Modelling to infer the role of animals in gambiense human African trypanosomiasis transmission and elimination in the DRC

Gambiense human African trypanosomiasis (gHAT) has been targeted for elimination of transmission (EoT) to humans by 2030. Whilst this ambitious goal is rapidly approaching, there remain fundamental questions about the presence of non-human animal transmission cycles and their potential role in slowing progress towards, or even preventing, EoT. In this study we focus on the country with the most gHAT disease burden, the Democratic Republic of Congo (DRC), and use mathematical modelling to assess whether animals may contribute to transmission in specific regions, and if so, how their presence could impact the likelihood and timing of EoT. By fitting two model variants—one with, and one without animal transmission—to the human case data from 2000–2016 we estimate model parameters for 158 endemic health zones of the DRC. We evaluate the statistical support for each model variant in each health zone and infer the contribution of animals to overall transmission and how this could impact predicted time to EoT. We conclude that there are 24/158 health zones where there is substantial to decisive statistical support for some animal transmission. However—even in these regions—we estimate that animals would be extremely unlikely to maintain transmission on their own. Animal transmission could hamper progress towards EoT in some settings, with projections under continuing interventions indicating that the number of health zones expected to achieve EoT by 2030 reduces from 68/158 to 61/158 if animal transmission is included in the model. With supplementary vector control (at a modest 60% tsetse reduction) added to medical screening and treatment interventions, the predicted number of health zones meeting the goal increases to 147/158 for the model including animal transmission. This is due to the impact of vector reduction on transmission to and from all hosts.

: PRIME-NTD criteria fulfillment. How the NTD Modelling Consortium's "5 key principles of good modelling practice" have been met in the present study.

Principle
What has been done to satisfy the principle?
Where in the manuscript is this described?

Stakeholder engagement
This study is one in a series of modelling analyses for the DRC. The modelling team lead the simulation and analysis work guided by members of the national sleeping sickness control programme in the DRC (PNLTHA-DRC) -coauthors E Mwamba Miaka and C Shampa. PNLTHA-DRC have supported model development through the context of how the human case data used in the study were collected and how this has changed over time (via in-person meetings, online meetings and by email). The GUI (and several variants of it) was designed in conjunction with PNLTHA-DRC to improve communication of the modelling outputs to non-modellers. It has been refined through various in-person meetings with different collaborators with the goal of providing understandable, policy-relevant outputs as well as scientific communication; over 20 nonmodellers have had opportunities to interact with and provide feedback on the GUI during development.

Complete model documentation
Full model fitting code and documentation is available through OpenScienceFramework (OSF). The model is fully described in the main text and SI.
See Materials and Methods section in the main text, Supplementary Information (file S1) and at OSF (https://osf.io/3xadf/)

Complete description of data used
The original data and how we aggregated the data for fitting were described in detail in the SI and also in the first model fitting paper for the DRC from our group [2]. Aggregate data can be viewed next to model fits in our GUI.

Communicating uncertainty
Structural uncertainty: The focus of this paper is to assess statistical support for two alternative gHAT model variants -one with and one without animal transmission. In previous studies eight variants were considered [5,3] and there was most (and similar) support for "Model 4" (without animal transmission) and "Model 7" (with animal transmission) which are examined here. We show both fits and projections side by side to show model differences driven by this structural uncertainty Structural uncertainty: Materials and Methods section in main text, Figs 2, 3, 6-8.
Parameter uncertainty: Parameter distributions are estimated by fitting to data. All model fits and projections include and propagate parameter uncertainty and include it in visual representations (either box and whisker plots or as probabilities, as appropriate) Parameter uncertainty: All main text figures (expect Fig 1), Supplementary Information (S1) Figs B-E and in the GUI (https://hatmepp.warwick.ac.uk/ animalfitting/v1/) Table A -continued from previous page  Principle What has been done to satisfy the principle?

Continued on next page
Where in the manuscript is this described?
Prediction uncertainty: All predictions incorporate structural and parameter uncertainty. Furthermore, whilst we show only two main strategies in the main text, additional future interventions which are a sensitivity analysis on active screening coverage and vector control effectiveness are shown in our GUI.

Testable model outcomes
Previous versions of this model have undergone validation exercises (data censoring) to examine the robustness of the predictive ability of the model [3,4,6]. Whilst this was not performed here, the model is an updated version of those that have undergone validation, with updates based on critical review of model fits as data for different regions or time periods is included (notable refinements included here were made in the related analysis of Crump et al. [2]). As more years' data become available in the future, the model outputs shown here will be able to be compared to reported active and passive case data to test model predictions.