Nonsteroidal anti-inflammatory drugs and risk of lung cancer
Article first published online: 4 JAN 2007
Copyright © 2006 Wiley-Liss, Inc.
International Journal of Cancer
Volume 120, Issue 7, pages 1565–1572, 1 April 2007
How to Cite
Hernández-Díaz, S. and García Rodríguez, L. A. (2007), Nonsteroidal anti-inflammatory drugs and risk of lung cancer. Int. J. Cancer, 120: 1565–1572. doi: 10.1002/ijc.22514
- Issue published online: 30 JAN 2007
- Article first published online: 4 JAN 2007
- Manuscript Accepted: 14 NOV 2006
- Manuscript Received: 28 JUL 2006
- lung cancer;
- nonsteroidal anti-inflammatory drugs;
Regular aspirin and non-aspirin nonsteroidal anti-inflammatory drug (NSAID) use is associated with a reduced risk of colorectal cancer. The effect of NSAIDs on the risk of other cancers remains unclear. To evaluate whether use of aspirin or other specific NSAIDs protects against lung cancer, we conducted a case–control study nested in a cohort of subjects 40–84 years old in 1995–2004, without a diagnosis of cancer before the study start date, and with at least 2 years of enrollment with a general practitioner providing data to the The Health Improvement Network (THIN) database in the UK. Patients who had a first diagnosis of primary lung cancer during the study period were considered cases. A random sample of 10,000 controls was frequency-matched to the cases for age, sex and calendar year. The index date for exposure definition was 1 year before the date of diagnosis for cases and 1 year before a random date within the study period for controls. Relative risks and 95% confidence intervals were estimated using conditional logistic regression stratified for matching factors. Factors such as smoking, chronic obstructive pulmonary disease, cardiovascular diseases and body mass index were introduced in the model. We identified 4,336 cases with primary incident lung cancer (incidence rate 7.6 per 10,000 person-years). Compared with subjects with no prescription of non-aspirin NSAID prior to the index date, the risk of lung cancer was 0.76 (0.61–0.94) among those who received a prescription the previous year and had a treatment duration of at least 1 year. The corresponding relative risk was 1.15 (0.99–1.34) for aspirin. In conclusion, prescription of non-aspirin NSAIDs for at least 1 year might be associated with a slightly reduced risk of lung cancer. Aspirin was not associated with a risk reduction, perhaps due to residual confounding. © 2006 Wiley-Liss, Inc.
Regular use of aspirin or other nonsteroidal anti-inflammatory drugs (NSAIDs) reduces the risk of colorectal cancer, and might reduce the risk of additional gastrointestinal cancers as well.1 The effect of NSAIDs on the development and progression of other types of cancer remains controversial.2, 3
The potential role of NSAIDs in the prevention of lung cancer is particularly relevant, given that lung cancer continues to be the leading cause of cancer mortality, accounting for over 160,000 deaths annually in the United States alone.4 Several lines of evidence suggest that NSAIDs may influence lung tumorigenesis through inhibition of cylooxygenase (COX) enzymes. In vitro studies have demonstrated an overexpression of COX-2 in lung cancer tissues,5, 6, 7 and NSAIDs have been shown to reduce COX-2 levels in lung cancer cell lines.8 Animal studies have shown a protective effect of these drugs against experimentally induced lung cancer.9, 10 Results from 3 randomized clinical trials, whose main objective was not the primary prevention of lung cancer, are compatible with a beneficial effect of aspirin on lung cancer mortality.11, 12 Results from observational studies on the association between lung cancer and NSAIDs, however, have been inconsistent.13
Previous epidemiological evidence was often limited by the relatively small number of lung cancer cases included and the potential bias from misclassification of the etiologically relevant exposure or residual confounding by underlying clinical conditions and smoking. In addition, limited information is available on the effect of specific NSAIDs or the role of dose, duration and latency of treatment. Therefore, we used data from The Health Improvement Network (THIN) database to evaluate the association between prospectively collected information on the frequency and duration of use of specific NSAIDs and the risk of primary lung cancer. The current study was specifically designed to assess these associations and it is the largest conducted on this topic so far.
Material and methods
We conducted a population-based cohort study with nested case–control analysis using data from the THIN database in the United Kingdom.14 Briefly, participating general practitioners systematically and prospectively retrieve and record on their computer the clinical information on around 3.9 million patients, including demographics, diagnoses from general practitioner's visits, specialist's referrals and hospital admissions, results of laboratory tests and comments in a free-text section. All prescriptions issued by the general practitioner, including dose and indication, are recorded and directly generated from the computer. Noteworthy, aspirin and ibuprofen were available both with and without prescription from the UK pharmacies during the study period, and this study ascertained only prescriptions for these drugs. Data are sent anonymously to THIN. THIN collects and organizes the information in order to be used for research projects. The READ classification is used to code specific diagnoses, and a drug dictionary based on data from the MULTILEX classification is used to code drugs.15, 16 The current study was approved by an ethics review board (MREC).
