Data sources and study population
We conducted this study within the entire Danish population of approximately 5.8 million persons. All Danish residents are provided free access to health care . Since 1968, the Civil Registration System has assigned a unique civil registration number to all residents at birth or upon immigration . The System also records sex and date of birth, and tracks changes in vital status, and migration for the entire population. The registration number allows unambiguous record linkage at the individual level.
The Danish National Prescription Registry contains complete data on prescription medications dispensed from community pharmacies in Denmark since 1995 . It does not include medications used in-hospital. We used this Registry (codes defined in supplementary material Table 1) to identify all adult (18 years or older) first-time users of ranitidine. We defined first-time users as persons who redeemed their first prescription for ranitidine between 1996 and 2008 and had no previous prescriptions for nizatidine, because nizatidine oral solution may contain high levels of NDMA . As active comparators, we identified all adult first-time users of cimetidine or famotidine, hereafter categorized as other H2RBs, and PPIs, defined as persons who redeemed their first prescription for these drugs between 1996 and 2008 and had no previous prescriptions for ranitidine or nizatidine.
We defined the index date as the date of the first prescription. We allowed ranitidine users to redeem prescriptions for other H2RBs or PPIs, but censored users of other H2RBs and PPIs if they redeemed a ranitidine prescription during follow-up, with a 2-year lag-time. Statistics on over-the-counter use of H2RBs and PPIs have been available since 1999 in Denmark. The proportion of ranitidine defined daily doses (DDDs) sold by prescription was 84% in 1999, declining to approximately 50% in 2004–2011, and to approximately 20% in 2012–2017 (Supplementary Material Table 2). Similarly, the proportion of total H2RB DDDs sold by prescription was 83% in 1999, declining to 60% in 2004–2007, and with a further decline thereafter (data not available for 2010–2017). In contrast, 97–98% of PPIs were sold by prescription throughout the period.
We also retrieved information on use of selected drugs (defined in Supplementary Material Table 1) potentially associated with risk of upper gastrointestinal cancer and thiazolidinedione recorded before or on the index date with a 90-day look-back period. We also identified use of drugs associated with bleeding risk, including platelet inhibitors, and anticoagulants, as such drugs could alter the threshold for diagnosis of cancer.
The Danish National Patient Registry has recorded all inpatient admissions to all Danish hospitals since 1977, and hospital outpatient clinic and emergency room visits since 1995 . We retrieved all diagnoses of gastrointestinal disease indicating H2RB treatment, including Barrett’s esophagus, gastroesophageal reflux disease, and peptic ulcers, recorded before or on the index date, with a 10-year look-back period. We also retrieved data on other chronic diseases including diabetes, chronic obstructive pulmonary disease (COPD), ischemic heart disease, and alcohol-related disease (see Supplementary Material Table 3) for definitions and disease codes). The Danish Pathology Registry contains information coded using SNOMED on all specimens examined in Danish departments of pathology. From this Registry, we retrieved information on Barrett’s esophagus diagnosed before or on the index date.
We followed members of the three cohorts starting one year after their index date (to avoid reverse causation) for an incident diagnosis of cancer recorded in the Danish Cancer Registry  which has recorded cancer diagnoses since 1943, with accurate and nearly complete case ascertainment. Cancer diagnoses are recorded using International Classification of Diseases, Tenth Revision (ICD-10) codes, and codes from the ICD for Oncology, Third Revision (ICD-O-3) for topography and morphology. The outcomes of interest were esophageal cancer (any, adenocarcinoma, and squamous cell carcinoma), stomach cancer (any, proximal, distal, and unknown/several regions), hepatocellular carcinoma, and pancreatic cancer. If a subject had cancer at more than one cancer site, all were included in the analysis. We excluded subjects with a cancer diagnosis recorded before the start of follow-up, except for non-melanoma skin cancer.
The Danish Register of Causes of Death includes age, place, and cause of death (coded according to ICD-10 since 1994) . We obtained information on deaths due to cancer. We included cancer-specific deaths in the outcome definition if the cancer was not identified through the Cancer Registry.
The study was reported to the Danish Data Protection Agency through registration at Aarhus University, with exemption from informed consent (record number KEA2017-36/812).
Starting one year after the date of the first prescription, we followed cohort members until the diagnosis date of a cancer of interest for a given analysis (i.e., without censoring for other cancer types), date of death, emigration, or end of the period (31 December 2018), using an intention-to-treat approach. We censored users of other H2RBs and PPIs if they redeemed a ranitidine prescription, with a 2-year lag-time. We computed the number of events (incident cancer or cancer-specific death) per person-years at risk and the 10-year cumulative risk (with death from other causes as a competing risk) of each cancer type by exposure status.
We used logistic regression, including all available covariates, to compute propensity scores for exposure to ranitidine. In the models, we included age and index dates as splines, along with use of PPIs in the previous two years when comparing with H2RBs and use of other H2RBs in the previous two years when comparing with PPIs. We then used the propensity scores to compute stabilized inverse-probability-of-treatment (sIPT) weights. We assessed the covariate balance after weighting using standardized mean differences. The use of sIPT weighting permitted estimation of the average treatment effect through comparison of exposed vs. unexposed populations with covariate distributions resembling the distribution in the overall population.
We compared the ranitidine-exposed cohort with each of the two comparison cohorts by estimating crude and sIPT-weighted risk curves, considering death from other causes as a competing risk. We obtained the 10-year cumulative risk and calculated crude and weighted 10-year risk differences with 95% confidence intervals (CIs) estimated by bootstrapping. We calculated hazard ratios (HRs) using Cox regression analysis including the crude and sIPT-weighted observations, with 95% CIs estimated by bootstrapping. We confirmed the proportionality assumption using log-log plots.
To address treatment duration, we conducted analyses restricted to persons who redeemed at least five or at least 10 prescriptions for ranitidine, other H2RBs, or PPIs. We started follow-up one year after the fifth prescription or the tenth prescription, respectively, and we included time between the first and the fifth/tenth prescription as a variable when computing propensity scores. In addition, in the analysis of subjects who redeemed at least 10 prescriptions, we separately analyzed the first 10 years of follow-up and follow-up after 10 + years, to focus on the subpopulation with relatively high exposure and long follow-up time.
We used SAS version 9.4 (SAS Institute, Cary, North Carolina, USA) for analyses.