This was a nationwide Swedish population-based cohort study during the study period 2006 throughout 2020 examining statin use in relation to mortality in patients who underwent curatively intended treatment for gastric adenocarcinoma. Data came from the Swedish Gastric Cancer Surgery Study (SWEGASS), which have been presented in detail in a recent cohort description [14]. In brief, SWEGASS included > 98% of all patients having undergone gastrectomy for gastric adenocarcinoma (including Siewert type III tumours of the gastric cardia) in Sweden from July 1, 2006, to December 31, 2015. Data on the exposure (statin use), were retrieved from the Swedish Prescribed Drug Registry, which electronically and automatically records all prescribed and dispensed drugs in Sweden, except for in-hospital use. The registration started in July 1, 2005 and inclusion started exactly 1 year later to allow assessment of dispensation of statins before surgery. The registry covers all Swedish pharmacies, and the recording is nearly 100% complete [15]. For the present study, the follow-up was updated until December 31, 2020. We excluded other (rarer) histological types of gastric malignancies than adenocarcinoma because of major differences in treatment and survival. Potentially eligible patients were initially identified in the Swedish Cancer Registry and Swedish Patient Registry [16]. The completeness of the registration in the Cancer Registry is 98% for gastric adenocarcinoma overall, and likely even higher for those who undergo gastrectomy [17]. Among all identified patients, the final cohort was selected after a review of medical records, including notes from multidisciplinary meetings, surgery charts, histopathology reports, and hospital discharge summaries [14]. Patients operated without a curative intent were excluded because statins are unlikely to influence survival at non-operable tumour stages. The study was approved by the Regional Ethical Review Board in Stockholm, Sweden (2017/141-31/2).

The study exposure was dispensation of a statin prescribed by a physician within 1 year prior to gastrectomy (or longer), defined by the Anatomical Therapeutic Chemical codes C10AA or C10BA. In separate sensitivity analyses, we examined statin use for 2 years and 3 years preoperatively, as well as statin use during the first postoperative year. All dispensed medications were considered to be taken and dispensation is hereafter termed as “use”. Statins are only available by prescription in Sweden and not over-the-counter.

The main outcome was 5-year disease-specific mortality, defined as death with gastric cancer as an underlying or contributing cause of death, occurring between the date of gastrectomy and 5 years postoperatively. The secondary outcome was 5-year all-cause mortality, defined as death from any cause within 5 years of the gastrectomy. Information on the mortality outcomes was obtained from the Swedish Cause of Death Registry. This registry has 100% completeness for date of death and 96% completeness for cause of death for all Swedish residents, including deaths among Swedish residents who die abroad [18]. Information on date of death is updated continuously, while causes of death are assembled at the end of each calendar year. Therefore, follow-up for disease-specific mortality ended December 31, 2019, which did not allow 5-year follow-up of all patients, while follow-up for all-cause mortality ended December 31, 2020, allowing 5-year follow-up of all participants.

Eleven covariates (categorisations in brackets) were considered potential confounders: age (continuous, linear modelling), sex (male or female), education (≤ 9, 10–12, or > 12 years of formal education, which represents primary school, secondary school, and tertiary education/university), calendar year (continuous, linear modelling), comorbidity (Charlson comorbidity index score 0, 1, or ≥ 2), low-dose (75–160 mg/day) aspirin use (yes or no), tumour sublocation (cardia or non-cardia), pathological tumour stage (0–I, II, III, or IV), neoadjuvant chemotherapy (yes or no), annual surgeon volume of gastrectomy (quartiles, i.e., four equal-sized groups), and radicality of the surgical resection (R0 or R1, [patients with R2 were excluded]). Information on tumour location, tumour stage, neoadjuvant chemotherapy, surgeon volume, and surgical radicality was retrieved from a review of medical records. Data on age, sex, education, calendar year, comorbidity, and aspirin use were obtained from three nationwide complete Swedish registries: Patient Registry, Longitudinal Integrated Database for Health Insurance and Labour Market Studies (LISA), and Prescribed Drug Registry [15, 16, 19]. Comorbidity was classified based on the most well-validated version of the Charlson comorbidity index (Supplementary Table 1) [20]. Aspirin use was included as a covariate because it is often used alongside statins and may improve gastric cancer survival [21]. Type of surgery (total versus subtotal gastrectomy) was not adjusted for due to collinearity with tumour sublocation, which was included in the main model. Postoperative complications were presented but not adjusted for, because these occurred after baseline (date of surgery).

Study patients were followed from the date of gastrectomy until death, 5 years after surgery, or end of the study period, whichever occurred first. The cumulative survival as a function of time was estimated using the Kaplan–Meier estimator for a descriptive comparison of users and non-users of statins (Supplementary Figs. 1 and 2). Cox proportional hazards models were used to calculate hazard ratios (HR) with 95% confidence intervals (CI) of the mortality outcomes among statin users compared to non-users of statins (reference group in all analyses). A multivariable model adjusted for the 11 covariates, described and categorised above. Effect modification analyses were conducted for covariates where a differential effect of statins was plausible. This analysis was conducted by including an interaction term in the models for the subgroups of age (≤ 66, 67–74, and ≥ 75 years), sex (male or female), low-dose aspirin use (yes or no), and pathological tumour stage (0–I, II, III, and IV), for which HRs with 95% CI were presented. For the interaction analyses, the variables were categorised. A likelihood-ratio test for each interaction term at the 0.05 level was computed by calculating the difference between the log likelihood statistics in the main adjusted model and the main adjusted model including the interaction term. To manage missing data, we used complete case analysis, i.e., excluded patients with any missing data in the models. Missing data were found in at least one of the 11 covariates in 10% of patients (n = 167) (education, n = 35; surgeon volume, n = 7; pathological tumour stage, n = 36; neoadjuvant chemotherapy, n = 6; tumour sublocation, n = 7; surgical radicality, n = 95). The proportional hazards assumption was evaluated using log–log survival plots and by calculating the correlations between Schoenfeld residuals for the covariates and ranking of individual failure time. The correlations were low, indicating that the proportional hazards assumption was met for all analyses. A senior biostatistician (FM) conducted the data management and statistical analyses according to a detailed and pre-defined study protocol and used the components SAS/BASE and SAS/STAT in the SAS software, Version 9.4 (SAS Institute Inc., Cary, NC, USA) for these purposes.