidal Anti-inflammatory Drug Use and Cutaneous Squamous Cell Carcinoma. Arch Dermatol. 2010;146(4):388-395. doi:10.1001/archdermatol.2009.374
Copyright 2010 American Medical Association. All Rights Reserved. Applicable FARS/DFARS Restrictions Apply to Government Use.2010
To examine the association between nonsteroidal anti-inflammatory drug (NSAID) use and cutaneous squamous cell carcinoma (SCC).
Retrospective case-control study.
Kaiser Permanente Northern California (KPNC), a large population based-health maintenance organization.
Random sample of 415 KPNC members diagnosed as having a pathologically verified SCC in 2004 and 415 age-, sex-, and race-matched controls with no history of skin cancer.
Main Exposure Measure
Self-reported NSAID use in the 10 years prior to baseline. Use of NSAIDs was categorized based on type (any NSAIDs, aspirin, ibuprofen, and nonaspirin NSAIDs). Odds ratios (ORs) and 95% confidence intervals (CIs) were calculated using conditional logistic regression to estimate the association of SCC with regular use, dose, and duration of exposure to the different NSAID types. Information on pharmacy-dispensed NSAIDs was also examined to assess its association with SCC risk. Models were adjusted for all ascertained SCC risk factors (fully adjusted model) and only those variables associated with both SCC risk and NSAID use (parsimonious model).
Fully adjusted analyses showed no statistically significant reduction in SCC risk with self-reported regular use of any NSAID (OR, 1.32; 95% CI, 0.92-1.89), aspirin (OR, 1.38; 95% CI, 0.96-1.97), ibuprofen (OR, 0.74; 95% CI, 0.46-1.19), or nonaspirin NSAIDs (OR, 0.84; 95% CI, 0.56-1.26). Analyses examining duration, dose, and variables combining duration and dose of NSAID exposure did not appreciably change results. An analysis using the parsimonious model showed similar results. The data on pharmacy-dispensed NSAIDs also showed no association with SCC risk.
Neither self-reported nor pharmacy-dispensed NSAID exposure was associated with cutaneous SCC risk.Published online February 15, 2010 (doi:10.1001/archdermatol.2009.374).
Nonsteroidal anti-inflammatory drugs (NSAIDs) inhibit the enzyme cyclooxygenase (COX), thereby blocking the synthesis of proinflammatory prostaglandins. In addition to their anti-inflammatory properties, NSAIDs also inhibit neoplastic proliferation by inducing apoptosis and inhibiting angiogenesis.1,2 Epidemiologic studies and randomized trials have shown protective effects of NSAIDs for several cancers including colorectal,3,4 breast,5 prostate,6 and lung.7 Laboratory studies suggest that NSAIDs exert protective effects against cutaneous squamous cell carcinomas (SCCs) both in vitro2,8 and in animal models.9- 12 However, few epidemiologic studies have examined the association between NSAID use and SCC risk, and these have yielded conflicting information.13- 18 The conflicting results may be due, in part, to differing methods for ascertaining NSAID exposure (eg, relying solely on pharmacy-dispensed NSAIDs),13,14 differing definitions of regular NSAID use,16,18 and lack of adjustment for potential confounding variables such as sun sensitivity.18
We conducted a case-control study to investigate whether NSAID use is associated with SCC using the Kaiser Permanente Northern California (KPNC) population. KPNC electronic records include a comprehensive pathology database through which cutaneous SCCs can be accurately identified. The epidemiology of cutaneous SCCs is otherwise difficult to study because unlike most malignant neoplasms, cutaneous SCCs are not reportable to national registries such as the Surveillance Epidemiology and End Results (SEER) program, they do not have unique International Classification of Disease, Ninth Revision (ICD-9) identifiers, and patients' self-reports of a history of SCC are not accurate.19 We used a self-administered questionnaire to ascertain SCC risk factors and exposure to NSAIDs by type, dose, and duration over the 10 years preceding the diagnosis date. Based on previously published literature,20 we hypothesized that NSAID use would be associated with a reduction in risk of SCC.
