Effect of NSAIDs on the recurrence of nonmelanoma skin cancer
Experimental studies have consistently shown a protective effect of nonsteroidal antiinflammatory drugs (NSAIDs) against nonmelanoma skin cancers (NMSC). However, little human epidemiological research has been done in this regard. We used data from the Skin Cancer Chemoprevention Study to explore the association of NSAID use and with the risk of basal-cell carcinoma (BCC) and squamous-cell carcinoma (SCC). 1,805 subjects with a recent history of NMSC were randomized to placebo or 50 mg of daily β-carotene. Participants were asked about their use of over-the-counter and prescription medications at baseline and every 4 months during the trial. Skin follow-up examinations were scheduled annually with a study dermatologist; confirmed lesions were the endpoints in the study. We used a risk set approach to the analysis of grouped times survival data and unconditional logistic regression to compute odds ratios [ORs] for various exposures to NSAIDs. The use of NSAIDs was reported in over 50% of questionnaires. For BCC, NSAIDs exhibited a weak protective effect in crude analyses, which attenuated markedly after adjustment. For SCC, the use of NSAIDs in the year previous to diagnosis reduced the odds by almost 30% (adjusted OR= 0.71, 95% CI 0.48–1.04). When we accounted for frequency of use, results for BCC were not striking, and there were inconsistent suggestions of an inverse association with SCC. There were some indications of a modest, nonsignificant reduction on the number of BCCs and SCCs with NSAID use. Our data suggest a weak and inconsistent chemopreventive effect of NSAIDs on BCC and SCC. © 2006 Wiley-Liss, Inc.
The chemopreventive effect of nonsteroidal antiinflammatory drugs (NSAIDs), particularly aspirin, on nonmelanoma skin cancer (NMSC) has been repeatedly shown in animal1, 2, 3, 4, 5 and in in vitro studies.6, 7 In humans, some experimental studies have reported that topical or oral NSAIDs may lead to regression of skin neoplasms.8, 9, 10
Findings from both observational studies11, 12, 13, 14, 15, 16, 17 and clinical trials18, 19, 20 are consistent with a protective effect of NSAIDs against colorectal neoplasia, but few epidemiological studies have investigated the preventive effects of NSAIDs on risk of nonmelanoma skin cancer,21, 22 the most commonly occurring cancer in the Caucasian population.23, 24 However, results from 1 single-arm, unblinded study and from a randomized trial of topical diclofenac for treatment of actinic keratoses, a well-known precursor of SCC, have been encouraging.25, 26 In the present analysis, we used data from a skin cancer chemoprevention trial,27 to explore the relationship between NSAID use and the occurrence of nonmelanoma skin cancer, both basal-cell carcinomas (BCC) and squamous-cell carcinomas (SCC).
Material and methods
The Skin Cancer Prevention Study was a randomized, double blind trial of oral β-carotene for the prevention of nonmelanoma skin cancer in patients with a recent history of these tumors. Details of the study design and methods have been described elsewhere.27, 28 The study involved 4 clinical centers: Dartmouth-Hitchcock Medical Center in Hanover, NH (also the coordinating center), the University of California at Los Angeles School of Medicine, the University of California Medical School in San Francisco and the University of Minnesota Schools of Medicine and Public Health, Minneapolis. Enrollment of patients began in February 1983. Eligible subjects had at least 1 histologically confirmed basal-cell or squamous-cell carcinoma after January 1, 1980. Eligibility was assessed through review of medical records and dermopathology reports. Of the 5,232 potentially eligible patients, 1,968 entered a 1-month placebo run in period and 1,805 subjects were randomized to receive either 50 mg of β-carotene (BASF, Wyandotte, Mich.) daily or placebo.
At baseline, subjects completed a questionnaire regarding demographic characteristics, sun exposure history, previous skin cancers, health behaviors, current use of over-the-counter and prescription medications and usual frequency of vegetable consumption. A study dermatologist assessed the skin type of each subject, summarizing its reactivity to sun exposure (erythema versus tanning), presence of solar damage and number and cell types of known previous skin cancers.
Every 4 months, participants completed interval questionnaires that included items regarding recent medical history and skin examinations and diagnoses since the previous questionnaire. Subjects were also asked what over-the-counter products and prescription medications had been used, although no information on doses or frequency of use was obtained. Skin follow-up examinations were scheduled annually with a study dermatologist; biopsies were performed on all lesions that were suspicious for cancer. Skin lesions removed in interim examinations were also tracked. All lesions were examined independently by a local pathologist as well as by the study dermatopathologist. If the 2 readings disagreed with regard to the presence or absence of nonmelanoma skin cancer, a third reading by a consultant dermatopathologist was used for a final diagnosis. The primary endpoint of the study was the occurrence of incident NMSC.
