Uncertainty Quantification in Cost-effectiveness Analysis for Stochastic-based Infectious Disease Models: Insights from Surveillance on Lymphatic Filariasis

Cost-effectiveness analyses (CEA) typically involve comparing effectiveness and costs of one or more interventions compared to standard of care, to determine which intervention should be optimally implemented to maximise population health within the constraints of the healthcare budget. Traditionally, cost-effectiveness evaluations are expressed using incremental cost-effectiveness ratios (ICERs), which are compared with a fixed willingness-to-pay (WTP) threshold. Due to the existing uncertainty in costs for interventions and the overall burden of disease, particularly with regard to diseases in populations that are difficult to study, it becomes important to consider uncertainty quantification whilst estimating ICERs.

To tackle the challenges of uncertainty quantification in CEA, we propose an alternative paradigm utilizing the Linear Wasserstein framework combined with Linear Discriminant Analysis (LDA) using a demonstrative example of lymphatic filariasis (LF). This approach uses geometric embeddings of the overall costs for treatment and surveillance, disability-adjusted lifeyears (DALYs) averted for morbidity by quantifying the burden of disease due to the years lived with disability, and probabilities of local elimination over a time-horizon of 20 years to evaluate the cost-effectiveness of lowering the stopping thresholds for post-surveillance determination of LF elimination as a public health problem. Our findings suggest that reducing the stopping threshold from <1% to <0.5% microfilaria (mf) prevalence for adults aged 20 years and above, under various treatment coverages and baseline prevalences, is cost-effective. When validated on 20% of test data, for 65% treatment coverage, a government expenditure of WTP ranging from $500 to $3,000 per 1% increase in local elimination probability justifies the switch to the lower threshold as cost-effective.

Stochastic model simulations often lead to parameter and structural uncertainty in CEA. Uncertainty may impact the decisions taken, and this study underscores the necessity of better uncertainty quantification techniques within CEA for making informed decisions.