Training of machine learning algorithms for the classification of accelerometer-measured bednet use and related behaviors associated with malaria risk

Malaria affects three billion people worldwide. Despite remarkable reductions in malaria incidence over the last 15 years, recent evidence shows that our traditional control tools are weakening. Long-lasting insecticide-treated bednets (LLINs) are the most widely used tool for malaria prevention and have contributed significantly to de- creases in malaria incidence, but recent studies suggest that LLINs are either less effective than before or people are not using them as reported. A rigorous assessment of the timing and circumstances of LLIN use could be vitally important to regain the initiative in malaria control, but we lack a reliable measure of LLIN use. Current measurement tools, like self-reported use, are subjective and unable to account for temporal variations in use.

Paul Krezanoski previously invented an electronic monitor of LLIN use to address these limitations. In the NIAID-funded K23 work these tools were used to measure LLIN use related to malaria exposure. Most recently, Krezanoski has pioneered a vastly improved approach using machine learning algorithms and accelerometer-based sensors to measure LLIN use. The central rationale for this project is that by leveraging this novel accelerometer LLIN use monitor can help to be more ambitious with future goals for measuring malaria risk related to LLINs. When combined with a carefully trained machine learning algorithm, it is believed that it is possible to accurately classify a far wider range of behaviors than merely when an LLIN is used.

This study will additionally 1) measure if there is someone under the LLIN, 2) determine how many people are under the LLIN, 3) identify interruptions in LLIN use (such as entering or exiting an LLIN), and 4) characterize who is under the LLIN (e.g. adult versus child). Longitudinal surveillance of exactly this type of data is crucial to disentangle the role of LLINs in malaria prevention. This high-risk, high-reward proposal represents an inno- vative approach to answer a pressing question that limits our understanding of how LLINs prevent malaria: when and how are LLINs used in practice? The long term goal for this novel surveillance platform is to develop a tool for measuring LLIN use and related behaviors to point the way towards identifying better interventions for malaria prevention. The single accelerometer and generality in the training of the machine were purposefully chosen to facilitate adoption of this approach for malaria surveillance in other settings. This proposal will establish a firm foundation upon which to pursue future research studies using this novel tool and prepare me for an independent research career using novel technologies and research methods to improve malaria control.

NIH Project details

Information Systems
Monitoring & Evaluation
Surveillance
Vector Control

Aug 2023 — Jul 2025

$115,020

United States