The AIDS epidemic continues to grow at an alarming rate, particularly among the poor and drug injecting population. The advent of effective anti-viral therapies ad earlier identification of HIV infected individuals is projected to increase the health care costs of AIDS. Estimating the demand for hospitalization and health care in general in the combing decades is complicated by this """"""""moving target"""""""". Current approaches to estimating the health care costs of the epidemic have used a """"""""spreadsheet"""""""", or actuarial, approach which ignores the substantial heterogeneity in the demand for medical care. Estimating demand from cross-sectional data is limited by our ability to control proximity to death since costs in the last months of life are known to differ substantially from earlier patterns. On the other hand, panel data analysis is complicated by non-random attrition (mortality) with respect to health, ad therefore with respect to the demand for health care. To estimate a lifetime demand for care requires a model which both predicts survival medical usage. We propose to develop an econometric framework for the simultaneous study of health care usage and survival in which there is an interpretable separation of the demand for care and the process of mortality. At the heart of the framework is a probabilities model for the evolution of patients' health, viewed as a latent, or unobserved, variable whose level affects both the demand for care and survival risk. Health is determined in a sequence of random draws from a probability distribution which itself may be changing over time due to the evolution of exogenous time- varying covariates and earlier medical treatment. Mortality arises when the health care use is triggered by a second threshold whose person specific value will depend upon the standard economic variables normally used in cross-sectional demand analyses. This model will be further developed and the necessary computer software written initially using a AHCPR funded cohort of AIDS patients non- randomly selected from participants in a service program. The model will be refined using the AHCPR funded AIDS Cost and Service Utilization Survey (ACSUS) which includes over 2000 HIV infected persons from hospital based clinics with early and late stage disease who were followed for 18 months. Emerging from this model development and application process will be a series of synthetic estimates of the per- person lifetime costs of HIV disease for various groups of infected individuals.