The study cohort included patients 40–84 years old between January 1995 and December 2004, with at least 2 years of enrollment with the general practitioner and at least 1 year since the first computerized prescription. Patients with any cancer before start date were excluded from the cohort (N = 49,237). Persons 70 years and older at start date with a follow-up greater than 1 year and no recording of any data during their follow-up time were also excluded (N = 5,205). This was done to exclude people whose data completeness is most likely seriously deficient. Members of this cohort were followed until the date of one of the following endpoints: lung cancer detection, exclusion criteria (other cancer), 85th birthday, death or end of study period, which ever came first. The final cohort consisted of 1,078,299 persons followed for an average of 5.4 years.
We identified patients with a first diagnosis of primary lung cancer during the study period (N = 4,419) and reviewed their de-identified computerized patient profiles, which included demographic and clinical data but not drug exposure information. The methods of diagnosis reflected the general clinical practice, i.e., most practitioners had referred the patient for an X-ray after clinical suspicion, and most patients had pathology reports, surgery or autopsy confirmation. We excluded 81 cases whose diagnosis was uncertain based on the available information (e.g., could be just a differential diagnosis). None of the patients had a clinical diagnosis of lung cancer in the database prior to the date assigned by the computerized algorithm. However, many of them had recorded suspicious symptoms (e.g., persistent cough) or procedures (e.g., chest X-rays) during the preceding months; the average time between potential initial symptoms and clinical diagnosis was 66 days, and none of them had warning symptoms recorded more than 365 days before diagnosis. Since it is impossible to assign with precision when the first symptoms had started for each patient, we incorporated a conservative lag time of 1 year between the presumably etiologically meaningful exposure date, refer to as index date, and the date of diagnosis (see exposure definition later). Patients with a diagnosis of lung cancer secondary to another cancer had been excluded from the study population (i.e., we excluded patients with other types of cancer). However, for 14 cases, it was uncertain whether the cancer was primary lung cancer with metastases or secondary metastasis in the lung from another cancer. Therefore, we classified these cases as “potentially” secondary and conducted subanalysis excluding them; results were identical.
For a random sample of 80 cases, we sent a brief questionnaire to their general practitioners asking them to obtain from their patients' written medical records all information related to the diagnosis of lung cancer. The questionnaire included specific questions on the histological type of the tumor and first date of diagnosis. Out of the 79 responses received, 75 confirmed the diagnosis of primary lung cancer; 17 were squamous cell carcinoma, 13 small cell carcinoma, 12 adenocarcinoma, 13 other type of lung cancer and 20 had insufficient histopathological information. The date when the diagnosis was first made according to the practitioner was, on average, 11 days prior to the date we had extracted from the computerized records. The 4 nonconfirmed cases were excluded from the study (2 of them had already been excluded in the systematic review of computerized profiles). Since the confirmation rate was close to 95%, we did not request original records for the remaining cases. The total number of cases considered for the analysis was 4,336.
Controls were randomly sampled from the same cohort. A date encompassed within the study period was generated at random for each of the members of the cohort. If the random date for a study member was included in his/her eligible person-time (follow-up period), we marked that person-day as an eligible control. The same exclusion criteria were applied to controls as to cases. A person selected to be a control was eligible to be a case later on. A group of 10,000 controls were randomly selected from the list of eligible person-days and frequency matched to the cases on sex, age within 1 year, and calendar year.
Early cancer symptoms in the subclinical phase (latent period), or after clinical suspicion but before the final diagnosis of lung cancer, might affect the prescription of NSAIDs. Therefore, we incorporated a lag time of 1 year into our definition of index date for exposure status assessment by discounting 12 months from the date of diagnosis for cases and from the random date for controls. We defined 3 mutually exclusive time windows of exposure: Recent use refers to prescriptions that lasted until the index date or ended in the year prior to the index date based on the length of drug therapy as prescribed by the general practitioner. Past use refers to prescriptions ending more than 1 year before the index date. Nonuse was defined as no recorded prescription at any time before the index date. Recent users were further subdivided into 2 categories according to duration of therapy: less and more than 1 year of use. The effect of dose was assessed in long-term (duration greater than 1 year) recent users. Also within long-term recent users, in addition to aspirin and non-aspirin (NA) NSAIDs, we studied separately the most widely prescribed individual NA-NSAIDs such as ibuprofen, diclofenac and naproxen.
