Estimation of multi-treatment effects from observational data with application to diabetes mellitus

Date of Approval: 
2016-05-17 00:00:00
Lay Summary: 
Background: Randomized clinical trials (RCTs) are the gold standard for identifying optimal treatment options. However, RCTs usually compare two treatments on a small homogenous group of patients. Observational data under certain assumptions are a possible tool to identify the optimal treatment among several in large cohorts. Objective: To provide guidance on statistical methods aimed at estimation of optimal treatment among three or more treatments. Apply this guidance to examine the effects of medications other than insulin used to control blood sugar levels on the occurrence of heart attack, stroke, or hospitalization for heart failure in patients with type 2 diabetes mellitus. Methods: We will use our knowledge from extensive simulations that were based on realistic settings to examine in a cohort of CPRD patients with type 2 diabetes mellitus whether non-insulin regimens differentially affect the occurrence of the aforementioned adverse events in. Our method will adjust for demographic and clinical characteristics that affect the use of the different regimens. Outcomes: We will provide researchers with guidelines for choosing the most appropriate statistical method when comparing more than three treatments. In addition, we will identify whether different non-insulin regimes that involve multiple agents are responsible to increased risk of adverse events.
Technical Summary: 
We will examine the performance of the methods by estimating effects of different antihyperglycemic drug regimens (mono- or poly-therapy) on the occurrence of myocardial infarction, stroke, and hospitalization for heart failure for diabetes mellitus patients. The design will consist of generalized propensity-score (GPS) matching based on pre-treatment covariates, with the primary method chosen via simulation work. This "as-matched" analysis based on the original exposure classifications has the advantage of preserving the randomization-like features of the GPS analysis. In this analysis, we will calculate incidence rates of acute myocardial infarction and use Kaplan-Meier plots to depict the cumulative probability of event-free time among the GPS matched cohorts (i.e., more than two).
Health Outcomes to be Measured: 
Myocardial infarction, stroke, and hospitalization for heart failure
Application Number: 

Roee Gutman - Chief Investigator - Brown University
Roee Gutman - Corresponding Applicant - Brown University
David Dore - Collaborator - Optum
Robert Smith - Collaborator - Providence Veterans Administration Medical Center
William Hiatt - Collaborator - University Of Colorado

HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation;Practice Level Index of Multiple Deprivation