Hypoglycemia (low blood sugar) may occur during insulin therapy, and can result in adverse clinical outcomes. Hypoglycemia endpoints of interest are the incidence (number of patients with one or more episodes during a specified time period) and the rate (average number of episodes during a specified time period). The distribution of monthly hypoglycemic episodes in a diabetes population tends to be zero inflated in the sense that many patients do not experience hypoglycemia during a specified time period. Current methodology used for analysis of such data in clinical trials does not adjust for the zero inflation, and thus results in biased estimates for both probability of hypoglycemia incidence and measures of central tendency.
Let Yi be the number of hypoglycemic episodes for ith patient during a period of insulin therapy. Assuming that responses for different patients are independently distributed as Y \sim 0 with probability ${p_i}$ and Poisson(li) with probability 1 - pi. Further, the parameter log li=X t b and logit(pi)=Xtg. In this work, the covariate matrix X arises from a 2-way crossover trial design that includes sequence, period and treatment effects. The proposed Bayesian model assumes over-dispersed priors and uses Gibbs sampling to produce estimates of the parameters. The model is implemented for several clinical trials.
Learning Objectives: To share limitations of the current Statistical Methodology in a analyzing Hypoglycemia (low blood sugar) in Diabetes clinical trials. Introduce and share results of improved statistical methodology using Bayesian Modeling
Keywords: Biostatistics,
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: Eli Lilly and Company
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.