This investigation was a case-control study of 415 KPNC members (ages 43 to 85 years) with pathologically confirmed SCC diagnosed in 2004 and 415 control subjects matched to cases by year of birth, sex, and self-reported race. To minimize risk of exposure misclassification due to poor recall of NSAID use, we excluded members older than 85 years and those with a diagnosis of dementia from January 1, 1994, to December 31, 2004. Members whose native language was not English were also excluded to reduce the likelihood of error in completing the self-administered questionnaire. This study was approved by the institutional review board of Kaiser Foundation Research Institute and was conducted according to the Declaration of Helsinki principles.
A pool of potential cases were identified from KPNC electronic pathology records by querying all electronic pathology reports of specimens collected between January 1, 2004, and December 31, 2004, that contained the word “skin” in the specimen line and the phrase “squamous cell carcinoma” or “SCC” in the diagnosis line. Each pathology report was reviewed by a dermatologist (M.M.A.), and only those tumors that were extragenital and nonmucosal and had a definitive diagnosis of SCC were included in the pool of eligible cases. The index date was defined as the date of SCC diagnosis.
In keeping with an institutional review board mandate that potential subjects with SCC be approached through their primary care provider, 1052 eligible cases were grouped and randomly selected by primary care provider. Of the 581 cases selected, 472 (81%) completed the questionnaire, and of those, 422 could be matched with a responding control. Control subjects were drawn from respondents to the 2005 Member Health Survey (MHS), a general health survey mailed to a random sample of KPNC adult members. We chose MHS participants because this survey asked questions on self-identified race and history of cancer. Of the MHS respondents who reported no history of cancer, potential controls were matched by year of birth ±1 year, sex, and race to the cases (n = 1801). One control was randomly selected within the case-control pairing and contacted to participate in the study. If they refused, another control was randomly selected within the pairing until a case-control pair was completed. We contacted 736 controls to achieve a matched control for all 422 cases (57% response).
Each participant was contacted by mail and asked to complete a 3-page self-administered questionnaire regarding various personal characteristics, medication use, health history, skin cancer history, and risk factors. The survey instrument was modeled off of a validated questionnaire to ascertain supplement use developed for the Vitamins and Lifestyle Study.21 For nonresponders who did not opt out of the study, the questionnaire was mailed a second time 3 weeks after initial contact. If no response was obtained from the second mailing and the patient did not opt out, the subject was contacted by telephone and the questionnaire administered by trained study personnel. Participants were offered a $5 gift card for completing the questionnaire. Subjects who had mailed in the questionnaire but had missing pertinent variables on NSAID exposure and SCC risk factors were contacted by telephone and, if they agreed, the missing variables were administered to them.
Given the significantly increased risk profile of organ transplant patients, case-control pairs in which 1 member reported a history of an organ transplant were excluded (6 cases). In addition, 1 case-control pair had discrepant self-reported race on the questionnaire and was excluded, leaving 415 case-control pairs for analysis.
Information was ascertained on over-the-counter and prescription NSAID use through a self-administered questionnaire that inquired about the use of aspirin, ibuprofen, naproxen, bismuth subsalicylate, diclofenac, etodolac, indomethacin, nabumetone, piroxicam, salsalate, sulindac, celecoxib, rofecoxib, and valdecoxib. Common brand names were given as examples to facilitate recall. Acetaminophen use was also ascertained as a comparison drug, since it is used for many of the same indications as NSAIDs (eg, analgesic, antipyretic) but is not hypothesized to have chemoprotective properties for skin cancer. For each drug, individuals were asked about their patterns of use over the preceding 10 years including years taken (categorical variable, <1, 1-3, 4-6, 7-9, and ≥10 years), days per week of use (categorical variable, 1-2, 3-5, and ≥6 d/wk), number of pills taken per day (1, 2, 3-4, 5-6, and ≥7), and dose (in milligrams; dose categories were based on the medication in question).
