Observational studies suggested that regular aspirin use may prevent cancer.[1, 2] A meta-analysis of individual patient data of 51 randomized controlled cardiovascular trials including more than 70,000 individuals showed a reduction of cancer incidence (meta-odds ratio [OR] 0.71, 95% confidence interval [CI] 0.57–0.89) and mortality (OR 0.63, 95% CI 0.49–0.82), which became most apparent 5 years after randomization to daily aspirin. However, these trials were primarily designed to assess prevention of cardiovascular events and prevention of cancer was assessed as a secondary endpoint. Long term follow up studies of these trials lacked statistical power to determine an effect of aspirin use on site-specific cancers and could only establish an effect on all cancers combined. Furthermore, trial participants are a highly selected group and therefore results cannot be extrapolated to the general population. The risk reduction of cancer mortality with aspirin use in real life appeared to be modest compared to the meta-analysis of trials (hazard ratio [HR] >5 years 0.84, 95% CI 0.75–0.95). The objective of this study was to investigate the possible protective effect of long term low dose aspirin use (≤100 mg daily) on the incidence of site-specific cancers in the Dutch general population. Because the protective effect of low dose aspirin became most apparent 5 years after randomization, we hypothesized that long term low dose aspirin users may have a lower cancer risk than short term users. Data from the Eindhoven Cancer Registry (ECR) and the PHARMO Record Linkage System (RLS) were used to obtain high quality and complete information on low dose aspirin exposure (≤100 mg daily) of all new low dose aspirin users and cancer incidence over a period of 12 years.
Observational and intervention studies suggest that low dose aspirin use may prevent cancer. The objective of this study was to investigate the protective effect of long term low dose aspirin use (≤100 mg daily) on cancer in general and site-specific cancer among low dose aspirin users in the Dutch general population. We conducted a population-based cohort study with detailed information on aspirin exposure and cancer incidence. Only incident (new) low dose aspirin users, who were included in the linkage between PHARMO and the Eindhoven Cancer Registry (1998–2010) and free of cancer before the start of follow up were included. A Cox proportional hazard model with cumulative aspirin use as a time-varying determinant was used to obtain hazard ratios (HR). Duration of aspirin use amongst 109,276 incident low dose aspirin users was not associated with a decreased risk of any of the site-specific cancers or cancer in general (adjusted HR per year of aspirin use for all cancers: 1.02, 95% confidence interval [CI] 1.00–1.04, HR of >6 years aspirin use compared to <2 years: 1.17, 95% CI 1.02–1.34). After adjusting for current and past aspirin use, 2–6 years of low dose aspirin use was associated with a reduced colorectal cancer risk compared to <2 years of aspirin use (adjusted HR 0.75, 95% CI 0.59–0.96). However, a clear dose-response relationship was not observed (adjusted HR >6 years aspirin use 0.95, 95% CI 0.60–1.49). Our results do not support the primary prevention of cancer among long term aspirin users.
The coverage area of both the ECR and the PHARMO RLS includes more than one million Dutch citizens. Briefly, the ECR is a population-based cancer registry in the south of the Netherlands covering 2.4 million inhabitants. The ECR includes more than 95% of all newly diagnosed malignancies and is based on the automated pathological archive (PALGA), the national registry of hospital discharge diagnosis (LMR), hematology departments and radiotherapy institutions.[6-8]
PHARMO RLS is a network of patient databases, which covers a demographic region of more than 3 million inhabitants including, among other things, community (out-patient) pharmacy data.[8, 9] The community pharmacy database includes all pharmacy dispensed healthcare products on the Dutch market, prescribed by general practitioners or specialists, including aspirin dispensings. Over the counter (OTC) aspirin use was not included. Patients were followed over a long period of time; until they moved away from the ECR-PHARMO catchment area, end of data collection of the specific community pharmacy, end of study period, or death, whichever occurred first. The date of death was obtained from the central bureau for genealogy (CBG), the local pharmacy or the hospital.
All citizens who lived in the ECR-PHARMO catchment area between 1 January 1998 and 31 December 2010 and were above 18 years were eligible (n = 1,263,935, Fig. 1). Participants were required to have a complete prescription history since the date of entry in PHARMO and at least 1 year of follow up to ascertain a new user design (n = 1,233,205, Supporting Information Fig. 1). Participants with a low dose aspirin dispensing during the first year of follow up were considered prevalent users and were excluded from the main analyses (n = 44,986). All incident (new) low dose aspirin users without cancer (excluding [excl] nonmelanoma skin cancer [NMSC]) before their first aspirin dispensing were included in the analysis (n = 109,276). Patients with a NMSC diagnosis prior to their first aspirin dispense were excluded from the NMSC analyses.
