Evaluating elimination thresholds and stopping criteria for interventions against the vector-borne macroparasitic disease, lymphatic filariasis, using mathematical modelling

Abstract

We leveraged the ability of EPIFIL transmission models fit to field data to evaluate the use of the WHO Transmission Assessment Survey (TAS) for supporting Lymphatic Filariasis (LF) intervention stopping decisions. Our results indicate that understanding the underlying parasite extinction dynamics, particularly the protracted transient dynamics involved in shifts to the extinct state, is crucial for understanding the impacts of using TAS for determining the achievement of LF elimination. These findings warn that employing stopping criteria set for operational purposes, as employed in the TAS strategy, without a full consideration of the dynamics of extinction could seriously undermine the goal of achieving global LF elimination.

Fig. 1. The SIR BM-based model fits to mf prevalence baseline data for two low-prevalence sites (DokanTofa and Piapung), two medium-prevalence sites (Missasso and Kirare), and two high-prevalence sites (Peneng and Dozanso).

Red circles indicate the observed age-prevalence data in the case of Dokan Tofa and Piapung, whereas age-stratified prevalence data derived from observed overall community baseline prevalences are shown for linear, plateau and convex age patterns of LF infection by the black, green, and blue curves, respectively, in Missasso, Kirare, Peneng, and Dozanso. Gray lines depict the model predictions for each site. The age-stratified and overall Monte Carlo p-values for the fits to baseline infection data for each of the six study sites were evaluated and are provided in Supplementary Table 3 in Supplementary Information. These show that apart from Piapung and Kirare, good fits were obtained for all other villages. Error bars indicate the 95% confidence intervals for each data point.

Fig. 2. Forecasting impacts of the MDAs carried out in each site on changes in mf prevalence for two low-prevalence sites (DokanTofa and Piapung), two medium-prevalence sites (Missasso and Kirare), and two high-prevalence sites (Peneng and Dozanso).

Red circles indicate the observed data and gray lines indicate the model predictions. The overall Monte Carlo p-values for the fits of post-MDA data for the six study sites were evaluated and indicated good fits in all sites (provided in Supplementary Table 3 in Supplementary Information). Error bars indicate the 95% confidence intervals of the data points.

Fig. 3. Bar diagram of recrudescence probability for two low-prevalence (DokanTofa, Piapung), two medium-prevalence (Missasso, Kirare), and two high-prevalence (Peneng, Dozanso) study sites.

a After 5 years of stopping annual MDA following breaching of the 1% mf threshold, and b after 5 years of stopping annual MDA following breaching of the 95% EP threshold as shown in Tables 1 and 2 and Supplementary Tables 4 and 5. Figures in red above each bar indicate the predicted ranges in mf prevalence while figures in black above each bar indicate the mf prevalence predicted as a proportion of baseline mf prevalence.

Fig. 4. Impact of annual MDA in DokanTofa, Nigeria.

a When annual MDA was continued to cross the 1% community mf threshold on mf prevalence in 6–7 years old children, b when annual MDA was continued to cross 1% mf threshold in the 40–70 years old population, c when annual MDA was continued to cross the corresponding 95% EP mf threshold in 6–7 years old children, and d when annual MDA was continued to cross the 95% EP mf threshold in the 40–70 years old population. Light gray lines indicate the model predictions with colors changing to black when trajectories begin to ascend and to dark gray when they move toward elimination. Numbers given in the black and dark gray colors indicate the probabilities (as percentages) of infection recrudescence and elimination respectively. Blue solid, blue dotted, and black solid horizontal lines indicate the 1% mf threshold, 0.5% mf threshold and model predicted 95% EP mf threshold, respectively, whereas blue dotted and blue solid vertical lines represent the time periods for crossing the 1% mf threshold and the 5 year post-MDA stoppage duration (the TAS period), respectively.

Fig. 5. Predicted transient dynamics in mf prevalence using annual MDA, biannual MDA, and annual IDA and the 95% mf prevalence EP threshold for Piapung.

Red curves indicate the model fits with transient behavior (curves having no significant positive or negative slopes) whereas the gray curves indicate the model fits having significant positive or negative slopes for different time periods (5, 10, 15, and 20 years) after stopping MDA following breaching of the 95% EP mf threshold. The figures in percentage show the percentage of the curves describing transient behaviors.

Fig. 6. Model-predicted impact of MDA on mf prevalence during seven rounds of MDA and after stopping MDA, and on CFA prevalence at the post-MDA survey in year 2008 for Leogane commune in Haiti.

Blue lines indicate the predicted impact of the interventions on mf prevalence whereas gray lines indicate the predicted impact of the interventions on CFA prevalence. Red dots indicate field-observed data points for mf while the blue dot is the corresponding data point observed for CFA. Horizontal and vertical black dotted lines indicate the WHO recommended 1% mf threshold and the time point of stopping MDA, respectively. Error bars indicate the 95% confidence intervals of the data points.

Fig. 7. Model-predicted impact of MDA alone (Group A villages) and MDA supplemented with VC throughout the intervention period (Group B villages) on mf prevalence.

Gray lines indicate the predicted impact of the interventions on mf prevalence and red dots indicate field data points. To implement simulations of the impact of integrated vector management (IVM) interventions carried out in Group B we fitted a segmented exponential function for monthly biting rate (MBR): MBR = MBR₀exp[a₁t], with a₁<0 for all t when VC is ON, otherwise a₁>0. Here we took a₁ = −5 for our model simulations. Error bars indicate the 95% confidence intervals of the data points.