Competing Risks Analysis of the Finnish Diabetes Prevention Study

Figures

• 

  FIGURE 1. The five‐state model developed for analyzing the outcome among participants in the Finish Diabetes Prevention Study and its non‐interventional follow‐up. The model captures the dynamic transitions between the following distinct states: IGT (State 1)—participants with impaired glucose tolerance not yet diagnosed with Type 2 diabetes, T2DM (State 2)—participants diagnosed with Type 2 diabetes within the intervention or follow‐up period, Drop‐out (State 3)—participants without a diabetes diagnosis who decided to leave the study, Drop‐out with T2DM (State 4)—participants who decided to leave the study and then were diagnosed with Type 2 diabetes, Death (State 5)—participants who died.

• 

  FIGURE 2. Visual Predictive Checks (VPCs) divided into control and intervention groups, where the solid line represents the observed proportion of participants in each state over time and the shaded area is the 95% confidence interval of the median percentile of model‐simulated data. State 1—participants with impaired glucose tolerance without Type 2 diabetes, State 2—participants diagnosed with Type 2 diabetes within the intervention or follow‐up period, State 3—participants without diabetes who left the study, State 4—participants who left the study and were then diagnosed with Type 2 diabetes, State 5—participants who died.

• 

  FIGURE 3. Forest plots illustrating covariate effects on the different hazard rates (λij). The reference patient is a 60‐year‐old male in the control group with the following characteristics: baseline Body mass index (bBMI) = 30.48 kg/m2, glycated hemoglobin at the previous visit (pHbA1c) = 5.6%, and quantitative insulin sensitivity check index at the previous visit (pQUICKI) = 0.316. Closed symbols denote point estimates, and horizontal bars represent 95% CI based on 200 samples of model parameters from the variance–covariance matrix. The shaded vertical band indicates the 0.8–1.25 interval. Values are expressed as fold‐change [95% CI] relative to the reference patient. Due to identical hazard rates (λ35=λ15 and λ45=λ25), λ35 and λ45 are omitted for redundancy. No covariates were included on λ34.

• 

  FIGURE 4. Kaplan–Meier graph of model‐simulated (left panel) and observed data (right panel) for Type 2 diabetes, displayed with 95% confidence intervals. In the graph, developing diabetes is assumed to be independent of drop‐out. The lower left panel of percentage at risk is the median value across all simulations.