Exposure to low dose aspirin
Dispensings were selected based on the Anatomical Therapeutic Chemical (ATC) codes of the World Health Organization (WHO) collaborating centre for drug statistics methodology. Dispensings with full ATC codes; B01AC06, B01AC08 and B01AC30 were considered low dose aspirin dispensings (≤100 mg daily). To calculate the duration of each dispense, the amount of dispensed drug was divided by the amount prescribed per day, as defined in the pharmacy data.
Cancer as outcome
The invasive (grouped) cancers which were analyzed in this study were cancer of the upper gastrointestinal (GI) tract (C15 and C16 of the International Classification of Disease 10 [ICD-10]), colorectal cancer (C18-C20), lung cancer (C33, C34), melanoma (C43), basal cell carcinoma (BCC) (C44) and other skin cancers (C44), breast cancer (C50), female genital cancer (C53-C56), prostate cancer (C61), urinary tract cancer (C64-C68) and lymphomas (C81-C88).
Age at start of follow up, sex, comedication use and comorbidities were considered potential confounders. As a proxy for comorbidities and the associated health care utilization we determined the unique number of dispensings (ATC codes) and the unique number of hospital discharge diagnosis (LMR) in the year previous to start of follow up.[11, 12] Comedication use which may influence risk of cancer was recorded for the year prior to start of follow up and included nonsteroidal anti-inflammatory drugs (NSAID), statins, angiotensin converting enzyme (ACE) inhibitors or angiotensin receptor blockers, glucocorticoids and other immunosuppressive drugs (Supporting Information Table 1).
|New aspirin users|
|Males (N, %)||53,679||(49.1)|
|Females (N, %)||55,597||(50.9)|
|Age in years (median, IQR)||59||(49–69)|
|Follow up in years since first aspirin dispense until death, end of study or lost to follow up (median, IQR)||4.4||(2.0–7.0)|
|Reasons for end of follow up (N,%)|
|End of study||89,406||(81.8)|
|End of data collection of the pharmacy||7,826||(7.2)|
|Moved away from the ECR-PHARMO catchment area||1,836||(1.7)|
|Time since first aspirin dispense until first cancer diagnosis, excl. NMSC (N, %)|
|Duration of low dose aspirin use (N,%)|
|Unique number of dispensings in the year before start of follow up (%)|
|1 or 2||30,149||(27.6)|
|Unique number of hospitalisation in the year before start of follow up (N, %)|
|Comedication in the year before start of follow up (N, %)|
|Immunosuppresive drug users||5,563||(5.1)|
|ACE inhibitor/AR blocker users||9,934||(9.1)|
The association between duration of low dose aspirin use and cancer was analyzed by using a Cox proportional hazard (PH) model with cumulative drug use as a time-dependent covariate. In these analyses the cumulative duration of aspirin use of the subject with the event of interest is compared with the cumulative duration of aspirin use of all other subjects at the same time point. Time since first aspirin dispensing was used as underlying timescale (Supporting Information Fig. 1).
A priori we decided that all analyses would be adjusted for age, sex, co-medication use and comorbidities. The linearity assumption was not met for age and therefore a restricted cubic spline function with four knots was included in the model for age. Final analyses were adjusted for unique number of ATC and LMR codes in the year prior to start of follow up (i.e., replacing ATC/LMR codes in the multivariable model for aforementioned co-medication use as binary variables (yes/no) resulted in similar results and the same conclusions). The correlation between the unique number of dispensings and hospitalizations was low (Pearsons correlation coefficient r = 0.22), indicating a low possibility of collinearity.