We estimated the relative risk and 95% confidence intervals for lung cancer associated with prescription of NSAIDs compared with no prescription at any time before the index date by means of conditional logistic regression models stratified for age, sex and calendar year. Other potential risk factors such as smoking status, history of chronic obstructive pulmonary disease (COPD), body mass index (weight in kilograms divided by height in meters squared), prescription of other drugs and alcohol use were ascertained from the database and introduced into the multivariate models. Given the importance of smoking on the occurrence of lung cancer, we conducted an alternative analysis using only patients with smoking status recorded in the database. Smoking classification was limited to never, recent and former, as a more detailed life-time smoking history was unavailable for most subjects. However, as a surrogate for smoking severity, we used information on smoking cessation interventions including general practitioner's advice and specific treatments prescribed to facilitate smoking cessation. Specific analyses were performed for men and women separately, as well as for smokers and nonsmokers.
We identified 4,336 patients with a first diagnosis of lung cancer, for an overall incidence rate of 7.6 per 10,000 person-years. Table I presents selected characteristics of cases and controls. Due to the matching, cases and controls had a similar distribution of age and gender. Compared with nonsmokers, recent or ex-smokers presented a greatly increased risk of lung cancer. Within controls, the prevalence of aspirin use was 17% among recent smokers, 26% among ex-smokers and 17% among nonsmokers, while the prevalence of NA-NSAIDs was 22–23% across smoking categories.
|Characteristic||Cases||Controls||Adjusted relative risk1|
|40–59||763 (17.6)2||1904 (19.0)2||NA3|
|60–69||1275 (29.4)||2901 (29.0)|
|70–79||1885 (42.8)||4177 (41.8)|
|80–85||443 (10.2)||1018 (10.2)|
|Male||2658 (61.3)||6053 (60.5)||NA3|
|Female||1678 (38.7)||3947 (39.5)|
|<20||320 (7.4)||337 (3.4)||1.44 (1.19–1.75)4|
|20–24||1304 (30.1)||2734 (27.4)||Reference|
|25–29||1102 (25.4)||3154 (31.5)||0.82 (0.74–0.91)|
|>29||453 (10.5)||1219 (12.2)||0.86 (0.74–0.99)|
|UNK||1157 (26.7)||2556 (25.6)||1.30 (1.13–1.48)|
|Never||764 (17.6)||5092 (50.9)||Reference|
|Recent||2200 (50.7)||1742 (17.4)||7.47 (6.72–8.29)|
|None||1752 (40.4)||1555 (15.6)||6.84 (6.13–7.62)|
|Advice only||156 (3.6)||74 (0.74)||16.64 (12.19–22.73)|
|Recent treatment||150 (3.5)||41 (0.41)||22.71 (15.70–32.86)|
|Past treatment||142 (3.3)||72 (0.72)||10.94 (8.00–14.96)|
|Former||848 (19.6)||1503 (15.0)||3.24 (2.87–3.65)|
|Unknown||524 (12.1)||1663 (16.6)||1.85 (1.55–2.20)|
|Alcohol (units per weeks)|
|None||1452 (33.5)||3372 (33.7)||Reference|
|1–9||1027 (23.7)||2529 (25.3)||1.09 (0.98–1.21)|
|10–19||429 (9.9)||918 (9.2)||1.12 (0.96–1.30)|
|≥20||448 (10.3)||740 (7.4)||1.24 (1.06–1.45)|
|Unknown||980 (22.6)||2441 (24.4)||1.03 (0.88–1.20)|
|COPD5||848 (19.6)||507 (5.1)||2.62 (2.26–3.03)|
|Asthma5||703 (16.2)||1031 (10.3)||0.98 (0.85–1.13)|
|Recent use||455 (10.5)||497 (5.0)||1.22 (1.03–1.45)|
|Past use||400 (9.2)||609 (6.1)||1.04 (0.88–1.23)|
|Nonuse||3481 (88.9)||8894 (88.9)||Reference|
|Hypertension5||1100 (25.4)||2966 (29.7)||0.81 (0.74–0.90)|
|IHD5||873 (20.1)||1647 (16.5)||1.09 (0.98–1.21)|
|CVD5||457 (10.5)||757 (7.6)||1.23 (1.07–1.42)|
|Diabetes5||331 (7.6)||744 (7.4)||1.01 (0.87–1.18)|
|Osteoarthritis5||1295 (29.9)||2874 (28.7)||0.83 (0.75–0.91)|
|Rheumatoid arthritis5||100 (2.3)||209 (2.1)||0.87 (0.66–1.14)|
|Recent use||1557 (35.9)||2593 (25.9)||1.40 (1.26–1.55)|
|Past use||1068 (24.6)||2255 (22.6)||1.16 (1.04–1.29)|
|Nonuse||1711 (39.5)||5152 (51.5)||Reference|
|Medical visits (no.)|
|0–10||725 (16.7)||2374 (23.7)||Reference|
|11–30||1477 (33.4)||3592 (35.9)||1.21 (1.07–1.36)|
|31–50||881 (20.3)||1803 (18.0)||1.33 (1.14–1.54)|
|>50||1283 (29.6)||2231 (22.3)||1.35 (1.15–1.60)|
|0||1155 (26.6)||3231 (32.3)||Reference|
|1–4||1814 (41.8)||4266 (42.7)||0.99 (0.89–1.10)|
|>4||1367 (31.5)||2503 (25.0)||1.08 (0.95–1.24)|
|Hospitalizations (≥1)5||1281 (29.5)||2321 (23.2)||1.08 (0.97–1.19)|
As shown in Table II, 22.1% cases and 17.3% of controls had been prescribed aspirin during the year preceding the index date (i.e., within the period of 13–24 months before the date of diagnosis for cases or the random date for controls). Compared with nonuse, the RR for lung cancer associated with aspirin use in the year preceding the index date was 1.12 (95% CI 0.99–1.28). For patients who had used aspirin for at least 1 year the RR was 1.15 (0.99–1.34). Higher doses were not associated with a lower RR (of note, very few subjects used more than 300 mg/day). For patients who had stopped using aspirin the risk was similar to that in nonusers after 1 year (RR = 1.03; 95% CI 0.84–1.27). When we analyzed the effect of aspirin in the subset of patients with a history of angina or myocardial infarction, the 2 most frequent indications for aspirin in our population, the RR for recent long-term use was 0.85 (0.64–1.12).
