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

The study addresses whether WHO-recommended Transmission Assessment Survey (TAS) stopping thresholds for lymphatic filariasis (LF)—specifically 1% microfilaremia (mf) or 2% circulating antigen (CFA) at community level and subsequent <0.5% mf or 1% CFA in 6–7-year-old children—reliably indicate interruption of transmission and justify stopping mass drug administration (MDA). LF elimination programs have reduced prevalence widely, pushing many areas to the endgame phase where robust stopping criteria are critical. Emerging field evidence and modeling suggest the globally fixed thresholds may not correspond to true local extinction thresholds and may not reflect adult age-group infection dynamics, raising risk of recrudescence if MDA is stopped prematurely. The purpose is to quantify, using data-fitted transmission models, the probability of elimination after crossing WHO thresholds versus model-derived site-specific extinction thresholds, and to evaluate intervention strategies (MDA frequencies, triple-drug regimens, and vector control) that best ensure sustained elimination. The importance lies in aligning operational stopping rules with underlying extinction dynamics to avoid undermining global LF elimination efforts.

The paper summarizes WHO’s TAS framework introduced in 2011, which sets 1% mf or 2% CFA community thresholds for pre-TAS and then surveys 6–7-year-old children at 2–3 year intervals to verify sustained suppression. Field studies have questioned whether crossing these thresholds indicates true transmission interruption or reflects adult prevalence, with reports of ongoing transmission after meeting TAS criteria. Recent modeling studies estimate lower and variable local interruption thresholds than WHO’s global targets and highlight that TAS thresholds were partly chosen for operational feasibility (sample sizes). Prior work with data-driven models has shown heterogeneous local dynamics, robustness of parameters, and the existence of biting rate and infection thresholds influenced by vector ecology and infection aggregation. Evidence from sites like Sri Lanka, American Samoa, Madagascar, and Leogane, Haiti indicates possible resurgence despite low child infection measures, underscoring concerns about global, one-size-fits-all thresholds.

• Data: Longitudinal human mf prevalence and intervention histories from six endemic sites spanning low, medium, and high transmission: DokanTofa (Nigeria), Piapung (Nigeria), Missasso (Mali), Kirare (Tanzania), Peneng (Papua New Guinea), and Dozanso (Mali). Inputs included baseline mf prevalence (age-stratified when available or reconstructed from overall prevalence), annual biting rate (ABR), dominant vector genus (Anopheles in all sites), drug regimens and coverage.
• Model: EPIFIL deterministic LF transmission model (vector genus-specific), representing human worm burdens (pre-patent, patent), mf and CFA intensities/prevalences, acquired immunity, and vector L3 larvae dynamics, with age-dependent exposure, density dependence, and infection aggregation.
• Calibration: Bayesian Melding with Sampling Importance Resampling (SIR). From wide priors, 200,000 parameter vectors were simulated; 500 best-fitting parameter sets per site were selected by likelihood against baseline prevalence data (observed or age-profile reconstructions) given site-specific ABR. Post-intervention predictions used these ensembles.
• Threshold estimation: Numerical stability analysis varied vector-to-human ratio to find threshold biting rate (TBR) where the system bifurcates between zero and endemic states; L3 threshold (L**) identified by increasing initial L3. Corresponding mf/L3 prevalence thresholds were computed at undisturbed ABR (for MDA-only targets) and at TBR (for MDA+vector control targets). From the ensemble of thresholds, the 95% elimination probability (EP) threshold (mf prevalence value whose crossing yields ≥95% chance of extinction) was used as model-derived elimination target.
• Intervention scenarios: Simulated three MDA strategies: (i) annual two-drug MDA (IVM+ALB or DEC+IVM per site), (ii) biannual two-drug MDA, (iii) annual IDA (IVM+DEC+ALB). Scenarios were run with and without vector control (VC), either from start until thresholds were crossed, or continued after stopping MDA for 5 years (TAS period) or 20 years. VC effects modeled per prior formulations (e.g., LLINs at ~65% coverage; MBR reduction function for integrated vector management). MDA targeted residents ≥5 years; worm kill rate estimated from post-intervention data; mf kill and sterilization durations fixed.
• Outcomes: For each site and scenario, computed (a) time to cross thresholds (WHO 1% mf vs site-specific 95% EP mf; also L3 95% EP in subset), and (b) probabilities by 5 and 20 years post-threshold of elimination (fraction of ensemble with negative slopes or convergence to zero) and recrudescence (positive slopes). Transient behavior (non-significant fluctuating slopes) was tracked to characterize protracted regime shifts.
• Validation: Compared model forecasts to field observations from Leogane, Haiti (seven MDAs, mf <1% by year 6 yet evidence of ongoing transmission in 2008), and to an Indian study comparing MDA-only vs MDA+VC with post-MDA surveys indicating VC preserved MDA gains and suppressed resurgence.

