Multivariate estimation of cumulative incidence, prevalence, and morbidity time of a disease when death is likely

Competing risk of death from other causes before developing the outcome of interest is a common phenomenon in clinical settings. In a previous article, we developed the "sandwiching method" as one approach to estimate disease incidence and morbidity time for populations at a high risk of death from other causes. In addition to its computational simplicity, the sandwiching method is also relatively assumption free. This article extends the original sandwiching method to incorporate patient characteristics into the estimation by using Cox's proportional hazards model. Data from the Johns Hopkins Hospital AIDS service were used to illustrate this extended method. The importance of estimates was discussed in the context of planning health care needs.