During the trial, 3,975 skin lesions in 1,093 participants (61%) were identified as possible cancers and removed. We excluded from the analysis lesions not confirmed microscopically (n = 49), those that were recurrence of a previous cancer (n = 286), malignant melanomas (n = 3) and nonneoplastic lesions (n = 1,685). Thus, 1,952 microscopically confirmed new skin cancers from 702 of the 1,805 randomized subjects (39%) were included in the present analysis (1,747 BCCs, 204 SCCs and 1 baso-squamous carcinoma). Given the high level of agreement between local and study dermatopathologists (96%), we included the local pathologist's diagnoses when the slides were not available for coordinating center review (n = 52). Of the 702 patients with confirmed cancers, 570 had only BCCs, 51 had only SCCs and 81 had both.
Every 6 months, an external safety and data-monitoring committee reviewed study data. This committee considered that we had sufficient data to evaluate the main study hypothesis in the summer of 1989, and participants were instructed to stop taking study pills on September 30, 1989.
To evaluate the association of NSAID use with the occurrence of NMSC, properly accounting for the time-dependent nature of NSAID use, we used a risk set approach to the analysis of grouped time survival data.29 Each 12-month period after randomization (except the first) was considered as 1 study year. The first study year was 15 months long to cover the time permitted in our protocol for the first annual follow-up appointment. Skin cancers were attributed to the study year of diagnosis. When the trial ended, 141 patients were in the first half of their fourth study year and therefore were censored at the end of the third year; 549 patients were in the first half of the fifth year and censored at the end of the fourth. In this analysis, we do not consider events that occurred after the end of a patient's fifth study year.
Of the 1,805 subjects randomized in the study, 84 never had an annual follow-up evaluation. Of the remaining 1,721 subjects, 1,019 remained cancer-free during the study and 702 had at least 1 NMSC. Randomization to β-carotene supplementation did not confer a significant reduction of risk for NMSC.27 In the present analysis, we defined risk sets for each study year as the group of all subjects who underwent that year's follow up dermatological exam without a prior incident NMSC since randomization. In each risk set, cases were subjects diagnosed with a NMSC at that year's annual examination or during an interim exam during that year, and controls were those who remained cancer-free up to that point in the study. A subject could be classified as a control in 1 risk set and subsequently become a case in a later one. To account for each type of NMSC, we applied the same approach to SCC and BCC separately.
In the overwhelming majority of instances, all 3 interval questionnaires for a study year were completed before that year's study exam [in 96.4 % of subject study years]. Overall, 1,256 subjects (70%) reported use of NSAIDs in at least 1 interval questionnaire during the study. Of these, around 87% only took medications containing aspirin (alone or in combination with other NSAIDs) and 13% exclusively used nonaspirin NSAIDs. Exposure was defined in 2 ways: as a binary variable indicating the report of NSAID use (including aspirin) in any questionnaires for a given study period, and as an ordered categorical variable that accounted for duration of exposure. In the latter case, we computed the proportion of completed questionnaires during the study period indicating NSAID use and defined a categorical variable as: none, sporadic use (less or half of questionnaires positive for NSAID use) and frequent use (more than half of the questionnaires positive).
To account for possible latencies of NSAID effect, we considered different exposure periods: reported use of NSAIDs at baseline, exposure during the same study year as the exam, exposure during the previous year and exposure during the same and the previous study years. To overcome possible response bias (as subjects who completed interval questionnaires could be different from those who do not), we repeated the analyses including only subjects who had completed the 3 questionnaires submitted during a given year. Subjects without interval questionnaires in a study period had missing information for the exposure in the corresponding risk set.
As appropriate for grouped time risk set data, unconditional logistic regression was applied to compute odds ratios (ORs) for various exposures to NSAIDs; covariates in the full models were age, sex, center, skin type, number of nonmelanoma skin cancers prior to study entry and number of completed questionnaires. Given the null effects of the randomized treatment assignment, this variable was not included in the final analyses.
We also accounted for multiple NMSC using overdispersed Poisson regression models. To ensure that NSAID use preceded the skin cancer diagnosis, we first considered the association of baseline use with the total number of NMSCs diagnosed during the trial. Second, we used the first 2 study years to classify NSAID use, and took as endpoints the number of diagnosed NMSC in the 3 later years. We also repeated these analyses independently for SCC and BCC.