We categorized the different NSAIDs into the following 4 groups: (1) any NSAIDs, (2) aspirin, (3) ibuprofen, and (4) nonaspirin NSAIDs. For each type, we examined 3 exposure measures. “Regular use” was defined as taking the medication at least once a week for at least 1 year. “Duration” represented categorized years of use in the past 10 years (range, 0-10 years). For subjects reporting more than 1 type of NSAID use, the longest reported duration was used for the exposure variable. The “dose” was the mean strength in milligrams per pill over the reported period of use and was calculated as:
(Days per Week/7) × Pills per Day × Dose per Pill.
The dose variable was further categorized for aspirin as “low dose” (≤81 mg) and “high dose” (>81 mg) and for ibuprofen as “low dose” (≤200 mg) and “high dose” (>200 mg) based on dosage during the stated period of use.
To simultaneously account for duration and frequency, the variable “pill-years” was defined as:
Pills per Day × Days per Week × Years of Use,
using the midpoint of each duration category. The pill-year variable was divided into 4 categories (0, >0 to <2, 2 to ≤5, and >5). If the respondent indicated that they had taken a specific NSAID but failed to complete the remaining information regarding that NSAID on the questionnaire, the lowest category of the missing frequency, dose, and/or duration was assumed.
The 2 most commonly reported NSAIDs, aspirin and ibuprofen, had wide variations in dose. To capture dose in addition to frequency and duration of use, we defined the variable “pill-year-dose” as:
Pill-Years × the Common Strength per Pill (in milligrams).
Since a pill-year is equivalent to taking 1 pill each day for 1 year, a 1 pill-year-dose for aspirin corresponds to an intake of one 325-mg tablet per day for 1 year, and for ibuprofen, to an intake of one 200-mg tablet per day for 1 year. There were not enough respondents to allow for meaningful analysis of pill-year-dose exposure for the remaining NSAIDs.
Longitudinal exposure to NSAIDs was ascertained for each study member using information found in automated KPNC pharmacy databases on filled prescriptions from the date the questionnaire was returned (baseline) to 10 years prior to baseline (encompassing 1995-2005). We searched for all known available pharmacy-dispensed NSAIDs requiring a prescription during the study period including nabumetone, etodolac, sulindac, indomethacin, piroxicam, salsalate, diclofenac, celecoxib, rofecoxib, and valdecoxib. We determined the number of refills for each medication type in the 10 years prior to the date the survey was returned. If the individual had at least 1 refill for a given NSAID, they were considered to be exposed. We also determined the cumulative days' supply for each NSAID type during the 10 years and calculated the years of exposure as a continuous variable. In addition, we created an ever/never exposure variable for any pharmacy-dispensed prescription NSAID (“any NSAID”) and used the maximum duration of any pharmacy-dispensed NSAID as the duration measure for the “any NSAID” variable.
Participants answered questions on variables known to influence SCC risk including skin type, history of freckling (yes/no), eye color, natural hair color, education, family history of skin cancer, history of sunburns, outdoor sun exposure, occupational sun exposure, tanning bed use (yes/no), high-risk exposures (UV light treatment, burn scar, nonhealing ulcers, radiation treatment, arsenic exposure, and exposure to industrial chemicals), and smoking (current vs former/never). For occupational exposure, we asked participants if their main occupation involved daily sun exposure of at least 2 hours between 10 AM and 4 PM (yes/no). We also inquired if respondents regularly (at least once a week) spent at least 2 hours outdoors between 10 AM and 4 PM, and if so, ascertained the average number of hours per week.