|Drug use1||Cases (N = 4,336)||Controls (N = 10,000)||Relative risk2||Adjusted Relative risk3|
|Recent use||958 (22.1)4||1,726 (17.3)4||1.38 (1.26–1.52)5||1.12 (0.99–1.28)5|
|<1 year||349 (8.1)||617 (6.2)||1.40 (1.22–1.61)||1.09 (0.92–1.30)|
|>1 year||609 (14.1)||1,109 (11.1)||1.37 (1.23–1.53)||1.15 (0.99–1.34)|
|75 mg/day||381 (8.8)||760 (7.6)||1.25 (1.10–1.43)||1.03 (0.86–1.22)|
|≥150 mg/day||204 (4.7)||298 (3.0)||1.70 (1.41–2.04)||1.53 (1.22–1.92)|
|Past use||204 (4.7)||375 (3.8)||1.35 (1.13–1.62)||1.03 (0.84–1.27)|
|Nonuse||3,174 (73.2)||7,899 (79.0)||Reference||Reference|
|Recent use||938 (21.6)||2,146 (21.5)||1.03 (0.94–1.13)||0.89 (0.79–1.00)|
|<1 year||763 (17.6)||1,700 (17.0)||1.05 (0.96–1.17)||0.92 (0.81–1.04)|
|>1 year||175 (4.0)||446 (4.5)||0.92 (0.76–1.10)||0.76 (0.61–0.94)|
|Low dose6||83 (1.9)||251 (2.5)||0.77 (0.60–1.00)||0.65 (0.49–0.87)|
|High dose||80 (1.9)||165 (1.7)||1.13 (0.86–1.48)||0.92 (0.67–1.27)|
|Past use||1,374 (31.7)||3,076 (30.8)||1.05 (0.97–1.14)||0.86 (0.78–0.95)|
|Nonuse||2,024 (46.7)||4,778 (47.8)||Reference||Reference|
Likewise, 21.6% of cases and 21.5% of controls had used prescription NA-NSAIDs during the year before the index date. Compared with nonuse, the RR for lung cancer associated with NA-NSAID use was 0.89 (0.79–1.00) overall and 0.76 (0.61–0.94) when use lasted more than a year. Further evaluation of the role of treatment duration resulted in a RR of 0.92 (0.82–1.05) for less than 6 months, 0.86 (0.62–1.19) for 6–12 months, 0.73 (0.54–1.00) for 12–24 months and 0.78 (0.59–1.03) for more than 24 months of treatment. Lower doses were associated with a lower RR. For patients who had stopped using NA-NSAIDs the RR was 0.86 (0.78–0.95) after 1 year. Within these past users, the RR was 0.87 (0.78–0.96) for short (up to 12 months) and 0.71 (0.48–1.05) for longer treatment durations. The RR for NA-NSAIDs in the subset of patients with a history of osteoarthritis or rheumatoid arthritis, the most frequent indications for long-term prescription NA-NSAIDs in our population, was similar to that in patients without arthritis. The RRs varied among individual NSAIDs, although the confidence intervals were overlapping (Table III).