Baseline characteristics of cases and controls are summarized in Table I. Cases were older and more likely to be male than controls. As expected, cases were more likely to have high/moderate risk skin type and a higher number of previous skin cancers.
Table I. Demographic and Clinical Characteristics of Study Participants
|Age (mean ± SD)||62.1 (10.3)||64.3 (9.1)|
|Males (%)||638 (62.6)||548 (78.1)|
|Study center (%)|
| Dartmouth||270 (26.5)||151 (21.5)|
| UCLA||136 (13.4)||177 (25.2)|
| UCSF||148 (14.5)||142 (20.2)|
| Minnesota||465 (45.6)||232 (33.1)|
|Skin type1 (%)|
| Burns||411 (40.3)||376 (53.6)|
| Moderate||406 (39.8)||252 (36.0)|
| Tans||202 (19.8)||73 (10.4)|
|Number of previous skin cancers2 (%)|
| 1||623 (61.6)||188 (26.8)|
| 2||201 (19.9)||143 (20.4)|
| 3||78 (7.7)||87 (12.4)|
| 4–5||68 (6.7)||122 (17.4)|
| 6–9||27 (2.7)||80 (11.4)|
| >10||14 (1.4)||81 (11.6)|
Compliance with interval questionnaires and use of medications containing aspirin is described in Table II. About 98% of participants completed at least 1 questionnaire in any given study period, and about 85% completed all 3 questionnaires. The response rate in the last study period was somewhat lower, reflecting the fact that the study ended prematurely and some information was censored as of the end of the previous year. Among subjects with at least 1 questionnaire, about 50% reported use of NSAIDs in each risk set. Among those with all questionnaires completed, more than a fourth indicated a regular use of aspirin, i.e., use in all 3 questionnaires.
Table II. Compliance with Questionnaires and NSAID Use by Risk Set Time
|1||1640||1621 (98.8)||1508 (92.0)||738 (45.5)||312 (20.7)|
|2||1310||1284 (98.0)||1153 (88.0)||643 (50.1)||285 (24.7)|
|3||1120||1105 (98.7)||958 (85.5)||600 (54.3)||272 (28.4)|
|4||912||899 (98.6)||784 (86.0)||516 (57.4)||247 (31.5)|
|5||500||487 (97.4)||375 (75.0)||214 (43.9)||113 (30.1)|
Table III shows the ORs for the association of incident BCC and SCC with NSAID use at different times in relation to diagnosis. For BCC, the crude estimates suggested a significant protective trend for baseline use (crude OR = 0.79, 95% CI: 0.65–0.95) and a borderline significant trend for use in the same and the previous year as the diagnosis (crude OR = 0.85, 95% CI: 0.70–1.03). However, these trends were attenuated after adjustment (adjusted OR for baseline use 0.89, 95% CI: 0.70–1.09 and for use in the same and previous year 0.91, 95% CI: 0.74–1.12). These results did not change when the analysis was restricted to subjects who were fully compliant with the questionnaires (data not shown).
Table III. Association of Risk of Nonmelanoma Skin Cancer with NSAID Use at Various Times Before Diagnosis
|Baseline||1488/4936||164/651||0.79 (0.65–0.95)||0.89 (0.73–1.09)||1889/6597||36/132||0.93 (0.63–1.37)||0.95 (0.64–1.42)|
|Same year||2524/4868||297/629||0.88 (0.75–1.04)||0.92 (0.77–1.10)||3307/6475||60/126||0.90 (0.63–1.28)||0.94 (0.66–1.36)|
|Previous year||2196/4873||244/639||0.88 (0.74–1.05)||0.96 (0.80–1.15)||2904/6487||43/128||0.67 (0.46–0.98)||0.71 (0.48–1.04)|
|Same and previous years4||1846/4833||198/624||0.85 (0.70–1.03)||0.91 (0.74–1.12)||2412/6420||40/123||0.76 (0.51–1.14)||0.79 (0.52–1.21)|
Associations with SCC tended to be somewhat stronger. There was an inverse association of risk with NSAID use during the previous year (adjusted OR= 0.71, 95% CI: 0.48–1.04) and during the same and previous year (adjusted OR = 0.79, 95% CI: 0.52–1.21). The findings for all BCC were very similar to those for all NMSC (data not shown).
