Estimating local LF models and assessment of predictive performance
We began by assessing the ability of our BM-based data-model fusion modelling approach for successfully estimating best-fitting LF models in each of the present study sites. Note that age-stratified mf prevalences had to be constructed for 4 of these sites that provided only overall community mf prevalence observations 15,16, and all mf prevalence values were corrected to reflect sampling of 1 ml blood volumes for carrying out this exercise15,17. Nonetheless, the curves in Fig. 1 show that the LF models identified for each site based on the SIR algorithm are indeed well able to satisfactorily reproduce the site-varying baseline mf infection prevalence observed for either of these types of data – ie. whether actually measured (circles) or constructed (circles joined by lines) (Monte Carlo-p-values > 0.05 for each site except in the case of Piapung and Kirare (Table S2 in Supplementary Information)).
The predictive performances of these locally data-informed models were assessed by comparing forecasts of the impacts of the specific MDAs carried out in each site (Table 1) for reducing mf prevalence against the corresponding actually observed reductions in these data per site (Fig. 2). These comparisons indicate that, as with predictions of site-specific baseline infection prevalences (Fig. 1), and apart from a tendency to overestimate prevalence reductions at some time points in the Piapung and Kirare sites, the estimated locally applicable LF models are also able to mimic the observed changes in the data due to the applied interventions sufficiently well (Monte Carlo p-values for assessing goodness-of-fit insignificant in each setting > 0.05 (Table S2 in Supplementary Information)).
Estimating timelines and probabilities of elimination due to the MDA alone interventions carried out in each site
The numerical forecasts of the timelines made by each site-specific model ensemble for reaching the 1% mf prevalence threshold due to the annual MDAs applied (at the average coverages reported in Table 4) in each of the present study sites are given in Table 1A. These highlight that apart for the case of Piapung the present models are able to reliably predict the required MDA intervention durations for breaching the 1% mf target in each site. Thus, where the number of years predicted by the locality-specific models for achieving the 1% mf threshold were either lower or equal to the actually applied annual MDA rounds, the final prevalences observed (following the applied MDAs) in these sites were found to be lower than the 1% mf prevalence threshold (Table 1A). This is in contrast to sites (eg. Kirare, Peneng) in which the model predicted annual MDA durations for reaching the 1% mf threshold are estimated to be longer than the actually applied MDA durations; in this case the end mf prevalences reached are, as expected, found to be higher than the 1% mf prevalence target (Table 1). This matching of predictions with observations futher confirm the conclusion made above that the present locally calibrated models are empirically adequate for capturing the dynamics of the applied MDA control in each site, and so can serve as reliable tools for addressing the chief objective of this study, viz. evaluating the consequences of applying the WHO TAS criteria for guiding decision making regarding the site-specific interruption of LF transmission.
Table 1 also provides the elimination probabilities over the 5-year TAS period estimated by the localized predictive models following the reduction of mf prevalences below the 1% mf threshold (Table 1A) or the significantly much lower model-estimated 95% EP mf prevalence thresholds estimated at ABR in each of the present sites (Table 1B). The calculated probabilities show that while achieving the 1% mf prevalence target in each site will require significantly fewer years of annual MDAs (from 3 to 9 years depending on intial endemicity), such achievement will, however, result in either low or moderate chance (4 to 50%) of transmission interruption with a corresponding significant probabilities of recrudescence (between 50 to 96%) (Table 1A). By contrast, even though it will take considerably longer (11 to 19 years) to achieve the much lower 95% EP thresholds (ranging from 0.001–0.007%) applicable in each site, the results show that once these thresholds are crossed, the probability of achieving elimination of transmission will be very high and close to nearly 100% (> 89%) in each site, with the corresponding recrudescence probabilities effectively close to zero, except in the case of Piapung village (Table 1B).
Table 1
A: Observed and model predicted post-MDA data for WHO recommended 1% mf threshold for six LF infected sites.
Settings | Site | No. of annual MDA rounds observed | Mf prevalence following interventions | Mean No. of years to reach 1% mf (model predicted) | Probability of elimination (%) 5years after crossing 1% mf threshold | Probability of recrudescence (%) 5years after crossing 1% mf threshold |
Low | DokanTofa, Nigeria18 | 7 | 0.4% | 3 | 9 | 91 |
Piapung, Nigeria18 | 7 | 2.1% | 4 | 4 | 96 |
Medium | Missasso, Mali19 | 6 | 0.0% | 5 | 8 | 92 |
Kirare, Tanzania20 | 6 | 2.7% | 7 | 20 | 80 |
High | Peneng, PNG21 | 5 | 3.7% | 9 | 50 | 50 |
Dozanso, Mali19 | 6 | 0.0% | 6 | 32 | 68 |
Table 1
B: Model predicted post-MDA data for model predicted 95% EP threshold for six LF infected sites.