|Drug use1||Cases (N = 4,336)||Controls (N = 10,000)||Relative risk2||Adjusted relative risk3|
|Recent use||327 (7.5)4||801 (8.0)4||0.95 (0.83–1.09)5||0.88 (0.76–1.04)5|
|<1 year||300 (6.9)||722 (7.2)||0.97 (0.84–1.12)||0.90 (0.76–1.06)|
|>1 year||27 (0.6)||79 (0.8)||0.78 (0.51–1.22)||0.78 (0.47–1.27)|
|Past use||1,044 (24.1)||2,252 (22.5)||1.09 (1.00–1.19)||0.92 (0.83–1.02)|
|Nonuse||2,965 (68.4)||6,947 (69.5)||Reference||Reference|
|Recent use||361 (8.3)||792 (7.9)||1.07 (0.94–1.22)||0.99 (0.85–1.16)|
|<1 year||297 (6.9)||638 (6.4)||1.09 (0.95–1.26)||1.00 (0.85–1.19)|
|>1 year||64 (1.5)||154 (1.5)||0.96 (0.72–1.29)||0.93 (0.66–1.31)|
|Past use||763 (17.6)||1,673 (16.7)||1.07 (0.97–1.18)||0.94 (0.84–1.06)|
|Nonuse||3,212 (74.1)||7,535 (75.4)||Reference||Reference|
|Recent use||90 (2.1)||229 (2.3)||0.92 (0.72–1.18)||0.85 (0.64–1.12)|
|<1 year||79 (1.8)||195 (2.0)||0.95 (0.73–1.24)||0.94 (0.70–1.27)|
|>1 year||11 (0.3)||34 (0.3)||0.75 (0.38–1.48)||0.48 (0.23–1.01)|
|Past use||441 (10.2)||866 (8.7)||1.20 (1.06–1.35)||0.98 (0.85–1.13)|
|Nonuse||3,805 (87.8)||8,905 (89.1)||Reference||Reference|
|Recent use||98 (2.3)||193 (1.9)||1.19 (0.93–1.53)||0.93 (0.69–1.25)|
|<1 year||85 (2.0)||162 (1.6)||1.23 (0.94–1.61)||0.95 (0.69–1.31)|
|>1 year||13 (0.3)||31 (0.3)||1.00 (0.52–1.91)||0.81 (0.38–1.72)|
|Past use||39 (0.9)||68 (0.7)||1.34 (0.90–2.00)||1.29 (0.80–2.09)|
|Nonuse||4,199 (96.8)||9,739 (97.4)||Reference||Reference|
Neither sex nor smoking status modified significantly the association between NSAIDs and lung cancer (Table IV). The inverse association between lung cancer and use of NA-NSAIDs for over a year was 0.72 for men (0.54–0.95) and 0.83 (0.59–1.18) for women; and was 0.75 (0.51–1.12) among recent smokers and 0.82 (0.54–1.26) among never smokers.
|Drug use and duration of use||Relative risk1|
|<1 year||1.30 (0.91–1.87)3||1.04 (0.60–1.78)||1.00 (0.62–1.62)||1.25 (0.72–2.17)|
|>1 year||1.11 (0.78–1.57)||1.11 (0.66–1.88)||1.04 (0.72–1.50)||1.23 (0.75–2.02)|
|Past use||1.08 (0.68–1.70)||1.25 (0.65–2.40)||0.96 (0.54–1.69)||1.28 (0.63–2.59)|
|<1 year||0.72 (0.56–0.95)||1.08 (0.75–1.56)||0.83 (0.60–1.15)||1.22 (0.81–1.83)|
|>1 year||0.65 (0.39–1.09)||1.10 (0.57–2.13)||0.81 (0.47–1.39)||0.82 (0.40–1.69)|
|Past use||0.84 (0.68–1.04)||0.98 (0.72–1.34)||0.71 (0.54–0.93)||1.23 (0.87–1.75)|
When we looked at fatal lung cancer cases (2,501 patients who died within 6 months of diagnosis), we saw similar effects. Overall, RR estimates were attenuated after inclusion of covariates in the model. The main confounders were history of cardiovascular diseases, COPD, health care utilization patterns and smoking. When the analyses were restricted to patients with information on smoking (over 80% of our study population), the adjusted RRs were 1.15 (0.97–1.35) for aspirin and 0.79 (0.63–1.00) for NA-NSAID use lasting more than a year. Out of the 1,109 control subjects classified as long-term users of aspirin in the main analysis, 87.7% were long-term users in the analysis without lag time; the corresponding proportion for the 446 controls using long-term NA-NSAIDs was 71.5%. When we repeated the analysis without incorporating any lag time, the RR for long-term use during the 0–12 months preceding the diagnosis was 1.20 (1.04–1.38) for aspirin and 0.92 (0.75–1.14) for NA-NSAIDs, and short term use of both aspirin and NA-NSAIDs was associated with a significantly increased risk. When we considered a lag time of 2 years, the RR associated with long-term use during the 24–36 months preceding the diagnosis was 1.11 (95% CI 0.95–1.30) for aspirin and 0.84 (0.67–1.06) for NA-NSAIDs.
We found a modest inverse association between prescription of NA-NSAIDs and the risk of lung cancer. The risk reduction was observed after 6–12 months of treatment and did not increase with the dose. Although the interactions of NA-NSAIDs with sex and smoking were not statistically significant, the inverse association between NA-NSAIDs and lung cancer was more apparent for males than for females, and more so for smokers than for nonsmokers. Results varied numerically but were not statistically different between individual NA-NSAIDs. We found no evidence of a protective effect of aspirin, even for long-term prescriptions and higher doses.