Differences in distributions of categorical covariates between cases and controls were analyzed using Pearson χ2 tests. For the matched case-control analysis, we used conditional logistic regression to estimate unadjusted and adjusted odds ratios (ORs) and Wald 95% confidence intervals (CIs) for each of the 4 NSAID exposure types with regular use, dose, duration, and pill-years as exposure measures. To ensure that our multivariate analysis was not overadjusted, we also analyzed the multivariate model limiting the number of variables to only those that were associated with both NSAID exposure and SCC risk at the P < .20 level (parsimonious model). The reference category for all analyses involving regular use and duration were users who reported consuming NSAIDs less than once a week for less than 1 year. The reference category for analyses involving dose were low-dose users (≤81 mg for aspirin and ≤200 mg for ibuprofen). P values were 2-sided. Power calculations indicated that a sample size of 415 cases and 415 controls with 59% of controls exposed to NSAIDs would result in a maximum detectable SCC risk reduction of 0.64 or a minimum detectable risk reduction of 36% (2-sided test; α = .05; power of 0.80, within matched-pair phi correlation coefficient of .20). All statistical analyses were performed using SAS version 9.1 software (SAS Institute Inc, Cary, North Carolina).
The mean (SD) age of participants at index date was 72.5 (8.6) years (range, 43-85 years). The majority of participants were male (n = 514 [62%]). Compared with controls, cases were more likely to have red or blond hair, blue or gray eyes, and lighter skin types. Cases were also more likely to report current smoking, a family history of skin cancer, a history of childhood freckles, routine sun exposure, and severe sunburns (Table 1).
The majority of study participants (61%) self-reported regular use of NSAIDs in the 10 years prior to baseline. The most commonly reported regularly used NSAIDs were aspirin (48%), ibuprofen (18%), naproxen (5%), and nabumetone (4%). Other NSAIDs (bismuth subsalicylate, diclofenac, etodolac, indomethacin, piroxicam, salsalate, sulindac, celecoxib, rofecoxib, and valdecoxib) were regularly used by less than 2% of the participants. Acetaminophen was regularly used by 19% of participants. There were no significant differences in sex, age at index date, or any measured SCC risk factors between regular and nonusers of any NSAID. Regular NSAID users were more likely to take acetaminophen (P = .005). There were no significant differences in reported type of NSAID regularly used between cases and controls except that 7 cases reported regular celecoxib use compared with only 1 control.
Regular use of any NSAID was not associated with a reduction in SCC risk (fully adjusted OR, 1.32; 95% CI, 0.92-1.89 [Table 2]). Although NSAID users whose exposure was of short duration (1-3 years) appeared to be at somewhat increased risk for SCC (adjusted OR, 1.94; 95% CI, 1.10-3.44), we found no consistent effects of duration of use of any NSAID on SCC risk (P value for linear trend, .69). In comparing regular aspirin users with nonusers, there was no negative association with SCC risk (adjusted OR, 1.38; 95% CI, 0.96-1.97). Duration of aspirin use was also not associated with SCC risk (P value for trend, .52). High-dose aspirin users did not have a different SCC risk than low-dose users. Regular ibuprofen users had a nonsignificant, slightly lower risk of SCC than never/occasional users (adjusted OR, 0.74; 95% CI, 0.46-1.19). Although there were no consistent effects of dose or duration, all ORs suggested a protective association with ibuprofen exposure. Among nonaspirin NSAIDs, there was also no significant SCC risk reduction (adjusted OR, 0.84; 95% CI, 0.56-1.26). For all 4 of our exposure variables, we found no evidence of a dose-response effect by duration of use. As expected, there was no association between regular use of acetaminophen (the control medication without COX activity) and SCC risk (OR, 1.15; 95% CI, 0.74-1.81) nor any association between dose and duration of acetaminophen use.
