Methods and issues in the projection of population health status

The discussion of strategies for forecasting health status changes in human populations often becomes immersed in efforts to utilize simple projection strategies that will produce crude projections. The motivation behind this effort is that simple projection strategies have limited data requirements and the crude projection strategies will be, in some ill-defined sense, robust (i.e., insensitive to assumptions). Actually there is a wide range of projection tools available. It seems appropriate to appraise the nature and attributes of each when considering the uses to which the projections will be put. For example, simple models are not necessarily more robust than more sophisticated procedures, especially for longer term temporal projections. Clearly we have many examples in developed countries where the use of simple actuarial or demographic projections has underestimated the true cost of a health programme by factors of 200-300%. The reason why the failures of such simple projection efforts become so rapidly manifest is that the programmes, once implemented, are expanded to meet the population's needs. In projecting only health services or utilization one has nearly a self-fulfilling prophecy--that resource constraints or the actual organization of the programme will directly determine the course of the level and mixture of health services consumption. Therefore failure to base the projections on a detailed model of underlying population needs leads in such cases to grossly inaccurate results. Clearly, projecting a population's health needs requires even more data than projecting health service requirements. Such information constraints require the use of a model to organize data from multiple objective and subjective sources, and to reflect the best scientific understanding of the processes involved. This article briefly discussed the application of 2 such models. One was designed for the analysis of discrete state health changes using population and vital statistics data, the other described both discrete and continuous changes using data from longitudinally followed community populations. One is designed to work only with detailed aggregate data with heavy inputs from scientific experts; the other deals with relatively information-rich measurements. Both can be modified on the basis of expert judgement to deal with simulations of a multiplicity of possible interventions. Both appropriately calculate the relative costs and benefits of select health initiatives.(ABSTRACT TRUNCATED AT 400 WORDS)