Several mechanisms have been proposed to explain the chemopreventive effects of NSAIDs, including enhancement of cellular immune response, induction of apoptosis and reduction of cell migration, angiogenesis and inflammation.1 Most of these mechanisms relate to their inhibition of the COX enzymes. By inhibiting COX enzymes, NSAIDs reduce the synthesis of prostaglandins from arachidonic acid and, as a result, attenuate inflammation. Prostaglandins may promote tumor growth by stimulating proliferation and migration of neoplasic cells, inhibiting apoptosis, or suppressing immune response to neoplasic cells; high levels of arachidonic acid may promote apoptosis; and inflammatory states may predispose to cancer.3 The proposed antiproliferative effect of NSAIDs presumably lies on their inhibition of the inducible isoform, COX-2. Recent clinical trials have focused on selective COX-2 inhibitors and have demonstrated an effect for secondary prevention of colorectal adenoma.17 While NSAID chemoprophylaxis might be restricted to cancers in the gastrointestinal area, it may also extend to other anatomic sites including cancers of the breast, prostate and lung.2, 3 COX-2 enzymes are overexpressed in human and rodent lung cancer tissues, particularly in nonsmall cell carcinomas,6 and NSAIDs have been shown to reduce COX-2 enzyme levels in lung cancer cells.5, 7, 8
Despite these fairly consistent data from experimental cell or animal models supporting a protective role for aspirin and NA-NSAIDs against lung cancer,5,7, 8, 9, 10 the data for humans has been more equivocal. A few epidemiological studies suggested reduced risk of lung cancer associated with NSAID use,18, 19, 20, 21, 22, 23 others did not support a protective effect either because the RR was below 1 but not statistically significant,11, 12, 24, 25 or because it was 1 or greater.26, 27, 28, 29 The pooled estimate of previous studies is compatible with a significant reduction of lung cancer risk associated with NSAIDs (Table V); the protective effect is stronger for NA-NSAIDs or NSAIDs overall than for aspirin.2, 3, 22 Some studies suggested a greater protective effect for males and for smokers.19,21, 22, 23 The discrepant findings across epidemiological studies may therefore reflect differences in the specific NSAIDs included in the studies, in the specific histological types of lung cancer ascertained or in the gender distribution and smoking habits of the study populations. Other possible reasons for the inconsistency include biased selection of controls, measurement error in assessing the etiologically relevant NSAID use and confounding. We analyze below each of these factors.
|Author||Cases (N)||Design||Exposure definition1||Control for smoking||RR|
|Peto et al.11||25||Clinical trial||7 pills per week, as randomized||Yes||0.64 (0.29–1.41)2|
|Lee et al.36||128||Clinical trial||3–4 times per week, as randomized||Yes||0.88 (0.62, 1.25)|
|Cook et al.12||205||Clinical trial||3–4 pills per week, as randomized||Yes||0.78 (0.59–1.03)|
|Paganini-Hill et al.26||111||Cohort||Any daily use of aspirin at start date||No||0.92 (0.54–1.55)3|
|Thun et al.25||NR||Cohort||≥16 pills per month for at least 1 year prior to start date||Yes||1.10 (0.99–1.22)|
|Schreinemachers et al.18||163||Cohort||Any use of aspirin in the 30 days prior to start date||Yes||0.68 (0.49–0.94)|
|Friss et al.37||447||Cohort||≥1 prescription||No||1.1 (0.9, 1.2)|
|Sorensen et al.38||692||Cohort||≥1 prescription||No||1.1 (1.0–1.2)|
|Holick et al.28||328||Cohort||≥2 times per week at start date||Yes||1.13 (0.89–1.43)||1.07 (0.69–1.66)|
|Ratnasinghe et al.23||162||Cohort||Use of any aspirin in the 1–6 months prior to start date||Yes||0.81 (0.62–1.07)|
|Langman et al.24||2,560||Nested case–control||≥7 prescriptions during months 13–36 before diagnosis||Yes4||0.84 (0.69–1.02)|
|Akhmedkhanov et al.21||81||Nested case–control||≥3 pills per week for ≥6 months at enrollment, 1-year lag time||Yes||0.66 (0.34–1.28)|
|Rosenberg et al.27||1,110||Case–control||≥4 days per week for ≥3 months, 1.5 years of lag time||Yes||1.0 (0.7–1.4)|
|Harris et al.19||489||Case–control||≥7 pills per week for ≥2 years at interview date||Yes||0.32 (0.23–0.44)|
|Moysich et al.20||868||Case–control||≥1 pills per week for ≥1 year vs. infrequent/no use||Yes||0.57 (0.41–0.78)|
|Muscat et al.22||1,038||Case–control||≥3 pills per week for ≥1 year||Yes||0.84 (0.62–1.11)||0.68 (0.53–0.89)||0.45 (0.30–0.65)3|
|Current study||4,336||Nested case–control||Use for >1 year, 1-year lag time||Yes4||1.15 (0.99–1.34)||0.76 (0.61–0.94)|
|Pooled RR||p < 0.05 (heterogeneity test)||0.91 (0.81–1.02)||0.66 (0.42–1.02)||0.81 (0.56–1.16)|
Since different NSAIDs might have different effects, studies should at least evaluate aspirin and NA-NSAIDs separately. Findings are compatible with a stronger chemopreventive effect of NA-NSAIDs for lung cancer, which is biologically plausible since aspirin inactivates preferentially the COX-1 isoenzyme, particularly at low doses as those used for cardioprophylaxis.33 In addition, aspirin might distinctively affect tumor growth through its antithrombotic actions. However, any finding could have been presented as biologically plausible under the current stage of knowledge. Moreover, the higher RRs estimated for aspirin when compared with those for NA-NSAIDs might be explained by different degrees of residual confounding, as we will argue later, which reinforces the importance of evaluating aspirin and NA-NSAIDs individually.