First, multivariable analyses were performed with duration of aspirin use as a continuous variable and a categorical variable (main analysis). Sensitivity analyses were performed to test if the main results were consistent with results obtained by alternative analysis strategies. Second, two lag time analyses were performed, discarding aspirin exposure 1 or 3 years before diagnosis to control for (i) misclassification of exposure time, due to the latency time, (ii) increased health care utilization before diagnosis and (iii) reverse causation. Third, because in our models with cumulative exposure, no distinction was made between current use and past use, we adjusted the analyses for the number of days elapsed since last aspirin use. Fourth, prevalent aspirin users will include many compliant and long term users, who are particularly important for evaluating long term effects of aspirin use. Excluding them may cause an overrepresentation of non-compliant and short term aspirin users in the main analyses. Therefore we included the prevalent users in a sensitivity analysis. Fifth, the reliability of recording of moving away from the ECR-PHARMO catchment area was unknown. Therefore patients were censored at the last date of any dispensing, because patients who pick up their dispensings did not move away and are most likely truly at risk in the analyses.
Finally, analyses were stratified by sex, age (<60, ≥60 years), number of unique dispensings (0, 1 or 2, >2) and number of unique hospitalizations (0, 1, >1) to assess the impact of effect modification. To check the PH assumption time interval specific HR were calculated. We observed no violation of the PH assumption. All analyses were performed using SAS 9.2 statistical software (SAS Institute, Cary, NC). All statistical tests were two-sided and considered significant at the p < 0.05 level.
The study population consisted of 109,276 incident low dose aspirin users with a median follow up time of 4.4 years (interquartile range [IQR] 2.0–7.0 years) since their first aspirin dispense (Table 1). The median age of low dose aspirin users was 69 (IQR: 49–69). Of all aspirin users, 12% had at least one hospital discharge diagnosis in the year before start of follow up and 72% used other medications. Most aspirin users used aspirin for 6 years or less. However, more than 14,000 people used low dose aspirin for more than 6 years. The daily low dose aspirin dose was ≥80 mg for 90.9% of all dispensings, ≤38 mg for 7.3% of all dispensings, and between 38 and 80 mg for 1.9% of all dispensings.
Each additional year of low dose of aspirin use was not associated with cancer risk among low dose aspirin users (adjusted HR per year of aspirin use for all cancers [excl. NMSC] 1.02, 95% CI 1.00–1.04) (Fig. 2, Table 2). Long term aspirin users were not at decreased risk of cancer compared to short term users either (adjusted HR >6 years aspirin use for all cancers 1.17, 95% CI 1.02–1.34). An inverse association was observed for upper GI tract cancer, although not statistically significant (adjusted HR per year of aspirin use 0.95, 95% CI 0.88–1.01, adjusted HR >6 years aspirin use 0.88, 95% CI 0.51–1.51).
|Duration of aspirin use|
|Total number||Per year||<2 years||2–6 years||>6 years|
|Na||of events||Adjusted HRb||95% CI||Events||HR||Events||Adjusted HRb||95% CI||Events||Adjusted HR||95% CI|
|All cancers (excluding NMSC)||109,276||5,415||1.02||(1.00–1.04)||3,021||1||1,919||1.06||(0.98–1.14)||475||1.17||(1.02–1.34)|
|Upper GI tract cancer||109,276||268||0.95||(0.88–1.01)||156||1||84||0.79||(0.57–1.11)||28||0.88||(0.51–1.51)|
|Female genital cancer||55,597||188||1.04||(0.95–1.14)||111||1||62||1.08||(0.72–1.60)||15||1.43||(0.69–2.95)|
|Urinary tract cancer||109,276||463||1.02||(0.96–1.07)||241||1||172||1.12||(0.86–1.45)||50||1.25||(0.81–1.91)|
The HR per year of aspirin use for all site-specific cancers decreased after including time since last aspirin use in the model to adjust for the difference between past and current aspirin use (Fig. 2, Supporting Information Table 2). Discarding exposure during 1 or 3 years prior to diagnosis (the latent period of tumor development) did not change the conclusions (Fig. 2, Supporting Information Table 2). After adjusting for time since last use, 2–6 years of low dose aspirin use was associated with a reduced colorectal cancer risk compared to <2 years of aspirin use (adjusted HR 0.75, 95% CI 0.59–0.96). However, a clear dose-response relationship was not observed (adjusted HR >6 years aspirin use 0.95, 95% CI 0.60–1.49) (Fig. 2, Supporting Information Table 2). The nonsignificant inverse association for upper GI tract cancer disappeared after adjustment for time since last aspirin use (adjusted HR 2–6 years 1.02, 95% CI 0.62–1.66, adjusted HR >6 years 1.45, 95% CI 0.59–3.59). The number of long term aspirin users more than doubled after inclusion of prevalent aspirin users (34,168 prevalent and incident users compared to 14,771 incident users with >6 years aspirin use). However, including prevalent aspirin users did not result in a protective association of long term aspirin use compared to short term use (Fig. 2, Supporting Information Table 2). Censoring subjects at the date of the last dispense of any type of medication to control for possible missed moving out of the ECR-PHARMO catchment area, did not change the conclusions (Fig. 2, Supporting Information Table 2). Similar results were observed in analyses stratified for sex, age group, and number of unique dispensings and hospitalizations (data not shown).
