Background: Monitoring and evaluation (M&E) is a key component of large-scale neglected tropical diseases (NTD) control programs. Diagnostic tests deployed in these M&E surveys are often imperfect, and it remains unclear how this affects the population-based program decision-making.
Methodology: We developed a 2-stage lot quality assurance sampling (LQAS) framework for decision-making that allows for both imperfect diagnostics and spatial heterogeneity of infections. We applied the framework to M&E of soil-transmitted helminth control programs as a case study. For this, we explored the impact of the diagnostic performance (sensitivity and specificity), spatial heterogeneity (intra-cluster correlation), and survey design on program decision-making around the prevalence decisions thresholds recommended by WHO (2%, 10%, 20% and 50%) and the associated total survey costs.
Principal findings: The survey design currently recommended by WHO (5 clusters and 50 subjects per cluster) may lead to incorrect program decisions around the 2% and 10% prevalence thresholds, even when perfect diagnostic tests are deployed. To reduce the risk of incorrect decisions around the 2% prevalence threshold, including more clusters (≥10) and deploying highly specific diagnostic methods (≥98%) are the most-cost saving strategies when spatial heterogeneity is moderate-to-high (intra-cluster correlation >0.017). The higher cost and lower throughput of improved diagnostic tests are compensated by lower required sample sizes, though only when the cost per test is <6.50 US$ and sample throughput is ≥3 per hour.
Conclusion/significance: Our framework provides a means to assess and update M&E guidelines and guide product development choices for NTD. Using soil-transmitted helminths as a case study, we show that current M&E guidelines may severely fall short, particularly in low-endemic and post-control settings. Furthermore, specificity rather than sensitivity is a critical parameter to consider. When the geographical distribution of an NTD within a district is highly heterogeneous, sampling more clusters (≥10) may be required.