A Bayesian framework for estimating a latent placental malaria recency signal from delivery histology and its association with birthweight: A retrospective analysis in western Kenya

Abstract Background Placental malaria causes poor pregnancy outcomes, such as low birth weight. Histology at delivery can identify broad placental infection states, but does not differentiate the week of gestation at which the infection occurred. This study aimed to assess whether the delivery histology could serve as a model-based latent signal of recency and whether this recency signal was related to birthweight. Methods All 174 pregnancies with evaluable placental histology from a biobank in Bungoma County, western Kenya (54 without previous or current placental malaria, 28 with past, and 92 with active Placental malaria) were included in this retrospective secondary analysis. An ordered three-class histology outcome was related to a latent gestational recency process via a hierarchical Bayesian model, fitted in R with Stan, with birthweight as a covariate. For the 120 pregnancies with active or past placental malaria, a recency summary was reported. Birth weight was adjusted for maternal age, primigravidity, and intermittent preventive treatment in pregnancy. Four chains were run each for 4,000 iterations, 2,000 of which were warm-up iterations. Results The average birth weight of the 174 evaluable pregnancies was 3,086 g (SD 571). The mean of the pregnancy-specific posterior median recency values was 32.3 weeks (SD 1.9, range 24.8–39.2) for active placental malaria cases and 23.0 weeks (SD 1.8, range 15.2–25.9) for past placental malaria cases. The first and second-trimester-centred birthweight coefficients were − 62.6 g (95% credible interval − 293.3 to 170.5) and 4.0 g (-235.8 to 232.9). The third-trimester-centred coefficient was − 264.8 g (-485.5 to -47.8). The largest R-hat was 1.013, and the smallest bulk effective sample size was 432.6. Conclusions Delivery histology provides minimal information on relative placental malaria recency or persistence, but not the observed date of infection. The late-gestation coefficient was linked to lower birth weight in this model with latent processes, but the estimate is subject to the assumptions of the latent processes model and should not be interpreted as a causal trimester-specific effect. The framework needs to be tested by prospective longitudinal studies where malaria is measured at multiple time points, treatment records are maintained, and fetal growth is assessed.