Our results show that long term aspirin use among low dose aspirin users may not prevent the incidence of cancer. Long term use among low dose aspirin users in the general population was not associated with a decreased risk of cancer in general or any site-specific cancer. After adjusting for current and past aspirin use, we observed a 25% risk reduction of colorectal cancer after 2–6 years of aspirin use compared to <2 years of aspirin use. A clear dose-response relationship was not observed with a longer duration of aspirin use. This may be due to the low number of long term incident aspirin users or to the low number of events in this category. We observed a non-significant modest inverse association between upper GI tract cancer and duration of aspirin use, which disappeared after adjusting for current and past aspirin use. This may indicate, that long term aspirin use appeared to be protective, because patients with gastric symptoms, who already have a higher risk of upper GI tract cancer, discontinue aspirin, whereas healthy patients continue aspirin use.
Results of other observational and intervention studies examining the association between aspirin use and cancer risk have been conflicting. A meta-analysis of individual patient data of 51 cardiovascular randomized controlled trials showed a reduction of cancer incidence 3 years after randomization to low dose aspirin (75–100 mg daily) in 6 trials including more than 30,000 individuals. A pooled analysis of these 6 trials and 26 trials of daily aspirin of any dose did not show a clear effect on many site-specific cancers, such as cancer of the gastrointestinal tract. A statistically significant risk reduction was only observed for cancers of the female reproductive organs. A direct comparison between this study and our study is not possible, because our study included only aspirin users. Other differences between our study and the meta-analysis of individual patient data include the choice of study population (sample of the general population vs. selected trial participants), analyzed treatment (dispensed drugs vs. scheduled treatment) and information of potential confounders (limited availability in automated health care databases vs. detailed information on an individual patient level). Other meta-analyses of observational studies on regular aspirin use and cancer risk showed a decreased risk of many site-specific cancers including colorectal cancer, esophageal and gastric cancer and other cancers of the digestive tract.[1, 2] In the meta-analyses, modest risk reductions were observed for breast and prostate cancer.[1, 2] Previous studies on lung cancer were inconclusive and no risk reduction was observed for pancreas, endometrium, ovary, bladder, and kidney cancer.[1, 2] Analyses conducted in the vitamins and lifestyle (VITAL) study showed no association between low dose aspirin use and hematological malignancies either. Observational studies of skin cancer and aspirin use also showed inconsistent results.[12, 16-18] The protective findings in some of these studies may be due to the use of a time-fixed analysis of aspirin exposure. This analysis can produce a protective HR in a dataset where there is no association between exposure and outcome, while a time-dependent analysis of exposure results in a correct HR of 1 in the same dataset. Data from the Nurses Health Study (NHS) was analyzed with a time-dependent definition of aspirin exposure showing no risk reductions of squamous and basal cell carcinoma and even a slightly increased risk of melanoma.
Conflicting results were also seen in randomized controlled trials designed to assess cancer risk with regular aspirin use. The Cancer Prevention Programme (CaPP2) did not demonstrate a protective effect of aspirin (600 mg daily) in 1,000 Lynch syndrome patients on the development of colorectal carcinomas 4 years after randomization, but found a sudden risk reduction after a mean follow up of 4.5 years.[20, 21] The Women's Health Study (WHS), (100 mg of alternate day aspirin) did not show an effect on cancer incidence after 10 years of follow up among almost 40,000 participants. The lack of effect was hypothesized to be due to the low dose or low frequency of aspirin use. A recent analysis of posttrial follow-up revealed a delayed effect on colorectal cancer, but not on total cancer or any other site-specific cancer. After a median follow up of 18 years among almost 34,000 women who participated in posttrial follow-up, 53 fewer cancers and 48 fewer cardiovascular disease (CVD) cases occurred, but there was an increase of 193 gastrointestinal bleedings and 214 peptic ulcers. The risk-benefit ratio of aspirin use in a primary prevention setting remains controversial because of its modest effectiveness and risk of bleedings.[24, 25]
Ongoing trials focus on the risks and benefits of daily aspirin: the ASPrin in Reducing Events in the Elderly (ASPREE) trial for example examines if the benefits of aspirin (100 mg daily) outweigh the risks in healthy participants above 70 years. This trial can provide information about the possible extension of a disability free life by using daily aspirin, which is a relevant primary endpoint to assess the risk-benefit ratio (NCT01038583).
