Competing Risks Analysis of the Finnish Diabetes Prevention Study.

Abstract

Clinical studies often observe one interesting event in the presence of other competing events. When both types of events can occur at any time but are only observed at clinical visits (i.e., interval censored), standard survival models may introduce bias in the estimated incidence of the interesting event over time. This can also lead to inflated relative differences between treatment groups. We developed a multi-state model for competing risks analysis of interval censored data from the Finnish Diabetes Prevention Study. The developed model predicted the participants' clinical outcomes and demonstrated that lifestyle changes significantly decreased the risk of both diabetes and death. The model showed that those who dropped out were at lower risk of developing diabetes, neglecting the assumption of independent censoring. Furthermore, the model identified the most important covariates predicting the future development of diabetes, which should be targeted for therapeutic intervention in likely clinical scenarios. These covariates are baseline BMI, HbA1c, and insulin sensitivity measurements by QUICKI for the onset of developing T2DM, baseline BMI for dropping out, and sex and age as the predictive covariates of death. Trial Registration: ClinicalTrials.gov identifier: NCT00518167.