CERTAINTY AND UNCERTAINTY IN MAPS: WHY WE NEED ERROR MAPS

Abstract

ISEE-185 Introduction: Maps are widely used to illustrate variation in disease, in disease risk factors, and in health delivery indicators. Such maps may be descriptive of disease distribution, showing geographical variation of disease prevalence or incidence (e.g. cancer, malaria, HIV), or descriptive of geographical variation of exposures (e.g. climate, facility based health delivery, community based interventions) or part of studies investigating the relationship between disease and risk or intervention. In many cases the quantities displayed in map form, whether based on sample surveys, health information systems, or outputs from models, are subject to considerable uncertainty. Explicit mapping of the uncertainty of mapped quantities is essential for the correct interpretation of maps. Yet this information is rarely provided. Methods: We use data from three unrelated studies (primary health facility indicators of management of sexually transmitted infections in South Africa, district level variation of malaria incidence in Malawi, and malaria prevalence of infection in relation to an intensive intervention in an equatorial region) to illustrate how maps can lead to interpretation and mis-interpretation of results. We apply Bayesian autoregressive modelling techniques to produce smoothed maps and to derive uncertainty maps. We briefly outline the methodology employed. Results: Unsmoothed, smoothed and error maps from the three studies will be presented and discussed. Conclusion: Maps that are processed using only GIS approaches can be highly misleading. Uncertainty in mapped quantities should be made explicit by accompanying error maps. GIS tools generally ignore this challenge.