Insights from mathematical modelling and quantitative analysis on the proposed WHO 2030 targets for visceral leishmaniasis on the Indian subcontinent

Disclaimer

The opinions expressed herein are those of the authors and do not necessarily reflect the views of the World Health Organization. Publication in Gates Open Research does not imply endorsement by the Gates Foundation.

Background

Visceral leishmaniasis (VL) is a neglected tropical disease (NTD) caused by Leishmania protozoa that are transmitted by infected female sand flies. Most individuals remain asymptomatic, but those that develop symptoms experience long-lasting fever and hepatosplenomegaly, and die when left untreated¹. About 5–10% of treated individuals develop a skin condition called post-kala-azar dermal leishmaniasis (PKDL)². Interventions include active case detection (ACD) of VL followed by prompt treatment, and vector control through indoor residual spraying of insecticide (IRS).

The World Health Organization (WHO) has targeted the elimination of VL as a public health problem on the Indian subcontinent by 2020, defined as <1 VL case (new case and relapse) per 10,000 population per year, at the district level in Nepal and sub-district level in Bangladesh and India³. To achieve this target, the associated WHO guidelines describe four phases: a pre-control preparatory phase; a 5-year attack phase designed to bring the incidence below 1 per 10,000 per year; a consolidation phase where incidence is kept below the target for 3 years; and a maintenance phase to ensure sustainable reductions in incidence⁴. The interventions associated with these phases entail different levels of ACD and IRS. With 2020 approaching, WHO is in the process of formulating targets for 2030 for the Indian subcontinent, asking stakeholders in the VL community whether to maintain the 2020 target or to modify it, while adding the target of zero mortality among detected cases. The stakeholders include endemic country representatives, implementing partners, donors, mathematical modellers, WHO staff, and others, and were all invited to provide ideas and feedback on the new 2030 WHO targets (April–July 2019).

Mathematical transmission models have proven useful tools in gaining insight into the transmission dynamics of infectious diseases, highlighting knowledge gaps, and predicting the impact of interventions, to ultimately inform policy makers in formulating optimal strategies⁵. The Bill-and-Melinda-Gates-Foundation-funded NTD Modelling Consortium was founded in 2014, bringing together multiple groups working on the modelling of NTDs, including VL, with VL modelling teams from The Netherlands (Erasmus MC) and the United Kingdom (Warwick University, London School of Tropical Medicine and Hygiene, and Oxford University)⁶. In close collaboration with disease experts and policy makers, the VL teams captured the transmission dynamics of VL with mathematical models^7,8, compared their models through geographical cross-validation⁹, and published a joint policy paper¹⁰. Even though the models had certain different underlying assumptions regarding the main knowledge gaps, such as the role of asymptomatic individuals, PKDL cases, the duration of immunity, and the actual quality of the interventions, the models generally agreed on the predicted trends towards elimination. In this Open Letter we present an overview of the main VL modelling outcomes to support the development of the new WHO 2030 VL targets.

Insights gained from mathematical modelling analyses

Model predictions have suggested that the current intervention guidelines recommended by the WHO are sufficient to reach the elimination target in areas with moderate VL incidence (up to 5 reported VL cases per 10,000 population per year) prior to the start of interventions. However, additional interventions, such as extending the WHO attack phase (intensive IRS and ACD), may be required to achieve the elimination target in regions with high pre-control incidence, depending on the relative infectiousness of different disease stages. As VL incidence decreases, the pool of susceptible individuals will grow, creating the potential for new outbreaks¹⁰.

Despite the overall decrease in VL incidence at sub-district level, clusters or ‘hotspots’ of high incidence within sub-districts remain. We studied these hotspots at hamlet (sub-village) level using geospatial analyses and found significant spatial and temporal heterogeneity in VL incidence. Based on these findings, we argue for considering lower geographical units for evaluation of control after the targets at district/sub-district level have been achieved¹¹. Another key issue is the spatial spread of infection at an even smaller scale, between human hosts. Our modelling work suggests that case-to-case transmission is local (≤500 m)^11–13, which would support the current range of IRS around affected villages. However, the value of IRS, as it is currently implemented, has been questioned^14,15, whereas there is some indirect evidence for the impact of diagnosis and treatment^14,16,17, making case detection pivotal in the elimination programme. Moreover, even though this is currently not the target, no geographical areas have been “cleared” of transmission, suggesting that current tools are not sufficient to achieve zero transmission and may have to be implemented indefinitely to maintain control. Incomplete understanding of the spatiotemporal variation in VL incidence and its importance for achieving elimination as a public health problem and/or interrupting transmission undermines our ability to provide advice on the most appropriate geographical level to evaluate programme impact.

