Risk Models for Individual Decision Making in Medicine
by
Malcolm Hudson
Macquarie University
Coauthors: John Simes

We develop a methodology for providing decision aids which integrate clinical trial evidence with other information relevant to the decision between alternative treatments. The better treatment option will depend on the health experience that follows it, together with the individual valuation of the alternative realizations (histories).

Since any future course of health states or health events is uncertain we cannot know exactly what history will eventuate. Our approach therefore will be to simulate, selecting many possible health state histories that have occurred in similar individuals on the same treatment in clinical trials and determining corresponding simulated histories for the new individual. The histories simulated under different treatment options are then valued by the individual and statistical comparisons made of alternative treatment options on this basis. Besides survival benefits, other factors that may have importance depending on individual circumstances include age and risk profile attributes. The risk profile allows a net valuation accounting for benefits, costs and harms. The risk profile attributes will depend on: . how trade-offs between different dimensions in outcomes are valued by doctor and patient; e.g. the individual's willingness to trade-off quality of life or other treatment effects for increased survival. . the preferences between alternative health states; . willingness to entertain the possibility of prolonged low quality states in anticipation of longer term benefits in health; . the discounting applied to the future as against the present. The uncertainty as to future health lead to uncertainty in the valuation which is best allowed for in repeated simulations of future histories (or realizations) of health states that may apply to the new individual. The variability in valuation may lead to different treatment preferences among patients.

In this presentation/paper we shall introduce the decision analysis framework for risk models employing trials data. Further we will consider the sensitivity of the decision about treatment to the absolute level of risk specified. Estimates of absolute risk for the individual treatment may be obtained from a matching group of patients in Clinical Trials databases. The effectiveness of such matched estimation will be evaluated. Finally, this analysis will be extended to value health state histories employing Markov models (Q-TWiST) of transitions in health states, where we examine the requirements for effective valuations of treatment alternatives for the individual.

Date received: September 27, 2001