Target trial emulation of aspirin after diagnosis of colorectal polyps

Colorectal cancer (CRC) is the second leading cause of cancer deaths worldwide [1]. Removing precursor lesions [2, 3] can reduce the risk of CRC incidence and mortality. However, after the removal of high-risk polyps, individuals still have an increased risk of future CRC compared to individuals without colorectal polyps [4]. Based on the extrapolation of cardiovascular prevention trials [5] and observational studies, daily aspirin intake has been proposed to prevent the development of CRC. The effectiveness has been comparable to endoscopic or fecal blood screening [6]. In 2016 the United States Preventive Services Task Force (USPSTF) advocated low-dose aspirin treatment for the primary prevention of cardiovascular disease (CVD) and CRC in adults aged 50–59 years [7]. However, a recently published randomized trial in healthy people age > 65—the ASPREE trial—was prematurely terminated as it showed increased all-cause mortality⁸ and, more specifically, increased CRC mortality [8] in aspirin users. Therefore, the USPSTF revisited the issue of aspirin prevention and published a draft evidence review in October 2021 [9]. The review concluded that current evidence (data from primary CVD prevention populations [9] and longer-term follow-up data from the Women's Health Study, WHS) did not support the claim that low-dose aspirin reduces CRC incidence or mortality. Contrary to the USPSTF review, the American College of Gastroenterology recommends aspirin for CRC prevention in individuals aged 50–69 years with a cardiovascular disease risk of ≥ 10% over the next 10 years but no increased risk of bleeding [10]. Outcome discrepancy (from a large reduction to increased risk of CRC and CRC mortality) is likely explained by differences in study populations (effect modification), study design (RCT vs. observational extrapolation of RCT after the end of study vs. observational studies) and lack of assessment of competing risk in selected populations. To enhance knowledge on aspirin in CRC prevention, we emulated a study of aspirin initiation in a selected group most likely to benefit from CRC prevention; individuals with newly diagnosed colorectal polyps, where prevention could also be thought of as “secondary” given the increased risk of CRC in this patient group.

Of 31,633 individuals meeting the inclusion criteria, 1716 (5.4%) initiated aspirin within 2 years of polyp diagnosis (Flowchart in Fig. 1). Of those 80% had more than a year in between first and last expedition of aspirin (over 50% had more than 5 years and around 25% had more than ten years of use). The median length of follow-up was 8.07 years and the median age of the study participants was 63 years. The number of study participants, events and crude incidence rates for CRC mortality according to initiation status at 2 years and pre-specified strata are presented in Table 1. Initiation of aspirin was more common in men, older individuals and those with ≤ 9 years of education. Aspirin initiators were compared with 29,917 non-initiators.

The HR (95%CI) for CRC mortality was 0.90 (0.75–1.06) for the full IPW model, 0.97 (0.75–1.06) for the baseline model and 0.97 (0.82–1.14) for the unadjusted model. In our posthoc model using a 12 months grace period corresponding numbers were 0.92 (0.77–1.60) for full model, 0.97 (0.76–1.07) baseline and 0.97 (0.82–1.15) unadjusted. The HR for CRC incidence was 0.88 (0.86–0.90) for essentially all three models (differing by < 0.02 in the CI estimates). The HR (95%CI) for all-cause mortality was 1.18 (1.12–1.24) in the full IPW model, 1.18 (1.11–1.23) in the baseline model and 1.31 (1.25–1.38) in the unadjusted model. IP weights for CRC incidence had a max value of 2003, mean = 1.81 and p99.8 = 1.28, for mortality outcomes IP weights max was 2077, mean = 1.79 and p99.8 = 1.27. In total, 18 (1.1%) of the aspirin initiators had a bleeding-related cause of death; this compares with 166 (0.6%) non-aspirin users, corresponding HR for bleeding-related cause of death were 1.45 (1.05–2.07) for the unadjusted, 1.45 (0.90–1.70) for baseline adjusted and 1.23 (0.89–1.73) for full IPW models.

The cumulative 10-year incidence of CRC standardized to baseline variables was about 6% in initiators and 8% in non-initiators, risk difference − 2.7% (− 3.0– − 2.5%) (Fig. 2). For CRC mortality, the 10-year cumulative incidence was 1% for both groups, risk difference − 0.2% (− 0.6 − ( +)0.1%) in our emulated models compared to crude rates of 0.8% in initiators vs. 0.7% in non-initiators for a median follow-up of 8 years (Table 1). The cumulative incidence of all-cause mortality at 10 years was 21% in initiators vs. 18% in non-initiators, corresponding risk difference 3.8% (2.0–5.4%). In individuals aged < 66 years we found that all-cause mortality increased in those who initiated aspirin, but the risk difference was smaller and confidence interval was non-significant(11% vs. 10%, risk difference 1% (− 0.05–4.2%) than that for all ages at 10-years due to fewer events (Fig. 3).

Stratified analyses (Table 2) showed that no subgroup had a significant reduction in CRC mortality. A decrease in CRC incidence was only significant in the subgroups < 66 years of age, women, villous polyps and without prior diabetes.