Settings | Site | 95% EP threshold values | Mean no. of years to reach 95% EP threshold (model predicted) | Probability of elimination (%) 5years after crossing model predicted 95% EP threshold | Probability of recrudescence (%) 5years after crossing model predicted 95% EP threshold |
Low | DokanTofa, Nigeria | 0.003109 | 11 | 95 | 5 |
Piapung, Nigeria | 0.001550 | 13 | 89 | 11 |
Medium | Missasso, Mali | 0.007500 | 15 | 98 | 2 |
Kirare, Tanzania | 0.001799 | 18 | 98 | 2 |
High | Peneng, PNG | 0.004534 | 19 | 99 | 1 |
Dozanso, Mali | 0.001101 | 16 | 95 | 5 |
Assessments of the impact of including vector control in MDA interventions on elimination and recrudescence probabilities
We next examined the relative effectiveness of three MDA scenarios with and without inclusion of VC for accomplishing sustainable LF elimination and reducing likely recrudescence in the present sites. The three MDA-based scenarios chosen for study were: (i) annual MDA, (ii) biannual MDA and (iii) annual IDA, firstly, with and without inclusion of supplementary VC from the beginning of MDA to when the WHO proposed 1% mf prevalence or model-estimated 95% EP thresholds are breached, and, secondly, after achievement of these thresholds using MDA continuing with VC alone for either over the 5-year TAS period or for up to 20 years post stoppage of MDA (Table 2). The drug combinations applied in each site (ie. IVM + ALB or IVM + DEC (Table 4)) were used for simulating the impacts of annual and biannual MDA respectively in each site, while the combined effects of IVM, DEC and ALB were modelled for forecasting the impact of annual IDA. The average MDA coverage and baseline mf prevalence pertaining to each site were used to run these scenario simulations (see Methods).
The results from carrying out the various MDA only and MDA plus VC intervention simulations are shown in the case of Piapung, Missaso and Dozanso villages, representing the low, medium and high baseline endemic sites investigated here respectively, for both the 1% (Table 2A) and their corresponding site-specific 95% EP mf prevalence thresholds (Table 2B). The time in years to cross each respective threshold and the probabilities of achieving elimination or recrudescence following reaching either threshold (and stoppage of MDA) are depicted in these tables with and without inclusion of VC for both the 5-year TAS period as well as over a longer 20-year term following the breaching of these thresholds. The results demonstrate, firstly, as expected, that timelines to reach the 1% mf theshold will increase with endemicity but decreased in each of the three sites as annual treatments are replaced by biannual two-drug MDAs and by the annual administration of IDA irrespective of whether VC is included or not (Table 2A). Thus, while annual single drug MDA interventions (with or without VC) took the longest times to reach the 1% mf prevalence target among the present drug interventions, this threshold was reached earlier in the the low transmission village of Piapung (5 years) compared to the need for 9 years of annual mass treatments required for Dozanso with the same IVM + ALB regimen. Biannual MDA, however, is predicted to be the most effective strategy for achieving the 1% mf threshold in the present villages, given that it takes only 1 (Piapung) to 4 years (Dozanso) compared to the 2 to 5 years required by annual IDA to cross this threshold in the same villages (Table 2A).
The second finding arising from the results depicted in Table 2A is that using annual MDAs, irrespective of the drug regimens used and inclusion of VC, counterintuitively resulted in greater elimination (between 4–51% probability of elimination after 5 years and between 26 to 100 probability of elimination after 20 years) and lower recruscrence (between 49–96% recrudescence probabilities after 5 years; between 0–74% recrudescence probabilities after 20 years) in these 3 villages compared to the application of the biannual MDA and IDA interventions (elimination probabilities generally between 0–33% with corresponding recrudescence probabilities reaching generally > 70–100%) following the achievement of the 1% mf threshold (Table 2A).
