Fig 1.
Data for Mosango health areas.
(A) A map of the location of Mosango, in the Kwilu province and Bandundu Sud coordination of the DRC (coordinations are the large geographic units, similar to provinces, for the organisation of gHAT activities). The map is divided into health area units, where possible, and health zone units otherwise. The Kwilu province is shown by the thick black border, the Bandundu coordination is shown in green and Mosango is dark green. (B–E) The smaller maps to the right are the distribution of the population (data provided by UCLA), the number of people screened (2000–2020) and the number of active and passive cases (2000–2020) across the health areas of Mosango (extracted from the WHO HAT Atlas [3]). The largest number of active and passive cases is 220 and 208 respectively, both in the health area of Kinzamba II. Shapefiles used to produce these maps were provided by Nicole Hoff and Cyrus Sinai under a CC-BY licence (current versions can be found at https://data.humdata.org/dataset/drc-health-data).
Fig 2.
The purple and blue boxes denote the low-risk and high-risk human infection states, respectively. Low-risk people, who randomly participate in active screening are given the subscript H1 and high-risk people, who do not participate, have the subscript H4 (this notation is used to align with previously published versions of this model). Tsetse dynamics are given by red boxes. All rates given by arrows are exponentially distributed, except where stated as Erlang. A proportion fA = 1 − fH of tsetse bites will be taken on non-human animals, but this does not contribute to the infection dynamics. This figure is adapted from Crump et al. [8] under a CC-BY licence.
Table 1.
Notation, a brief description of the parameter, either the fixed value used or the prior distribution used in fitting, and a source for the fixed parameters.
Table 2.
Events table for the stochastic tau-leaping model for humans and deterministic ODEs for the tsetse dynamics.
Fig 3.
(A) Estimates of the negative log posterior probability (NLPP) given by different numbers of particles used by our particle filter. Each box plot consists of 10,000 values consisting of five stochastic realisations from a posterior of 2,000 values. (B) Iterations of example pMCMC chains. Two independent chains and values for the NLPP for 50 particles. (C) The between-chain convergence diagnostic as the number of iterations increases after the burn-in. (D) The effective sample size (ESS) after the burn-in is taken to be the minimum value of the autocorrelation of the thinned sample for all the fitted parameters. Dashed lines on sub-panels C and D show where thresholds for completion lie. Background colours on sub-panels b–D indicate the phase of the pMCMC: red for the (relatively short) transient phase, blue for the adaptive phase, and green for the sampling phase.
Fig 4.
Model fitting outputs in example health areas.
(A and E) Data for the number of people actively screened and the active and passive cases reported are displayed as solid black lines for two health areas: Kinzamba I and Kinzamba II (labelled as A2 and A3 respectively in this analysis). (B–D and F–H) The modelling outputs for active and passive cases and the number of new infections in each year are given as coloured box plots. The median value of the box plots is shown as a white line, with outer boxes and whiskers representing 50% and 95% credible intervals respectively. No data line is shown for new infections, since this cannot be directly observed. Maps of the health area locations with the Mosango health zone are shown in the top right of each column. Shapefiles used to produce these maps were provided by Nicole Hoff and Cyrus Sinai under a CC-BY licence (current versions can be found at https://data.humdata.org/dataset/drc-health-data).
Fig 5.
Geographical distribution of model parameters in Mosango health zone.
The geographical distribution of the median values for: (A) the basic reproduction number R0 and (B) the proportion of stage 2 cases reported from passive screening u(1998) is shown in maps of the health zone. Each health area is labelled A1–A16, with the full posterior distribution given in histograms on the right. The median value is shown by a horizontal coloured line, aligned with the color scale of the map. The prior distribution is shown on the histograms as a solid black line, although with a relatively low probability density in the displayed region for the R0 histograms. It is the same for each health area. The names of health areas A1–A16 are given in S1 Text. Shapefiles used to produce this map were provided by Nicole Hoff and Cyrus Sinai under a CC-BY licence (current versions can be found at https://data.humdata.org/dataset/drc-health-data).
Fig 6.
Model fitting outputs for Mosango health zone.
(A–D) Data for (A) the number of people actively screened and (B) the active and (C) passive cases reported is displayed as solid black lines. Since new infections are not observed there is no data in panel D. The coloured box plots show annual model outputs for different fitting approaches, where blue boxes show the deterministic health zone model, red shows the stochastic health zone model, and green shows the aggregated stochastic health area models. The median value of the box plots is shown as a white line, with box plot whiskers representing 95% credible intervals. (E) The bottom panel shows the probability of EoT from 2000–2050 assuming a continuation of interventions.
Fig 7.
Model fitting outputs for Kinzamba II (A3) health area.
(A–D) Data for the number of people actively screened and the active and passive cases reported is shown by solid black lines. Since new infections are not observed there is no data in panel D. Coloured box plots show annual model outputs for different methods for both model fitting and model projections. The light blue boxes show the deterministic model fitting with deterministic projections, the light purple shows the deterministic model fitting with stochastic projections, and the light red shows the stochastic model fitting with stochastic projections. The median value of the box plots is shown as a white line, with box plot whiskers representing 95% credible intervals. (E) The bottom panel shows the probability of EoT from 2000–2050 assuming a continuation of interventions.
Fig 8.
Elimination of transmission (EoT) in Mosango.
(A) Probabilities of EoT for health area by a given year are shown by coloured lines in the top panel, with aggregated health area results shown as a black line and separate health zone model results shown by a dotted grey line. (B–D) The below panel shows maps of Mosango with median expected years of EoT given by colour. The three maps show three approaches for displaying EoT: (B) for each individual health area, (C) for the health zone using the health area models, and (D) for the health zone using the health zone model. Shapefiles used to produce this map were provided by Nicole Hoff and Cyrus Sinai under a CC-BY licence (current versions can be found at https://data.humdata.org/dataset/drc-health-data).