Modelling disease frequency measures in schizophrenia epidemiology

Abstract

Background Recent systematic reviews have compiled estimates related to the incidence, prevalence and mortality associated with schizophrenia. The aims of this study were (a) to model various frequency measures, (b) to examine the consistency of the published versus the modelled estimates, and (c) to explore the relative change in prevalence estimates after adjustments were made to incidence, remission, and mortality estimates. Methods We identified studies that provided matched incidence and prevalence estimates. We applied the DisMod software program to model incidence from observed prevalence estimates and vice versa. The accuracy of the modelled data was compared to the published data using Mann-Whitney Signed Rank tests. Finally, we conducted several 'thought experiments' to explore the impact of changing the incidence, remission, and mortality rates on prevalence estimates. Results We identified 24 matched-pairs of incidence and prevalence estimates. The distributions of modelled versus published estimates were significantly different. In 20 pairs, DisMod calculated modelled prevalence estimates that were higher than published estimates, while modelled incidence estimates were lower than published estimates in 21 pairs. In the majority of pairs, the difference between published and modelled estimates was greater than 50%. With respect to the 'thought experiments', a 25% reduction in mortality was associated with a 5-7% increase in prevalence, while 25% reduction in incidence or remission rates resulted in 18-23% and 1.2-2.4% decrease in prevalence estimates, respectively. Conclusion The consistency between published incidence and prevalence estimates of schizophrenia is poor. Models can help interrogate these inconsistencies and provide insights into the dynamics of schizophrenia epidemiology.