The effect of NSAIDs may21 or may not22 vary according to the histological subtype of the tumor.1, 21 If it does, one could speculate that a different histological distribution of tumors in females compared with males or in smokers compared with nonsmokers may also explain the different effects of NSAID found in females and males or smokers and nonsmokers, respectively. Unfortunately, the histological types of lung cancer were not addressed in our study due to incomplete pathology information.
Regarding biases in control selection, it is worth noting that the most protective RR estimate came from a case–control study that ascertained cases from a cancer hospital and controls from health screening clinics.19 Since screening participants might be more health conscious than the general population, they might overrepresent the use of long-term aspirin for cardioprotection in the source population. Indeed, this study reported the highest frequency of NSAID use; 26% of controls had used 1 or more pills per day for at least 2 years. In another hospital-based case–control study, which reported the second most protective RR estimate, the authors acknowledge a potential greater likelihood of aspirin use among controls, who could have used aspirin to treat their conditions.20 Such selection biases would have produced spurious negative associations, therefore overestimating any protective effect of NSAIDs. Another hospital-based case–control study excluded from the analysis controls with conditions for which NSAIDs are either indicated or contraindicated. This exclusion criteria was apparently not applied to cases, thus potentially under- or overestimating the use of NSAIDs in the source population and consequently introducing bias.22
Exposure definition in previous epidemiological studies was quite heterogeneous. According to the proposed etiologic mechanisms, long-term use of NSAIDs would be needed in order to observe the hypothesized protective effect. In addition, a lag time of at least 6 months seems appropriate, since use during the months preceding the diagnosis might be (i) not etiologically relevant given the subclinical latent periods of tumors and (ii) potentially biased if initial lung cancer symptoms determined the prescription of NSAIDs, as confirmed in our study by the increased short-term use of both aspirin and NA-NSAIDs during the months preceding the diagnosis. However, studies using relatively broad exposure definitions reported associations similar to the ones observed in studies with more a priori “valid” exposure definitions. While it is difficult to justify the validity of results based on short-term exposures, particularly those without lag time (e.g., including 1 dose the day before diagnosis), long-term exposures might be less sensitive to inadequate lag time definitions. Our own results were relatively resilient to changes in lag time specification for long-term use. This might be due to the fact that long-term NSAID users receive prescriptions throughout long periods of time and, therefore, patients classified as exposed right before diagnosis, or at study baseline, still have a relatively high probability of being exposed during the etiologically relevant period.24
Results from observational studies that collected exposure information through questionnaires or retrospective personal interviews were susceptible to recall bias, while studies based on prescription or dispensing data did not account for over-the-counter use and noncompliance. In the current study, detailed information on dates of use and type of drugs used was ascertained prospectively from computerized prescriptions, and was therefore not affected by patient recall or survival. However, we were not able to capture over-the-counter use of aspirin and ibuprofen, which were available without prescription in the UK. Yet, the degree of underestimation of NSAID use would be particularly limited in the current study, given our focus on long term exposure mainly in elderly people who had access to these drugs for free through general practitioner prescription. We nevertheless performed a sensitivity analysis to quantify the impact of misclassification due to unrecorded over-the-counter use or to incomplete compliance.34 With false negative probabilities beyond 30%, the net impact of nondifferential under-recorded use of NSAIDs would have been a small bias towards the null. That is, the protective effect found for NA-NSAIDs could not be explained by the unrecorded use. Moreover, although misclassification of exposures collected prospectively is usually close to nondifferential among cases and controls, we also examined the impact of differential misclassification, in case certain characteristics of the cases (e.g., higher prevalence of smokers) made them more prone to use over-the-counter NSAIDs. Only unrealistically high differential under-recording (e.g., 20% for cases and 0% for controls) cancelled the inverse association found for NA-NSAIDs. The negligible impact of missing over-the-counter anti-inflammatory drugs use had been previously reported.30 More directly, using similar methods, our group found the currently widely accepted reduced risk of colorectal cancer,31 suggesting that we would have been able to detect associations of similar magnitude for lung cancer.