We examined which of the covariables served as potential confounding factors by virtue of being associated with both NSAID use and SCC risk at the P ≤ .20 level. Only 2 variables emerged as potential confounding variables: hair color and occupational sun exposure. We analyzed all multivariate models using only those 2 potential confounding variables (parsimonious multivariable model) and found no significant associations between NSAID use (categorized as any NSAID, aspirin, ibuprofen, nonaspirin NSAIDs) and SCC risk (data not shown). Of note, the slightly protective effect of ibuprofen use on SCC risk was not evident in the parsimonious multivariable model (adjusted OR, 1.03; 95% CI, 0.71-1.50).
Finally, we examined the association of pharmacy-dispensed NSAIDs and SCC risk. In total, 27% of cases and 26% of controls were prescribed an NSAID (not including aspirin, ibuprofen, or naproxen) and had refilled the prescription for the NSAID at least once during the 10-year observation period. There was no association between any pharmacy-dispensed NSAID use and SCC risk, whether examining overall NSAIDs or examining individual NSAID types (ever/never use). There was also no statistically significant effect of duration (Table 3).
Results from this case-control study do not support the hypothesis that NSAIDs are inversely associated with risk of cutaneous SCC. For self-reported NSAID exposure, there was no clear effect of dose, duration, or combined dose-duration variables for any of the 4 NSAID exposure categories on SCC risk. Short-term, low-dose use of any NSAID as well as acetaminophen were both associated with an increased risk of SCC, suggesting that the effect is not class specific and possibly confounded by indication.22 Furthermore, there was no association between pharmacy-dispensed NSAIDs and SCC risk during the same observation period (10 years prior to baseline).
Our results are largely consistent with 3 of the 4 published articles examining the association of NSAIDs with SCC risk.13- 16 Two large population-based cohort studies examined the association of NSAID exposure, both nonaspirin NSAIDs13 and low-dose aspirin,14 and found no association with nonmelanoma skin cancer. These studies defined nonmelanoma skin cancer using diagnosis codes and were therefore unable to separately analyze basal cell carcinomas and SCCs, which may have different sensitivities to chemoprevention with NSAIDs. In a study of 132 SCCs arising among 1093 participants of the Skin Cancer Prevention Study, a randomized trial of oral beta carotene for the prevention of NMSC, NSAID use in the year prior to diagnosis was not associated with a statistically significant risk reduction of SCC (adjusted OR, 0.71; 95% CI, 0.48-1.04).15 The analysis did not differentiate between aspirin and nonaspirin NSAIDs, although the majority of NSAID users (87%) were aspirin users. The trial was limited to people with a history of skin cancer, which may limit generalizability of the findings. Our point estimates are consistent with a published abstract, which reported that regular NSAID use for more than 24 months was associated with an increased risk of nonmelanoma skin cancer.18 The full study has not been published to date, prohibiting a more detailed comparison of their findings and ours.
Our results differ from those of a case-control study that compared 86 subjects with SCC with 187 age- and sex-matched controls drawn from 1621 sun-exposed residents of Queensland, Australia, and ascertained NSAID use over the previous 10 years with face-to-face interviews.16 The authors reported that NSAID use more than 8 times a week for more than a year was associated with a significantly reduced SCC risk (adjusted OR, 0.07; 95% CI, 0.01-0.71). Full-dose NSAID use 2 or more times per week for more than 5 years was also associated with a risk reduction for SCC (OR, 0.20; 95% CI, 0.04-0.96). Limitations of this study include its small sample size (n = 86 cases and n = 187 controls) and the fact that their exposure measurement strictly relied on recall and was also not validated against an external source, such as pharmacy records. The difference between our results and the results of this Australian study16 may also be attributed to study populations (community-based cohort in Australia vs United States), which may have different risk profiles, such as sun-exposure history. Also, the type of NSAID used by their participants differed from our sample: the most common nonaspirin NSAID reported in their sample was celecoxib (17%), whereas less than 1% of our participants reported celecoxib use. In their analysis, Butler et al16 did not distinguish among different types of NSAIDs; however, different subsets of NSAIDs may have differential chemoprotective properties.