Future research may aim to identify biomarkers associated with a reduced cancer risk with aspirin use. Recently, a PIK3CA (the phosphatidylinositol-4,5-bisphosphonate 3-kinase, catalytic subunit alpha polypeptide) gene mutation was identified in colorectal cancer as a potential useful genetic marker to help target adjuvant treatment with aspirin more effectively. Aspirin use among colorectal cancer patients with a mutation in the PIK3CA gene was associated with a 46% reduction in all-cause mortality and a 82% decrease in cancer related mortality. Aspirin use was not associated with mortality in wild type cancers.
Strengths and Limitations
The strengths of this study include the large population-based sample and the detailed and high quality information on both aspirin use and cancer, limiting both selection and information bias. The validity of both exposure and outcome was high, because the cancer registry completeness exceeds 95% and the PHARMO RLS has detailed information on dosage, duration and frequency of each dispensing during follow up. Because of our study methodology, important biases in pharmacoepidemiology, such as immortal time bias,[28, 29] prevalent user bias and lag time bias[12, 16] were circumvented.[13, 30] The limitations include the lack of information on OTC aspirin use and potential confounders. The nondifferential misclassification of exposure due to OTC low dose aspirin use is likely to be minimal, because low dose aspirin for the indication “platelet aggregation inhibition” is only available on prescription in the Netherlands. The proportion of high dose OTC aspirin use is unknown, which could have biased our results towards the null hypothesis. However, it has been shown that, pharmacy data can give valid associations even though a high proportion (25%) of the drugs is available OTC. This suggests that the effect of missing high dose OTC aspirin may not be a large source of bias in the present study. Important confounders, such as use of other medications and comorbidities were included in our analysis. Other important confounders, such as smoking, overweight, alcohol use and other lifestyle factors were not available. This could have influenced our results, although we expected that this would have resulted in a too low HR, rather than towards no effect. In our analysis, adherent long term users were compared to those who discontinue aspirin use. A protective HR may be a result of the healthy adherer bias (i.e., adherence to the drug is associated with other healthy behaviors). Instead, we observed an increased cancer risk among long term users, which may indicate the opposite of the healthy adherer effect. An alternative explanation for our lack of a protective association is an increased ascertainment of cancer among long term users, which could have biased our results towards the null. This is unlikely, because the specificity of the linkage in a random sample with at least 1-year follow up was as high as 99.5%. In addition, an increased ascertainment of cancer among long term users could also have been caused by short term users of whom moving out of the ECR-PHARMO catchment area was not recorded. Therefore, we censored the subjects at the last date of any dispensing, but this did not alter the results. Aspirin users may have died prematurely, which resulted in a high number of short term users, but this could not be confirmed by an analysis of all-cause mortality (adjusted HR per additional year of low dose aspirin use 0.97, 95% CI: 0.96–0.98).
The negative findings of this study cannot be extrapolated to populations at high risk of developing cancer (e.g., patients with a history of cancer or patients with premalignant lesions, such as polyposis, Barett's esophagus or actinic keratoses).
Long term aspirin use among low dose aspirin users was neither associated with a decreased risk of cancer in general, nor with a clear risk reduction of any site-specific cancer in the Dutch general population. Our results do not support the primary prevention of cancer among long term aspirin users. Ongoing clinical trials may provide information about the balance between risks and benefits of aspirin use among subgroups at higher risk for cancer.
The authors thank Mieke Aarts and her colleagues of the Eindhoven Cancer Registry (ECR) for their dedicated data collection and linkage of the data. They also gratefully acknowledge Huub Straatman and his colleagues from the PHARMO institute for drug outcomes research for their dedicated data collection, data processing and help with the SAS programming. They thank both the ECR and the PHARMO institute for drug outcomes research for the availability of their data on favorable terms. They thank Emmilia Dowlatshahi and Satu Siiskonen for critically reading the manuscript.