We accounted for intensity of exposure by examining the proportion of completed questionnaires that indicated NSAID use during a given study interval (Table IV). In this analysis, we did not include baseline use, since it was a one-time report in the initial questionnaire. For BCC, the ORs for frequent use of NSAIDs did not suggest significant associations in any of the exposure periods considered (Table IV). For SCC, there was an inverse association of sporadic use with risk. For example, the adjusted OR for NSAID use in the same or the previous year was 0.45 (95% CI: 0.28–0.87). However, ORs for frequent use were much closer to the null (Table IV). Again, results for all BCC were very similar to those for all NMSC (data not shown).
Table IV. Association of Nonmelanoma Skin Cancer Risk with Duration of NSAID Use During Various Periods Before Diagnosis
| None||2344||332||1.00 (Reference)||1.00 (Reference)||3618||66||1.00 (Reference)||1.00 (Reference)|
| Sporadic||650||79||0.87 (0.67–1.13)||0.86 (0.65–1.13)||846||9||0.51 (0.25–1.04)||0.57 (0.28–1.16)|
| Frequent||1874||218||0.89 (0.74–1.07)||0.94 (0.78–1.15)||2461||51||1.04 (0.72–1.51)||1.07 (0.73–1.57)|
| None||1718||183||1.00 (Reference)||1.00 (Reference)||2445||52||1.00 (Reference)||1.00 (Reference)|
| Sporadic||481||63||1.25 (0.92–1.69)||1.29 (0.93–1.77)||673||6||0.43 (0.18–1.00)||0.48 (0.20–1.12)|
| Frequent||1309||118||0.88 (0.69–1.12)||0.96 (0.74–1.24)||1770||29||0.80 (0.50–1.26)||0.85 (0.53–1.36)|
| None||1347||149||1.00 (Reference)||1.00 (Reference)||1927||45||1.00 (Reference)||1.00 (Reference)|
| Sporadic||917||96||0.98 (0.75–1.29)||1.05 (0.79–1.39)||1285||12||0.42 (0.22–0.80)||0.45 (0.28–0.87)|
| Frequent||1279||124||0.91 (0.70–1.17)||1.01 (0.77–1.32)||1731||33||0.84 (0.53–1.33)||0.91 (0.57–1.45)|
To evaluate the effect of NSAIDs on the number of recurrences, we used 2 approaches. First, we studied the relationship between NSAID use reported at the intake questionnaire and the number of cancers diagnosed during the trial (Table V). For BCC, baseline NSAID use conferred a significant reduction of risk in the crude analysis (unadjusted ratio of the average number of tumors 0.86, 95% CI: 0.66–0.95). Adjustment lowered this point estimate, but statistical significance was lost (adjusted ratio = 0.78, 95% CI: 0.66–1.12). We obtained similar estimates for all NMSC. Results for SCC were also not striking.
Table V. Association Of NSAID Use on The Average Number of Nonmelanoma Skin Cancers Diagnosed
|Association of NSAID use at baseline with NMSC diagnosed during the study|
|None||1.01||1.00 (Reference)||1.00 (Reference)||0.11||1.00 (Reference)||1.00 (Reference)|
|Any||0.86||0.86 (0.66–0.95)||0.78 (0.66–1.12)||0.09||0.81 (0.49–1.34)||0.89 (0.57–1.40)|
|Association of NSAID use during the first 2 years in the study with the number of NMSC diagnosed during the last 3 years of follow up3|
|None4||0.62||1.00 (Reference)||1.00 (Reference)||0.07||1.00 (Reference)||1.00 (Reference)|
|Any||0.44||0.69 (0.52–0.93)||0.91 (0.74–1.13)||0.05||0.68 (0.41–1.13)||0.80 (0.50–1.29)|
| Sporadic4||0.40||0.60 (0.40–0.88)||0.78 (0.59–1.03)||0.06||0.77 (0.41–1.44)||0.94 (0.53–1.67)|
| Frequent4||0.48||0.76 (0.54–1.08)||1.02 (0.80–1.30)||0.04||0.61 (0.32–1.15)||0.69 (0.38–1.25)|
In addition, we explored the association between NSAID use during the first 2 years of the study and the average number of NMSC diagnosed in the subsequent 3 years. Again, there were suggestions of a reduction in the number of BCCs among NSAID users in the crude models, but not after adjustment (adjusted ratio = 0.91, 95% CI: 0.74–1.13). When we took into account the frequency of use, the rate was nonsignificantly lower for sporadic users (adjusted ratio = 0.78, 95% CI: 0.59–1.03), but not for frequent users. Results for all NMSC were virtually the same. Overall, rates for SCC were lower for NSAID users than for nonusers, but this was compatible with a chance association in all analyses (Table V).
