Malaria surveillance estimates: Relationship between routinely reported metrics, malaria incidence, environmental covariates, and gender in high malaria transmission areas of Uganda

Abstract

Background: Like many African countries, Uganda relies on routine health facility data to monitor the malaria burden, changing disease epidemiology, and impact of control interventions. The data is often presented as test positivity rate (TPR, defined as number of laboratory-confirmed malaria cases per 100 suspected cases examined by either microscopy or rapid diagnostic test (RDT)), and total laboratory confirmed cases of malaria (TCM, defined as Total number of malaria cases confirmed positive by either microscopy or RDT). Unfortunately, these proxy metrics are unable to quantify changes in malaria burden over time and space which is possible when malaria incidence (the gold standard estimate) is used. In addition, the estimates are influenced by a number of factors which are not well appreciated in the Ugandan setting. This thesis compared the relationship between the routine reported estimates and malaria incidence, and investigated how environmental covariates and gender affect these estimates in high malaria transmission settings of Uganda. In addition, this study compared how the estimates vary when used to evaluate the impact of mass bed-net distribution in these settings. Methods: Two sub-studies were conducted including: 1) the use of public health facility-based surveillance, cross-sectional survey, and secondary publicly available environmental data from remote sensing sources to determine the relation between routine estimates and malaria incidence, environmental covariates, gender; 2) a quasi-experimental design utilizing data from patients who present to Outpatient departments of 12 public health facilities to determine how the estimates vary when used to evaluate the impact of mass bed-net distribution. Both linear and exponential models were used to compare TPR and TCM relative to malaria incidence estimates over time. Distributed lag nonlinear model was used to determine the associations between environmental covariates and malaria incidence. Poisson regression models adjusting for a calendar date was used to evaluate gender differences in the incidence of diagnosed malaria around public health facilities. The fractional and Poisson regression models adjusting for repeated measures within the sites were used to evaluate how the different estimates vary when used to estimate the impact of the 2020/2021 long-lasting Insecticide-treated bed net (LLIN) distribution campaign in high malaria transmission settings of Uganda. Results: TCM was a better predictor of malaria incidence (adjusted R-squared (adj. R2) range 0.81-0.98) compared to TPR (adj. R2 range 0.10-0.84) across sites. High values of xxiv

environmental covariates were significantly associated with increased cumulative incidence rate ratio (IRR) of malaria (rainfall; IRR =1.99, 95% confidence (CI): [1.22,2.27, temperature; IRR=8.16, 95% CI: [3.41,20.26], and normalized difference vegetation index (NDVI); IRR=1.57, 95% CI: [1.09,2.25] at lag-month 4). Female gender had a high incidence of diagnosed malaria at public health facilities compared to males (IRR=1.72, 95% CI [1.68,1.77], p<0.001). All surveillance estimates showed decreasing malaria burden following LLIN distribution (TPR; coefficient= -0.30 95% CI: [-0.59, -0.01], TCM; RR=0.90, 95% CI: [0.86,0.95], and malaria incidence; IRR=0.72, 95% CI [0.69,0.75]). Conclusion: Temporal changes in TCM correlated better with changes in malaria incidence compared to TPR and would be a better estimate when comparing changes over time using routine surveillance data. High temperature, Rainfall, and NDVI significantly increased the cumulative IRR of malaria and would need to be adjusted for when using incidence as an estimate for malaria burden. Females disproportionately contributed to the burden of malaria diagnosed at public health facilities, especially once they reached childbearing age. Reduction in malaria burden was observed following mass LLIN distribution. These findings demonstrate the variability of routinely reported metrics, and compositions of environmental covariates which may require consideration in planning for control interventions.