Combining VC with MDA from the start until the 1% mf threshold is reached did not overly improve the results obtained with the use of MDA alone, irrespective of the MDA strategy; however, including VC over the long-term (ie for 20 years after achievement of the 1% mf threshold by MDA) had an dramatic effect in increasing the probability of elimination (to > 71%) and reducing the corresponding probability of recrudescence (to as low as < 10%) depending on site and MDA strategy. Including VC after achievement of the 1% mf threshold by MDA for over the 5-year TAS period alone, however, while improving the results obtained with the use of MDA alone irrespective of site or MDA strategy (Table 2A), resulted in lower elimination probabilities compared to when it was included over the longer 20-year period for all interventions. As a result of this, all estimated recruscence probabilities were found to be higher for the 5-year TAS period compared to the corresponding values calculated for the longer term 20-year period, highlighting the need for longer continuation with VC beyond the 5-year TAS period to ensure the breakage of parasite transmission under this scenario.
The corresponding results on timelines to extinction and probabilities of elimination and recrudescence using the corresponding site-specific 95% EP thresholds as the extinction target in each village is shown in Table 2B. These show that using these thresholds will first of all significantly lengthen the durations of interventions required to break parasite transmission. Thus, while using annual single MDAs will take between 14 to 24 years to cross these thresholds in the three example villages - as compared to just 5 to 9 years it would take for this intervention to cross the 1% mf threshold in the same villages depending on site and inclusion of VC (Table 2A) - this will be reduced to between 8 to 13 and 10 to 17 years of treatments required when the corresponding biannual and annual IDA MDA strategies are used in these villages (in comparison with just the 1 to 4 and 2 to 5 years required by these interventions to reach the 1% mf threshold ). Adding VC to each MDA strategy from the beginning, by contrast to the results arising from using the 1% mf threshold, however, despite still taking longer, will result in reducing the overall durations required (by at least 2 to 6 years depending on strategy and site) to reach the 95% EP threshold (Table 2B).
The estimated probabilities of elimination and recrudescence depicted in Table 2B, on the other hand, highlight the major benefit of using the 95% EP threshold as targets for LF elimination programs, viz. that once these thresholds are breached by an intervention, it will inevitably lead to interruption of transmission (> 98% probability at least) and result in the occurrence of close to zero probabilities of infection recrudescence, particularly over the longer 20-year period irrespective of site or type of MDA-based strategy (Table 2B). Another interesting, but perhaps expected finding, is that simply using MDA alone for achieving the 95% EP threshold, in contrast to using the 1% mf target, is also adequate for accomplishing transmission elimination as well as reducing the probability of infection recrudescence to close to zero over the long-term irrespective of the vagaries of site conditions and whether VC is included or not (compare Tables 2B and A). However, increasing the frequency of MDA (to biannual treatments) and use of annual IDA resulted in lower elimination and higher recrudescence probabilities following the 5-year TAS period irrespective of whether VC was included or not when these 95% EP targets are used (Table 2B)
Inclusion of VC with MDA from the beginning helped to reduce the time to reach the site-specific 95% EP thresholds and to increase the probability for transmission elimination for the 5-year TAS period, but it was found not to add any additional benefit to increasing the probability for achieving transmission elimination or reducing the probability of recrudescence over that accomplished by MDA alone (carried out until the 95% EP thresholds are reached) for a longer term 20 years period (Table 2B), in contrast to these outcomes when using the 1% mf threshold (Table 2A). Applying VC during 5 years (TAS period) after achievement of the 95% EP threshold by MDA alone will also result in almost the same benefit in increasing the probability for achieving transmission elimination or reducing the probability of recrudescence as that accomplished by carrying out either MDA alone or by implementing longer-term supplementary VC (Table 2B).
Table 2
A. Recrudescence probability and elimination probability for different scenarios with 1% mf threshold for one low (Piapung), one med (Missasso) and one high prev (Dozanso) sites.
