Assessing sensitivity to unobserved confounders in observational studies: a Bayesian approach

Hospitals and other health care organizations collect data in areas including admissions, diagnostic tests, and hospital discharges. Health services researchers use this information because it is abundant, readily available and inexpensive to access. However, this type of data differs from experimental data, which is collected for the purpose of research. With experimental data, half of participants receive a treatment and the other half receives a placebo based on random allocation, which allows researchers to measure the impact of the treatment in a manner which is unbiased. Results may be less reliable for non-experimental data when comparing groups of patients because uncontrolled circumstances can influence the outcomes in a process known as confounding. Lawrence McCandless is examining whether a new approach to using health care databases, Bayesian sensitivity analysis, can improve the accuracy and reliability of statistical studies. Lawrence is investigating the effectiveness of the approach in studying waitlists for coronary artery bypass surgery in BC. The research could suggest ways to more broadly use health care databases to study and improve the health system.