Accuracy of Plasmodium falciparum genetic data for estimating parasite prevalence and malaria incidence in Uganda

TL;DR

This study evaluates how genetic data from malaria parasites in Uganda can estimate malaria spread and prevalence, finding that diversity metrics like COI are reliable for prevalence but less so for incidence.

Contribution

The study demonstrates that within-host diversity metrics, particularly COI and eCOI, are strong predictors of malaria prevalence across diverse transmission settings.

Findings

COI and eCOI were strongly and consistently positively associated with parasite prevalence.

Between-host relatedness was the strongest predictor of incidence but showed weaker overall performance in incidence estimation.

Genomic metrics captured spatial variation in malaria prevalence but had limited ability to predict incidence.

Abstract

Genetic metrics derived from Plasmodium falciparum infections offer a potential complement to conventional malaria surveillance by utilizing features of parasite diversity and relatedness to estimate transmission intensity. However, the performance of molecular metrics to predict epidemiologic metrics across a wide range of transmission settings remains understudied.

Dried blood spots from 3563 symptomatic malaria cases were collected from 26 sentinel health facilities across Uganda during two collections in 2023. Amplicon deep sequencing of 165 polyallelic microhaplotypes was performed using MAD4HatTeR. Within-host diversity metrics (complexity of infection (COI), effective complexity of infection (eCOI), percent polyclonality, within-host relatedness) and between-host relatedness metrics were calculated. Associations with prevalence and recent incidence were evaluated using correlation and regression analyses, and estimation accuracy was examined using nested grouped cross-validation.

Marked geographic heterogeneity in malaria burden was evident across sites; parasite prevalence ranged from 5.0% to 49.23% in Round 1, while incidence ranged from 91–1062 cases per 1,000 person-years (PY) in Round 1 and 33–1667 cases per 1,000 PY in Round 2. COI and eCOI were strongly and consistently positively associated with parasite prevalence. The proportion of highly related infection pairs was negatively associated with both prevalence and incidence and was the genetic metric most consistently associated with incidence. Nested grouped cross-validation identified single-predictor models using COI or eCOI as optimal for estimating prevalence, yielding a pooled cross-validated correlation of r = 0.79. Models estimating incidence showed weaker performance, with models incorporating both diversity and relatedness metrics achieving a pooled correlation of r = 0.37.

Microhaplotype-based metrics of within-host diversity, particularly COI and eCOI, reliably reflected spatial variation in malaria prevalence across Uganda, while between-host relatedness provided complementary information and was the strongest predictor of incidence. These findings indicate that parasite genomic metrics derived from polyallelic microhaplotypes can capture broad differences in transmission intensity reflected by parasite prevalence, but may have more limited ability to predict incidence. Integration of genomic metrics with harmonized epidemiologic data and expanded sampling of asymptomatic infections will be important next steps to understand the potential utility of parasite genetic metrics for malaria surveillance and subnational stratification.

Figures