| Piapung | Missaso | Dozanso |
| | Elimination probability (%) | Recrudescence probability (%) | | Elimination probability (%) | Recrudescence probability (%) | | Elimination probability (%) | Recrudescence probability (%) |
Scenario | Time to cross threshold (year) | After 5 years | After 35 years | After 5 years | After 35 years | Time to cross threshold (year) | After 5 years | After 35 years | After 5 years | After 35 years | Time to cross threshold (year) | After 5 years | After 35 years | After 5 years | After 35 years |
Annual MDA upto 1% mf | 5 | 4 | 33 | 96 | 67 | 6 | 8 | 26 | 92 | 74 | 9 | 32 | 48 | 68 | 52 |
Annual MDA + VC upto 1% mf | 5 | 5 | 36 | 95 | 64 | 6 | 13 | 33 | 87 | 67 | 9 | 39 | 54 | 61 | 46 |
Annual MDA upto 1% mf, endgame VC | 5 | 23 | 99 | 77 | 1 | 6 | 40 | 100 | 60 | 0 | 9 | 51 | 99 | 49 | 1 |
Annual MDA upto 1% mf, then only VC for next 5years | 5 | 23 | 58 | 77 | 42 | 6 | 40 | 48 | 60 | 52 | 9 | 51 | 61 | 49 | 39 |
Biannual MDA upto 1% mf | 1 | 0 | 0 | 100 | 100 | 3 | 2 | 14 | 98 | 86 | 4 | 6 | 21 | 94 | 79 |
Biannual MDA + VC upto 1% mf | 1 | 0 | 0 | 100 | 100 | 3 | 2 | 16 | 98 | 84 | 4 | 7 | 23 | 93 | 77 |
Biannual MDA upto 1% mf, endgame VC | 1 | 0 | 71 | 100 | 29 | 3 | 12 | 89 | 88 | 11 | 4 | 10 | 82 | 90 | 18 |
Biannual MDA upto 1% mf, then only VC for next 5years | 1 | 0 | 3 | 100 | 97 | 3 | 12 | 30 | 88 | 70 | 4 | 10 | 33 | 90 | 67 |
Annual IDA upto 1% mf | 2 | 0 | 0 | 100 | 100 | 4 | 0 | 5 | 100 | 95 | 5 | 0 | 4 | 100 | 96 |
Annual IDA + VC upto 1% mf | 2 | 0 | 0 | 100 | 100 | 4 | 0 | 6 | 100 | 94 | 5 | 0 | 11 | 100 | 89 |
Annual IDA upto 1% mf, endgame VC | 2 | 0 | 76 | 100 | 24 | 4 | 0 | 92 | 100 | 8 | 5 | 0 | 80 | 100 | 20 |
Annual IDA upto 1% mf, then only VC for next 5years | 2 | 0 | 6 | 100 | 94 | 4 | 0 | 15 | 100 | 85 | 5 | 0 | 15 | 100 | 85 |
Table 2
B. Recrudescence probability and elimination probability for different scenarios with 95% EP threshold for one low (Piapung), one med (Missasso) and one high prev (Dozanso) site
| Piapung | Missasso | Dozanso |
| | Elimination probability (%) | Recrudescence probability (%) | | Elimination probability (%) | Recrudescence probability (%) | | Elimination probability (%) | Recrudescence probability (%) |
Scenario | Time to cross threshold (year) | After 5 years | After 35 years | After 5 years | After 35 years | Time to cross threshold (year) | After 5 years | After 35 years | After 5 years | After 35 years | Time to cross threshold (year) | After 5 years | After 35 years | After 5 years | After 35 years |
Annual MDA upto 95% EP | 20 | 89 | 100 | 11 | 0 | 22 | 98 | 100 | 2 | 0 | 24 | 95 | 100 | 5 | 0 |
Annual MDA + VC upto 95% EP | 14 | 97 | 99 | 3 | 1 | 19 | 99 | 99 | 1 | 1 | 20 | 98 | 99 | 2 | 1 |
Annual MDA upto 95% EP, endgame VC | 20 | 89 | 100 | 11 | 0 | 22 | 98 | 100 | 2 | 0 | 24 | 95 | 100 | 5 | 0 |
Annual MDA upto 95% EP, then only VC for next 5yrs | 20 | 89 | 100 | 11 | 0 | 22 | 98 | 100 | 2 | 0 | 24 | 95 | 100 | 5 | 0 |
Biannual MDA upto 95% EP | 11 | 22 | 99 | 78 | 1 | 12 | 63 | 99 | 37 | 1 | 13 | 45 | 100 | 55 | 0 |
Biannual MDA + VC upto 95% EP | 8 | 85 | 98 | 15 | 2 | 11 | 96 | 99 | 4 | 1 | 11 | 94 | 98 | 6 | 2 |
Biannual MDA upto 95% EP, endgame VC | 11 | 22 | 100 | 78 | 0 | 12 | 64 | 100 | 36 | 0 | 13 | 45 | 100 | 55 | 0 |
Biannual MDA upto 95% EP, then only VC for next 5yrs | 11 | 22 | 99 | 78 | 1 | 12 | 64 | 100 | 36 | 0 | 13 | 45 | 100 | 55 | 0 |
Annual IDA upto 95% EP | 13 | 29 | 99 | 71 | 1 | 16 | 75 | 100 | 25 | 0 | 17 | 63 | 100 | 37 | 0 |
Annual IDA + VC upto 95% EP | 10 | 79 | 98 | 21 | 2 | 13 | 99 | 100 | 1 | 0 | 15 | 98 | 100 | 2 | 0 |
Annual IDA upto 95% EP, endgame VC | 13 | 29 | 100 | 71 | 0 | 16 | 75 | 100 | 25 | 0 | 17 | 63 | 100 | 37 | 0 |
Annual IDA upto 95% EP, then only VC for next 5yrs | 13 | 29 | 99 | 71 | 1 | 16 | 75 | 100 | 25 | 0 | 17 | 63 | 100 | 37 | 0 |
