Evaluating health policy under uncertainty: An application to early warning systems

The usefulness of predictive information for improving public health is mitigated by large gaps in our understanding of the economic costs and benefits of this information. Despite the potential for reducing the incidence of communicable disease, uncertainty in forecasts and in the effectiveness of interventions often results in warnings that are not heeded. In this dissertation we develop a method for providing policy guidelines when little is known about a policy's cost or effectiveness. Using this methodology, we then estimate the benefits of an early warning system for a specific public health threat: epidemics of the mosquito-borne disease dengue fever in Puerto Rico.;In the first section, economic impacts are developed through Monte Carlo simulations based on data from Puerto Rico on disease incidence and direct and indirect economic impacts. Disease transmission is reduced by mosquito abatement. Since virtually no data is available on the cost and effectiveness of mosquito abatement policies, we develop an alternative to traditional cost-effectiveness measures. We characterize a successful intervention by determining the bounds of cost and effectiveness at which a mosquito control effort is superior to a policy of no abatement. These bounds on policy cost and effectiveness are estimated through an algorithm that determines the presence of stochastic dominance for a range of risk aversion coefficients.;The second section examines the changes in expected disease distributions that occur with an early warning system. We consider four types of predictive information from an early warning system: no predictive information; predictive information based on off-seasonal reporting rates; and two improvements to predictive information. A Bayesian approach transforms the original distributions of economic impacts into distributions conditional on signals from the early warning system. Using the methodologies of section one, we estimate the changes to the bounds on cost and effectiveness that occur with predictive information.;In the third and final section, we consider the possibility of loss aversion among decision-makers. A decision-maker exhibiting loss averse preferences may prefer a riskier policy which offers a chance of low expenditures to a policy which requires some expenditures but represents overall lower expected costs.