Our study showed that CVD-naive individuals initiating aspirin within 2 years of polyp diagnosis had a 2% lower 10-year cumulative incidence of CRC. No difference was seen in CRC mortality. Start of aspirin use was further associated with an 18% increased relative risk of all-cause mortality and an absolute mortality risk difference of + 4% after 10 years. Because most individuals with myocardial infarction and stroke should have been prescribed aspirin at the first cardiovascular event, the increased risk of all-cause mortality may include a lack of positivity, i.e. probability of having the exposure, conditional on covariates, is greater than 0 and less than 1, for all strata and exposure levels of interest. However, we checked that positivity criteria were held in the data and also adjusted for these variables in our IPW model without more than marginal effects. Our estimated HR of 1.18 for all-cause mortality is close to the HR of 1.14 reported from the ASPREE trail [21], and also older meta-analyses of all previous primary prevention trials of aspirin including high risk CVD groups and health care workers have failed to show a beneficial effect of aspirin on all-cause mortality (HR/OR 0.94 ranging from 0.80 to 1.25 in individual studies) [22]. Therefore, it is possible that aspirin intervention may be linked to a real increase in all-cause mortality in older [23] individuals lacking manifest CVD risk factors. Another prevention study of aspirin initiation in individuals with a history of atherosclerotic risk factors but without manifest atherosclerotic disease at age 60–85 years was terminated prematurely because no beneficial effect was observed [24]. However we canont rule out that CVD in being an indication for aspirin initiation may still have caused undetectable positivity problems that impacted our all-cause estimate to be higher than a true causal estimate, but should however not have had any impact on CRC incidence and mortality. Our findings add to the accumulating evidence that aspirin has no net mortality benefit when used as a chemoprevention agent in CRC, even in individuals with increased risk of CRC without CVD risk factors at polyp diagnosis. Further we acknowledge that the results of this emulated trial should be interpreted as effects of aspirin initiation in addition to polyp removal. However for individuals with polyps detected we believe that aspirin initiation without polyp removal is not a realistic treatment alternative. Because we chose not to include individuals with a history of previous CVD events, this selection, and hence lack of baseline risk of competing CVD death, may explain some discrepancies from previous extensions of cardiovascular trials showing beneficial results [25]. However, it is impossible to perform a good observational study without excluding individuals with previous CVD, given that they are unlikely to be eligible for aspirin as chemoprevention if they did not initiate the drug for well-documented secondary prevention following a CVD event.

Our study is the first to specifically emulate a target trial of aspirin initiation in individuals with a newly diagnosed polyp. This group represents a selection of the general population expected to benefit most from CRC chemoprevention with aspirin, particularly for those with villous components (not recommended surveillance in Sweden during the study period unless the size was > 10 mm [26]) at a high risk of incident cancer [4]. Here, we found a reduced risk of incident CRC but no impact on CRC mortality; and no significant benefit in CRC mortality was found in any subgroup. On the contrary, crude/unadjusted numbers showed higher CRC mortality in the initiators (0.8% vs. 0.7%) after a median follow-up of 8 years.

Our study has several strengths, including a large cohort, high validity of the nationwide Swedish registers (National Patient Register [27], Cause of Death Register [28], Cancer Register [29], Prescribed Drug Register [15] and ESPRESSO [12]), a well-defined intervention in terms of time in relation to polyp detection, a relevant population selected to benefit the most from potential chemoprevention and the use of a robust methodology allowing for causal inference. Causal inference methods design, when data satisfy assumptions, avoid several common biases of observational studies (e.g., immortal time and prevalent user bias). In addition, it relates to a well-defined intervention that can be translated into clinical practice [30]. Finally, our median follow-up of 8 years is longer than the 4.7 years of the ASPREE trial and spans a time frame that could detect differences in long-term CRC mortality.

Our study has several limitations worth noting. We lacked information on the size and exact colonic location of polyps, smoking status (although we used COPD as a proxy measure for heavy smoking), body mass index, alcohol use, lifestyle risk factors and diet. We also lacked information on the number of endoscopic examinations during the follow-up. Hence there is a potential for residual or unmeasured confounding from the above mentioned factors as well as from lack of adjustment for other medications. The direction of this bias would likely overestimate all-cause mortality among aspirin users due to higher prevalence of risk factors of all-cause mortality among aspirin users, but how it would affect CRC incidence/mortality is not clear as CVD is also a competeing event in individuals with increasing number of combined risk factors. In this study competing events (ie. death from non CRC) was handled as a censoring event, no other analytical approach was used, however the risk of competing CVD events should be lower in this population, where previous history of such events were exclusion criteria than in studies where the population base was indeed individuals with CVD risk factors. Because most of our population was Caucasian, our results may not apply to non-Caucasian populations. At the time of the study, CRC screening in Sweden was only practiced in the capital region of Stockholm and not nationwide [31]. Therefore, our results are mainly applicable to polyps detected from symptoms. The two-year grace period was chosen to allow completion of the clinical workup for symptoms such as anemia so that any prevalent cancers were discovered during the grace period and that anemia could be reversed before initiation was mandated. We acknowledge that the effect of aspirin initiation may not be homogenous during the two year grace period (i.e. may be smaller if initiation took place 24 months vs 1 month post baseline). But as CRC is generally thought of as a slow growing cancer we do believe the average effect is representative of the overall potential clinical benefit of aspirin use. This is further supported by our posthoc analysis showing very similar results (non-significant HR = 0.92 vs 0.90) for CRC mortality using a 12 months grace period. Furthermore, we did not consider the length of aspirin use after the initial introduction, as modeling of discontinuation would cause intractable bias from discontinuation being heavily associated with shorter life expectancy, side effects or other contraindications [32]. Because aspirin initiation was slightly more common in this study's first 3.5 calendar years, we cannot exclude the possibility that some prevalent users without previously registered prescriptions in the Prescribed Drug Register were included within the first calendar period. A median follow-up of 8 years may also be too short to detect a benefit on CRC mortality.

In conclusion, this study found that in individuals with polyp removal, initiation of aspirin led to a 2% lower cumulative incidence of CRC over 10 years but had no effect on CRC mortality and was associated with a 4% higher cumulative incidence of mortality.