Assessing validity of model forecasts using field data
Field data to corroborate the above model predictions are still scare, but data from two studies allowed evaluation of the impact of adding vector control to MDA programmes as a measure for reducing infection recrudescence after stopping MDAs once prevalences are reduced to near 1% mf prevalence22, and if this would indeed lead to sustained interruption of transmission once crossed and MDAs are stopped 13,23−26. By contrast, surveillance data on the impact of annual MDA using DEC + ALB assembled from the commune of Leogane in Haiti allowed the direct investigation of whether using the WHO recommended 1% mf threshold is sufficient to bring about LF transmission interruption and hence support the stopping of intervention13,23−26. In the latter setting, seven rounds of MDA were delivered starting from year 2000 with variable coverages and compliances and mf prevalence was reduced below the 1% threshold by the sixth year of treatments (in 2005); however, a subsequent survey carried out in whole population in year 2008 indicated that transmission was still ongoing in the community despite the apparent success of the applied MDAs13. Figure 5 depicts both the changes in mf prevalence in the community until year 7 post-MDA and the prevalences for both CFA (blue points/lines) and mf (red points/lines) at the post-MDA survey in year 2008. It also portrays the predictions of our model calibrated to the baseline mf data for the annual MDA interventions carried out in the commune (by taking also into account the actual coverages attained annually). These results provide important empirical evidence indicating that the WHO 1% mf threshold has not led to interruption of LF transmission in this setting, and indeed that, as the model forecasts show (and further mimicking the theoretical results above), the use of this threshold will instead result in significant corresponding resurgences of infection if such arbitrarily thresholds are used for deciding the stopping of interventions.
In the study by Sunish et al22, on the other hand, one group of villages received two annual MDAs of DEC + IVM), while Group B villages received the same MDA schedule in combination with VC, chiefly using expanded polystyrene beads (EBP), biolarvicide and larvivorous fish (Table 3). MDAs were stopped after two annual treatments and the communities were resurveyed for infection levels three years after stoppage of MDAs in both groups of villages. VC, however, continued in the Group B villages following cessation of MDA. The results from the post-treatment surveys showed that VC preserved the effects of MDA while resurgences occurred in Group A villages (Table 4). Overall, continuing with vector control suppressed the mf prevalence attained three years following stoppage of MDA in the Group B villages by up to 55.8% compared to the mf prevalence reached in Group A (Table 3). Figure 6 shows the corresponding model predictions in comparison with observed data, and highlights that the theoretical findings described above regarding the impact of using supplementary vector control for both retaining the gains of MDA and arresting LF recrudescence may indeed operate in the field.
Table 3
Pre- and post-treatment data for two groups of villages at Tirukoilur in India.
Treatment Groups | M&E data 22 |
Pre-treatment (1994 October-December) mf positive (%) (no. of sample) | First survey (1997 October-December) mf positive (%) (no. of sample) | Second survey (1999 April-June) mf positive (%) (no. of sample) |
Group A (MDA alone for 1995 (June-August) and 1996 (July-September) | 15.19 (724) | 1.81 (609) | 4.74 (591) |
Group B (MDA + VC for 1995 (June-August) and 1996 (July-September) and VC up to 1998 (December)) | 15.09 (795) | 1.24 (645) | 2.08 (673) |