Sensitivity and Uncertainty Analysis of Complex Models of Disease Transmission: An HIV Model, as an Example

Abstract

Summary HIV transmission models have become very complex. The behavior of some of these models may only be explored by uncertainty and sensitivity analyses, because the structural complexity of the model are coupled with a high degree of uncertainty in estimating the values of the input parameters. Uncertainty analysis may be used to assess the variability (prediction imprecision) in the outcome variable that is due to the uncertainty in estimating the input values. A sensitivity analysis can extend an uncertainty analysis by identifying which parameters are important in contributing to the prediction imprecision (i.e., how do changes in the values of the input parameters alter the value of the outcome variable). In this paper an uncertainty and a sensitivity analysis are described and applied; both analyses are based upon the Latin Hypercube Sampling (LHS) scheme, which is an extremely efficient sampling design proposed by McKay, Conover & Beckman (1979). The methods described in this paper have not previously been applied to deterministic models of disease transmission, although these models have many characteristics in common with the risk assessment models that the strategies were designed to investigate. The utility of the LHS uncertainty and the LHS/PRC (Latin Hypercube Sampling/Partial Rank Correlation) sensitivity analysis techniques are illustrated by analyzing a complex deterministic model of HIV transmission.