Date of Award


Document Type

Campus Access Dissertation


Epidemiology and Biostatistics



First Advisor

James R Hebert


Angiotensin receptor blockers (ARBs) are commonly prescribed drugs that have proven benefits among patients with cardiovascular and diabetic diseases. Recently a concern of modest increase in the risk of solid cancers among ARB users was reported. This became controversial, as many experts argued that it was not supported by current understanding on biological plausibility. There is a need to conduct individual-level analysis on the relationship between ARB and cancer to discern both cause and the magnitude of effect, if present. Pharmaceutical phase I, II and III research conducted by drug companies focus on efficacy and early adverse events; such research is not powered to detect long-term adverse events. Regulatory authorities such as the FDA rely on information received via post-marketing safety surveillance, reports generated by independent investigators and other voluntary reporting systems. Unfortunately, such reports are subject to biases resulting from selective reporting standards and non-standardized practices, with a net consequence of low sensitivity. Publication bias commonly plagues medication safety reporting. Investigators are more likely to publish in scientific journals results that show safety concerns and suppress findings with negative results.

In the era of electronic medical records, the volume of comparative effectiveness research on medication safety using real-world data is increasing. Given this volume, although the probability of Type-1 error for a single project is constant, there is now higher likelihood that a publication that reports a significant relationship between exposure and outcome has arisen after suppression of many individual findings of non-significant relationship, aided by bias of selective publication. Such reports, although might have robust internal validity, will most definitely have extremely poor external validity. Current research methods, that are guided by the principle of α=0.05 (Type-1 error), are designed to prove an association but not disprove it. So, once a type-1 error of deleterious association has been reported, it is difficult to call into question this proven relationship even after demonstrating the absence of such an association in an independent report with higher-power. This has led to several initiatives to develop best practice-based approaches in comparative effectiveness research, such as that recommended by the International Society for Pharmacoeconomics and Outcomes Research (ISPOR).

For this dissertation we have conducted a medication safety study designed to help confirm or refute a concern raised regarding the relationship between the ARB class of drugs and cancer. Three primary cancers were evaluated as primary outcome measures: hepatocellular cancer, colon cancer and prostate cancer. For our study we utilized the facilities and data resources provided by the Department of Veterans Affairs (VA). The VA approved this research and provided us access to data on nearly 14 million unique veteran patients, including patient demographics and medical records through the Veterans Affairs Informatics and Computing Infrastructure (VINCI) program. This is the only national individual-level data source in the United States that provides researchers access to detailed longitudinal clinical, pharmaceutical, pathological and laboratory data at the individual level. Our approach in this project is to incorporate a "best practice" approach to conduct high quality research. The goal is to produce results that are easily interpretable, generalizable, and have the highest internal and external validity. We conducted analyses using the theory of counterfactuals, the results of which are expected to be closest to those results from a prospective randomized experiment designed to answer the same research question.

The project was conducted using an innovative design that uses multiple "independent" cohorts, the findings of each cohort being mutually exclusive of each other. Using such mutually exclusive cohorts, we were able to significantly reduce the collective probability of chance finding and thus reduce the overall false positive rate. Our design allows for an extremely high power to detect any true association if it truly exits. If our project concludes no association or an association in the opposite direction, this novel approach would strongly refute the initial report of increased cancer risk and suggest that the original finding was indeed a Type 1 error, i.e. false positive. By using such standardized, reproducible and transparent analytical approach on real world data, we believe we are well positioned to provide strong evidence that would help address similar medication safety controversies in the future.