Assessing the impact of aggregating disease stage data in model predictions of human African trypanosomiasis transmission and control activities in Bandundu province (DRC)

Since the turn of the century, the global community has made great progress towards the elimination of gambiense human African trypanosomiasis (HAT). Elimination programs, primarily relying on screening and treatment campaigns, have also created a rich database of HAT epidemiology. Mathematical models calibrated with these data can help to fill remaining gaps in our understanding of HAT transmission dynamics, including key operational research questions such as whether integrating vector control with current intervention strategies is needed to achieve HAT elimination. Here we explore, via an ensemble of models and simulation studies, how including or not disease stage data, or using more updated data sets affect model predictions of future control strategies.Author summary: Human African tryposonomiasis (HAT), also known as sleeping sickness, is a parasitic disease with over 65 million people estimated to be living at risk of infection. Sleeping sickness consists of two stages: the first one is relatively mild but the second stage is usually fatal if untreated. The World Health Organization has targeted HAT for elimination as a public health problem by 2020 and for elimination of transmission by 2030. Regular monitoring updates indicate that 2020 elimination goals are likely to be achieved. This monitoring relies mainly on case report data that is collected through medical-based control activities—the main strategy employed so far in HAT control. This epidemiological data are also used to calibrate mathematical models that can be used to analyse current interventions and provide projections of potential intensified strategies. We investigated the role of the type and level of aggregation of this HAT case data (staging data and truncated data) on model calibrations and projections. We highlight that the lack of detailed epidemiological information, such as missing stage of disease or truncated time series data, impacts model recommendations for strategy choice: it can misrepresent the underlying HAT epidemiology (for example, the ratio of stage 1 to stage 2 cases) and increase uncertainty in predictions. Consistently including new data from control activities as well as enriching it through cross-sectional (e.g. demographic or behavioural data) and geo-located data is likely to improve modelling accuracy to support planning, monitoring and adapting HAT interventions.