European Journal of Operational Research vol:24 pages:449-457
We derive the proper form of the Akaike information criterion for variable selection for mixture cure models, which are often fit via the expectation-maximization
algorithm. Separate covariate sets may be used in the mixture components. The selection criteria are applicable to survival models for right-censored data with multiple competing risks and allow for the presence of an insusceptible group. The method is illustrated on credit loan data, with pre-payment and default as events
and maturity as the insusceptible case and is used in a simulation study.