No Association Between Chronic use of Ranitidine, Compared With Omeprazole or Famotidine, and Gastrointestinal Malignancies

Yeseong D. Kim; Jiasheng Wang; Fahmi Shibli; Kamrine E. Poels; Stephen J. Ganocy; Ronnie Fass

Disclosures

Aliment Pharmacol Ther. 2021;54(5):606-615. 

In This Article

Materials and Methods

Database Description

Explorys (IBM ®Explorys) is a private, cloud-based data store which provides aggregated and deidentified electronic medical record data from 360 and over 73 million individual patients across the United States. This software places a healthcare gateway server behind the firewall of each participating healthcare institution, and precludes the investigators from seeing institution or patient-level data. The data are real-time updated at least once every 24 hours and retrieved from a variety of health information systems including billing inquiries, electronic health records (EHRs), and laboratory systems, which is then uploaded onto a data grid. This grid is accessible via web application and allows the user to search and analyse the aggregated, standardised, normalised, and de-identified population level data.[16]

The data in Explorys are exported according to common clinical standards; diagnoses are derived from the International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM), and mapped into Systemised Nomenclature of Medicine—Clinical Terms (SNOMED-CT). Procedures and findings are also directly mapped into the SNOMED-CT hierarchy. Laboratory tests are mapped onto Logical Observation Identifiers Names and Codes (LOINC), and prescription drugs are mapped onto RxNorm. This standardisation of medical linguistics facilitates rapid searching, indexing, and cohort generation. For Health Insurance Portability and Accountability Act (HIPAA) compliant statistical de-identification purposes, population counts are reported to the nearest 10, or as less than 10 if between 0 and 10.[17]

Study Design

Population-based data retrieval using Explorys was performed to identify 3 cohorts of adult patients at or above the age of 18 who were prescribed either ranitidine, famotidine, or omeprazole (Figure 1). Each cohort was subsequently subdivided into subcohorts to identify the incidences of first-time diagnoses of 5-gastrointestinal malignancies; oesophageal, gastric, hepatocellular, pancreatic, and colorectal. Incidence data for each cancer were queried up to 10 years (2009–2018) in 1-year intervals, with the first-time cancer diagnosis indexed at a minimum of 1 year following initiation of each anti-reflux therapy. This one year lag-time was incorporated to mitigate the effects of immortal time bias, as acute NDMA exposures of less than 1 year is unlikely to affect an individual's risk of cancer formation.[18] Additionally, the effects of protopathic bias were mitigated by excluding patients with prior history of any malignancy, and by incorporating the aforementioned 1 year lag-time. Furthermore, it was ensured that each anti-reflux medication was continued until the initial diagnoses of each gastrointestinal malignancy by querying cohorts only with a concurrent prescription of given anti-reflux therapy at the time of initial cancer diagnosis. Diagnostic terms based on Explorys SNOMED-CT codes that were used included "malignant tumour of oesophagus," "malignant tumour of stomach," "malignant neoplasm of liver," "malignant tumour of pancreas," and "malignant tumour of large intestine." Pharmacological nomenclature used based on RxNorm were "ranitidine," "Zantac," "famotidine," "Pepcid," "omeprazole," and "Prilosec." This study was exempt by the MetroHealth Institutional Review Board (IRB). All authors had access to the study data and reviewed/approved the final manuscript.

Control and Confounding Factors

Several measures were undertaken to mitigate the effect of confounding factors. First, patients with prior diagnosis of any gastrointestinal malignancy were excluded. Patients who were prescribed more than one anti-reflux medication, in any combination of the three medications studied, were excluded from the study. Parallel search parameters and retrieval criteria were used to query patient cohorts started on either ranitidine, famotidine, or omeprazole. Famotidine was used as a same-class medication comparator, and omeprazole as a different-class medication comparator. Both drugs are widely used to treat acid-peptic disorders, and are not known to contain NDMA or cause malignancy, serving as controls to maintain internal validity of the study design, search strategy, and data from the database. Adjusted odds ratios (ORs) were calculated by multivariable logistic regression analysis accounting for three common risk factors for each type of cancer studied. For oesophageal cancer, tobacco use, alcohol use, and Barrett's oesophagus (BO); for gastric cancer, atrophic gastritis, tobacco use, and obesity; for pancreatic cancer, tobacco use, diabetes mellitus (DM), and obesity; for liver cancer, cirrhosis, obesity, and chronic viral hepatitis; for colorectal cancer, obesity, DM, and IBD were controlled for. A separate multivariable logistic regression model was performed accounting for demographic factors including age above or below 65, male or female gender, Caucasian or African American race, and the most significant risk factor from the common risk factor analysis model. Two separate regression models were run so the Explorys cohort sizes would return as non-zero integers. Continuous variables were subdivided into categorical variables. Demographic terms based on SNOMED-CT codes that were used were "male," "female," "adult between 18 and 65 of age," "senior above 65 of age," "Caucasian race," "African American race." Diagnostic terms based on SNOMED-CT codes for confounders were "tobacco use," "current drinker of alcohol," "Barrett's oesophagus," "atrophic gastritis," "obesity by BMI > 30," "diabetes mellitus," "cirrhosis of the liver," "chronic viral hepatitis," and "inflammatory bowel disease."

To validate the data generated by the database, we generated five separate cohorts to examine the annual incidence rates of each gastrointestinal malignancy studied from 2009 to 2018 for comparison to known values reported in the literature.

Statistical Analysis

All demographic and incidence data in this study queried from Explorys were categorical and thus were presented as counts and percentages. Multivariable logistic regression analysis on aggregated data was calculated to assess risk, in the form of ORs, of gastrointestinal malignancies for ranitidine, famotidine, and omeprazole usage in patients while adjusting for tobacco use, alcohol use, DM, obesity, cirrhosis, IBD, atrophic gastritis, and BO as covariables in predetermined combinations, as previously described, depending on the cancer queried. A separate multivariable logistic regression analysis was performed adjusting for age, gender, race, and highest-weight common cancer risk factor in predetermined combinations as previously described. Chi-squared test was employed to analyse the significance of differences in baseline characteristics among the cohorts. All tests were two-sided. Using Bonferroni correction for multiple comparisons, statistical significance was established at P < 0.002 throughout the study. R version 3.6.1 (R Foundation for Statistical Computing) was used to compute all statistical tests.

processing....