• Model fit and predictive adequacy: Site-specific calibrated models reproduced baseline age-prevalence patterns in most sites (except some over/underfit in Piapung and Kirare) and matched observed post-MDA prevalence reductions, supporting use for policy evaluation.
• WHO 1% mf threshold: Fewer MDA years needed (≈3–9 years) to reach 1%, but low to moderate probabilities of elimination over 5-year TAS period (≈1–41%) and substantial recrudescence risk (≈15–55%). After stopping at 1%, predicted mf prevalence in 5 years can rebound to ≈0.1–3% across sites. Elimination probabilities increase over 20 years but remain sensitive to site and regimen.
• Model-estimated 95% EP thresholds: Much lower mf thresholds (examples given: ≈0.001–0.007% mf) require longer MDA (≈11–19+ years) but, once crossed, yield high elimination probabilities over 5 years (often >50%, up to ≈98%; Piapung ≈31%) and near-zero recrudescence across sites; by 20 years, elimination probabilities generally >95% with recrudescence ≈0.
• L3 thresholds and timing: Crossing L3 95% EP thresholds occurs roughly in half the time needed to cross mf 95% EP thresholds; stopping at 1% mf typically occurs well before L3 thresholds are breached, explaining higher resurgence under TAS stopping.
• Intervention regimen effects: Biannual MDA and annual IDA reduce time to thresholds relative to annual two-drug MDA. However, more frequent/intense regimens, while crossing thresholds earlier, often show lower elimination probabilities over the immediate 5-year post-stopping period due to protracted transient dynamics; elimination probabilities converge high by 15–20 years.
• Vector control (VC): Adding VC from the start until threshold crossing provided limited improvement versus MDA alone. By contrast, continuing VC long-term (≈20 years) after stopping MDA at 1% mf dramatically increased elimination probabilities (approaching 100%) and reduced recrudescence (≈0–8%), across sites and MDA strategies. VC limited to only the 5-year TAS period gave less benefit.
• Age-structure implications: Meeting TAS criteria in children can mask ongoing transmission in older ages; models show mf in 6–7-year-olds can be <0.5% while older adults exhibit rising prevalence if MDA stops at 1%.
• Transient dynamics: After crossing thresholds, system exhibits a slow regime shift with long-lasting transients (years to decades) characterized by very low but persistent infection; this prolongs time to certain extinction, especially following rapid threshold crossing via intense MDA.
• Field corroboration: Leogane, Haiti data align with model predictions that mf <1% did not ensure interruption; Indian villages study supports that sustained VC post-MDA preserves gains and prevents resurgence.

The analysis indicates that WHO’s globally fixed 1% mf TAS threshold does not reliably signal interruption of LF transmission across diverse settings, largely because it is typically reached prior to crossing vector (L3) thresholds and without accounting for protracted extinction transients. Consequently, stopping MDA at 1% often leads to substantial recrudescence, especially in unmonitored older age groups. Using site-specific, model-derived 95% EP thresholds aligns stopping decisions with local extinction dynamics and substantially improves elimination likelihood and minimizes resurgence; however, these thresholds are very low, variable across sites, and challenging to measure operationally. Integrating long-term vector control after MDA—if programs continue using the 1% threshold—can compensate by suppressing transmission during the slow regime shift, achieving high elimination probabilities and preventing recrudescence. The findings emphasize the importance of extinction dynamics, heterogeneity, and transient behavior in endgame policy design, arguing for tailored thresholds or integrated endgame strategies that consider local transmission ecology and prolonged transition periods.

This study couples data-driven EPIFIL models with field data to evaluate TAS-based stopping rules versus model-derived extinction thresholds for LF. It shows that stopping MDA at the 1% mf threshold frequently fails to ensure sustained transmission interruption due to unbreached vector thresholds and long transient dynamics, leading to recrudescence. In contrast, crossing site-specific 95% EP thresholds yields durable elimination with minimal resurgence risk, though requiring longer MDA timelines. If programs retain the WHO threshold for operational reasons, sustained post-MDA vector control greatly improves outcomes, effectively ensuring elimination over longer horizons. Policy implications include: revisiting global thresholds in favor of locally informed targets; planning for extended surveillance and control during protracted extinction transients; integrating vector control into endgame strategies; and developing practical measurement approaches or proxies for very low prevalence thresholds. Future research should refine measurement of low-valued thresholds, characterize transient dynamics (including critical slowing down), evaluate integrated strategies in diverse ecologies, and compare deterministic and stochastic model predictions with long-term field data.

• Data limitations: Age-stratified baseline data were unavailable in four sites and reconstructed from overall prevalence patterns; some model fits (Piapung, Kirare) were weaker at certain timepoints. Field validation datasets are few and short-term.
• Model structure: Deterministic population-based model may estimate lower thresholds than stochastic individual-based frameworks at very low prevalence; stochastic fade-outs and small-population effects may alter threshold behavior.
• Measurement feasibility: Model-estimated 95% EP thresholds are extremely low and difficult to measure operationally; inference about their practical attainment relies on modeling proxies.
• Generalizability: All sites had Anopheles vectors; results may differ with other vector genera or distinct ecological contexts and migration patterns. Assumptions on drug effects, coverage, and VC efficacy introduce uncertainty.
• Time horizons: Elimination probabilities depend on long post-stopping horizons (10–20 years); programmatic feasibility and adherence to prolonged VC/surveillance are uncertain.