In this closely monitored cohort of high risk subjects, there were only inconsistent, weak suggestions of an inverse association of use of aspirin and other NSAIDs with the incidence of NMSC in years following use. At most, our data suggest a weak chemopreventive effect of NSAID use on SCC in the year prior to diagnosis, and on the number of BCCs and SCCs.
The chemopreventive attributes of NSAIDs against carcinogenesis have been widely recognized.30, 31, 32 In relation to the skin, the effect of NSAIDs, particularly aspirin, on carcinogenesis has been investigated in a few in vitro studies7, 33 and in many animal studies; the majority have reported an anticarcinogenic effect of NSAIDs.1, 2, 3, 4, 5, 6, 33, 34, 35, 36 Uncontrolled human studies have reported a decrease in the number of new skin tumors or regression of existing ones with the use of topical or oral NSAIDs, supporting the antineoplastic properties of these drugs.8, 9, 10, 37
In spite of this, little human epidemiological research has been done to explore this association. Two population-based studies21, 22 that used administrative databases assessed the risk of various cancers, including NMSC, in relation to use of NSAIDs, with null results. However, a randomized trial has provided good evidence of the therapeutic properties of NSAIDs, showing the efficacy of topical diclofenac against actinic keratoses, proliferative lesions that are thought to be precursors of SCCs.25 Another single-arm, open-label study of topical diclofenac and actinic keratosis reached similar positive conclusions.26
The best-known mechanism of action of NSAIDs is the inhibition of COX-2, the inducible isoform of the cyclooxygenase enzyme, implicated in inflammation and promotion of neoplastic tumors.30, 31, 38, 39, 40 Overexpression of COX-2 seems to be a factor in almost all types of neoplasia, and in various stages of carcinogenesis.3, 30, 31, 32, 41, 42 In the skin, experimental studies have shown that acute UVB exposure can cause COX-2 overexpression in murine and human skin and high levels of the enzyme are present in neoplastic lesions following UV irradiation.33, 39, 40, 41 Furthermore, many in vitro and animal studies provide strong evidence for a role of COX-2 as an endogenous promoter of skin neoplasia.3, 34, 41, 43, 44, 45 However, our data only show inconsistent suggestions of effects of NSAIDs on BCC and SCC.
One potential explanation for our mixed findings could be that the NSAID use we observed was too short to have affected the occurrence of NMSC. For colorectal cancer, for example, 10–20 years of consistent use may be required for a protective effect.30 Unfortunately, our data do not permit us to assess the association of risk with the duration of use. Also, we observed a limited number of SCCs, and had only limited statistical power to detect significant associations for this endpoint. Another plausible reason could be inaccuracies in the assessment of NSAID use, since our questionnaires did not collect information about dose or frequency of use and this measurement error could have biased our findings. Moreover, patients were not asked specifically about use of aspirin or other NSAIDs, possibly leading to underreporting. If this affected cases and controls equally, the result would be a conservative bias in our estimates. Finally, confounding seems to be an important factor in the analysis, since our estimates were often closer to the null and in the full multivariate models.
Nonetheless, our analysis has several strengths. To our knowledge, our study is the first to address the association between use of aspirin and other NSAIDs and the risk of NMSC in a closely monitored cohort of high risk patients. We used data from a carefully controlled cohort of patients that were part of a randomized clinical trial. Study dermatologists performed annual exams and removed all suspicious lesions. Biopsy specimens were read locally and uniformly by the coordinating center pathologist and the level of agreement was 96%. We used several statistical approaches to analyze the data, exploring different exposure periods, and various case–control definitions within the cohort design. We also explored the possibility of separate associations for BCC and SCC, acknowledging the many indications that BCC and SCC are biologically distinct: essential differences in histogenesis,46, 47, 48, 49 molecular pathways50, 51 and vascularization patterns.52, 53 The major limitation of our study is the lack of data regarding the doses and duration of NSAID use in our subjects.
Thus, despite the role of COX-2 in skin carcinogenesis, our study provides only weak suggestions of chemopreventive effect of NSAIDs on BCC and SCC. Of course, these results must be interpreted in the context of the limitations of our data. At this point, more observational and randomized studies would be very valuable, especially studies that can assess more closely the duration of NSAID use.
We are indebted to the subjects of the Beta Carotene Skin Cancer Trial, as well as to Nicholas Lowe, J. Corwin Vance, Peter Elias and Steven K. Spencer, all of whose enthusiasm and cooperation were essential for the successful completion of the study.