Measuring effects of risk factors on cumulative incidence and remaining lifetime risk in the presence of competing risks

To analyze competing risks survival data, a commonly used approach employs a Cox model to assess the covariate effects on the cause-specific hazards treating competing risks as censoring. Subdistribution models directly assess the covariate effects on the cumulative incidence function (CIF) treating the competing risk as a second event. We study the relationship of regression coefficient estimates from the Cox model and from the subdistribution model through extensive simulations. We observe the coefficient estimates from the two models are very different when the probability of competing events is above 0.1. The magnitude and directions of the differences depend on the coefficients of both events. Generally, the subdistribution model is more appropriate and performs better. However, when the probability of competing events is low, the simpler Cox model can be used. Furthermore, when the probability of competing events is below 0.4 and the censoring rate is above 0.5, the subdistribution model displays some bias in coefficient estimates potentially due to small numbers of competing events.;Survival analyses of diseases of the elderly such as dementia often use survival age as the outcome, and account for left truncation. In such scenarios, we may consider death as a competing risk instead of censoring event. An existing SAS macro estimates the CIF of a disease and the remaining lifetime risk (LTR) conditional on survival alive and event-free to a specified age. However, the macro doesn't compare the whole LTR curves for different groups and doesn't measure the effects of risk factors on the LTR. We study a method which uses the product of the inverse probability of censoring weight and the inverse probability of left truncation weight. We further extend it to model the LTR using the survival ages as the outcome and excluding those events that occurred prior to the event-free base age.;We present two SAS macros to model the CIF using survival time as the outcome and the LTR using survival age as outcome and accounting for the left truncation and competing risks, respectively. We illustrate the use of these SAS macros through real datasets.