Strengths of the present study include a large sample size (N = 830) and thorough measurement of exposure including type, dose, and duration over the preceding 10-year period. Our exposure measurement included both self-reported NSAID use, which captures over-the-counter and prescription NSAIDs, as well as detailed information on pharmacy-dispensed NSAIDs, derived from KPNC's comprehensive electronic pharmacy database. Both data sets yielded consistent findings of no association of SCC risk with NSAID exposure. After we adjusted for only those variables shown to be potential confounders (parsimonious model) as well as all ascertained SCC risk factors (fully adjusted model), our data on self-reported NSAID use showed similar findings.
There are several potential limitations to this study, including the possibility of recall bias, selection bias, and limitations of generalizability. The high response rates from both cases (81%) and controls (57%), combined with the fact that both groups largely conformed to established risk factors for SCC, suggests minimal selection bias. Our control population came from respondents to the MHS and not from the Kaiser Permanente membership at large and may be prone to selection bias. However, previously published articles have reported that respondents to the MHS are representative of the KPNC population.23 Recall bias is also a potential problem for the self-reported variables. Although our questionnaire was modeled on a previous self-administered questionnaire on supplement use (including NSAIDs) and cancer risk,24 which has shown high test-retest reliability and validity in capturing supplement intake over the past 10 years,21 the validity of using a self-administered questionnaire to retrospectively collect information on drug use may need further study. Differential misclassification would result if the cancer diagnosis served as a stimulus for cases to recall NSAID exposures more or less thoroughly than controls. However, the questionnaire collected data on a variety of exposures and NSAIDs are not recognized to be an important factor in the risk of skin cancer. The generalizability of our study may be limited because we only studied KPNC members, although previous studies have shown that the KPNC membership is highly representative of the surrounding region except for the tail ends (lowest and highest) of the income distribution.23,25
In this case-control study, we did not detect any consistent relationships between SCC risk and overall NSAID, aspirin, ibuprofen, or nonaspirin NSAID use. Dose and duration of NSAID use did not appear to alter risk of SCC. Given the potential toxic effects of NSAIDs, including platelet dysfunction and gastric ulcers, more uniformly efficacious chemopreventive agents with safer adverse effect profiles need to be explored.
Correspondence: Maryam M. Asgari, MD, MPH, Division of Research, Kaiser Permanente Northern California, 2000 Broadway, Oakland, CA 94612 (email@example.com).
Accepted for Publication: September 19, 2009.
Published Online: February 15, 2010 (doi:10.1001/archdermatol.2009.374).
Author Contributions: Dr Asgari had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Asgari, Friedman, and White. Acquisition of data: Asgari, Warton, and White. Analysis and interpretation of data: Asgari, Chren, Warton, Friedman, and White. Drafting of the manuscript: Asgari and Warton. Critical revision of the manuscript for important intellectual content: Asgari, Chren, Warton, Friedman, and White. Statistical analysis: Asgari, Warton, and White. Obtained funding: Asgari. Administrative, technical, and material support: Asgari, Warton, and White. Study supervision: Chren, Friedman, and White.
Financial Disclosure: Dr Friedman has served on an advisory committee to Roche Laboratories in June 2008, and in the past 3 years has consulted for law firms serving both plaintiffs and Ortho-McNeil-Janssen Pharmaceuticals regarding litigation concerning celecoxib and Ortho-Evra, respectively. These consultancies had no relationship with or influence on the present study.
Funding/Support: This study was supported by grants from the National Institute of Arthritis Musculoskeletal and Skin Diseases (K23 AR 051037 [Dr Asgari] and K24 AR 052667 [Dr Chren]) and from the National Cancer Institute (R01 CA 098838 [Dr Friedman]).
Role of the Sponsors: The sponsors had no role in the design and conduct of the study; in the collection, analysis, and interpretation of data; or in the preparation, review, or approval of the manuscript.