Recent xenodiagnostic studies have found PKDL can be at least as infectious as VL, depending on skin lesion type and severity¹⁸. When incorporating this into our models, PKDL becomes an important source of transmission, especially when nearing the low-incidence VL elimination target^10,19, likely challenging the maintenance of the elimination target on the Indian subcontinent. When adding a hypothetical PKDL control strategy (e.g. preventing 95% of PKDL) to the existing WHO strategy, the elimination target can be achieved as much as 8 years earlier, when keeping the assumption that only 2.5% of past VL cases develop PKDL constant¹⁹. Currently, the main intervention programmes do not include PKDL control. The main reason is that the options for diagnosing and treating this skin condition are limited. Ideally, active PKDL case-finding and treatment would be implemented, through long-term follow-up of VL patients, patient education to report themselves to the health clinic when symptoms of skin discoloration or rash appear, and educating community workers on recognizing PKDL. However, the suggested strategies are operationally difficult and require the availability of safe and effective treatment. There is a clear need for better VL treatment options, including a vaccine or immunomodulator to avert any PKDL development, and diagnostics for PKDL.

It is unclear to what extent interventions have contributed to the decline in VL incidence in addition to the secular trend in VL incidence, as some decline may be due to long-term (~15-year) natural cycles^20–22. However, transmission models suggest that interventions are likely to have contributed significantly to the decline over the last 5 years. Large village-level outbreaks of disease are still to be expected and have been observed in the field (e.g. Kosra²³)^24,25, with recent stochastic modelling work reproducing those trends¹². From the point of view of achieving the target of 1 per 10,000 cases in each block (sub-district level in India), due to the high variability in case numbers at this level, it is statistically implausible that the target can be achieved in all blocks simultaneously. This variation means that whilst low endemicity is maintained, the current target will fail at some point, as has occurred recently in Nepal²⁶. This suggests that the choice of administration level at which to assess elimination as a public health problem is critical to attaining and maintaining the target.

Table 1 summarizes the modelling insights and challenges for reaching the WHO 2030 targets. In Table 2 we summarize the priority questions that were identified in discussions with WHO and present how modelling can address this.

Practical implications of the currently proposed targets

Where are the risks that need to be mitigated to achieve the stated goals?

The current focus of the control programme is in the so-called “endemic” administrative units, although the actual definition used for this classification is unclear. This also means that the “non-endemic” units can have unobserved and uncontrolled transmission. The experience of the control programme has been that the “endemic” region appears to have expanded even as the incidence of cases has fallen²⁵ possibly due to human mobility combined with late development of PKDL and potential transmission from asymptomatic individuals, and/or an increase in case detection. Moreover, human mobility and migration might provide a constant “trickle” of transmission. Infrequent large epidemics (e.g. Kosra²³), and continued low-level transmission in Nepal^26,34 suggest that both of these occur.

Overall, poor understanding of the spatiotemporal variation in VL incidence and its importance for achieving control and/or interrupting transmission undermines our ability to provide advice on the most appropriate geographical level to evaluate programme impact. Fine-scale geographical measurement will lead to high variation over time and space whereas a lower spatial resolution (i.e. aggregating M&E results over larger areas) will come with a risk of missing programme failure at a more local level. Additional risks include the perverse incentive to not detect/register cases in order to ensure the target is achieved; natural disasters that could disrupt detection of cases (e.g. flooding in Raghopur block, India in August 2017 disrupted control); and the potential rise of HIV and its interaction with VL that could drive local resurgent epidemics³⁵.

Where are the places of perverse incentive?

Diagnosis and treatment are critical for ongoing VL control (i.e. to attain and maintain elimination as a public health problem). Targets should encourage diagnosis and treatment, but the “knife-edge” target encourages cases not to be diagnosed/treated/reported if it will mean that the block (and therefore country) fails to reach or sustain the target. The WHO validation process includes analyses to ensure that the claimed elimination is not due to underreporting⁴. The average block size in India is ~200,000, at which the target is 20 cases: 21 is failure and 19 is success. At the recent meeting WHO suggested to create other categories instead of success/fail such as very low/low/medium/high/very high. Such a target could encourage case finding and detach the success/failure of the target from finding of a single case in a block. Another suggestion is to report a detected case as a person who has been diagnosed, treated, and has not died. In this way it could be a positive incentive by saying “number of lives saved”. Other suggestions include setting an incidence target of 1/20,000 at the district level, or setting a target to increase the proportion of cases coming from low-incidence blocks, such as done in the Americas.

Priority questions

Data availability

No data are associated with this article.