Performance of diagnosis-based risk adjustment measures in a population of sick Australians

Australia is beginning to explore 'managed competition' as an organising framework for the health care system. This requires setting fair capitation rates, i.e. rates that adjust for the risk profile of covered lives. This paper tests two US-developed risk adjustment approaches using Australian data.

Data from the 'co-ordinated care' dataset (which incorporates all service costs of 16,538 participants in a large health service research project conducted in 1996-99) were grouped into homogenous risk categories using risk adjustment 'grouper software'. The grouper products yielded three sets of homogenous categories: Diagnostic Groups and Diagnostic cost Groups. A two-stage analysis of predictive power was used: probability of any service use in the concurrent year, next year and the year after (logistic regression) and, for service users, a regression of logged cost of service use. The independent variables were diagnosis gender, a SES variable and the

Age, gender and diagnosis-based risk adjustment measures explain around 40-45% of variation in costs of service use in the current year for untrimmed data (compared with around 15% for age and gender alone). Prediction of subsequent use is much poorer (around 20%). Using more information to assign people to risk categories generally improves prediction.

Predictive power of diagnosis-base risk adjusters on this Australian dataset is similar to that found in

Low predictive power carries policy risks of cream skimming rather than managing population health and care. Competitive funding models with risk adjustment on prior year experience could reduce system efficiency if implemented with current risk adjustment technology.