Finally, heterogeneity among study results might be explained by different degrees of residual confounding. Analgesics might be prescribed to treat conditions that could be directly or indirectly associated with the risk of cancer (e.g., patients with osteoarthritis might have a reduced risk of lung cancer).32 We adjusted for cardiovascular and osteoarticular disorders, the main indications for long-term use of aspirin and NA-NSAIDs, respectively, as well as for other relevant medical and demographic characteristics. The lack of detailed smoking history (e.g., time from starting or quitting) limited our ability to control for smoking habits. However, the smoking information was enough for us to find the expected strong association between smoking and lung cancer, and neither model adjustment nor restriction to subjects with smoking information suggested a major role for potential residual confounding by smoking once we adjusted for cardiovascular clinical risk factors. This agrees with prior evidence suggesting that adjustment for smoking status does not change the relative risk estimate as much as one would expect.21, 23 The relatively modest confounding role of such a strong risk factor for lung cancer is due to the weak association between smoking and NSAID use.22, 23 Similarly, although we did not have data on diet or family history, previous studies have shown that neither of these factors were important confounders.28
Yet, several lines of evidence support the existence of residual confounding for the effect of aspirin by either underlying clinical conditions or common risk factors for cardiovascular disease and lung cancer (e.g. smoking): (i) the RR for aspirin changed from 1.37 to 1.15 upon adjustment for measured confounders, suggesting that better control for these factors would tend to lower the RR towards a protective effect. In fact, the RR moved further to 0.85 upon restriction of the analysis to the subset of patients with a clear cardiovascular indication. (ii) Smoking is the strongest known risk factor for lung cancer. Therefore, the more frequent use of aspirin found among ex-smokers (probably because cardioprotective aspirin therapy was accompanied by smoking cessation advices) would create a positive association between aspirin and lung cancer that might be difficult to fully control in the analyses. Residual confounding by smoking is further supported by the increased risk associated with aspirin in patients without a history of ischemic heart disease or cerebrovascular disease, among whom the association between smoking and aspirin use was stronger (perhaps because smoking is a more prominent determinant of cardioprophilaxis in the absence of previous serious cardiovascular events). (iii) Results from 3 randomized controlled trials (i.e., studies without confounding) are compatible with a protective effect (pooled RR = 0.80; 95% CI 0.65–0.99) (see Table V). Unfortunately, these trials were not designed to study lung cancer and had inadequate statistical power. Altogether, these considerations suggest that the increased risk of lung cancer (15%) associated with aspirin in the current study might just reflect residual confounding. On the other hand, it is unlikely for confounding to explain the protective RR found for NA-NSAIDs, since NA-NSAID use was not associated with smoking habits (i.e., the potential impact of residual confounding by smoking is limited) and adjustment for measured confounders moved the RR for NA-NSAIDs towards a protective effect (i.e., better control would probably result in lower RR).
The overall incidence of lung cancer in the UK is around 66 cases per 100,000 persons per year.35 If the 20% risk reduction associated with long term NA-NSAID treatment proves to be true, it would translate into 13 cases prevented every 100,000 persons treated for at least 1 year. This is a modest impact compared with the number of cases that would be prevented through smoking cessation. However, the benefit might be greater for specific groups of patients at a particularly high risk of lung cancer. For example, for males over 60 who smoke, whose incidence rate is above 1,000 cases per 100,000 person-years, the suggested 30% risk reduction among smokers would translate into more than 300 cases prevented every 100,000 persons on prolonged NA-NSAID treatment.
In conclusion, prolonged use of NA-NSAIDs might be associated with a modest reduction of lung cancer risk. Evidence for a reduced risk associated with long-term aspirin use at cardioprophylactic doses remains weak and difficult to evaluate under a scenario of potential residual bias.
- 14Feasibility study and methodology to create a quality-evaluated database of primary care data. Inform Prim Care 2004; 12: 171–7., , .
- 15NHS Terminology Service. Available at http://www.connectingforhealth.nhs.uk/terminology/readcodes 2006.
- 16Multilex Drug Data File. Available at http://www.firstdatabank.co.uk/products/multilex/ 2006.
- 30Using prescription claims for drugs available over-the-counter (OTC). Pharmacoepidemiol Drug Saf 2000; 9: S37., , , , , , .
- 35GLOBOCAN 2002. Cancer incidence, mortality and prevalence worldwide. IARC CancerBase No. 5, version 2.0. Lyon: IARC Press, 2004., , , .
- 36Low-dose aspirin and risk of cancer: the Physicians Health Study [abstract]. Am J Epidemiol 1995; 141: S28., , .