Strong effect of demographic changes on Tuberculosis susceptibility in South Africa

Abstract

South Africa is among the world's top eight tuberculosis (TB) burden countries, and despite a focus on HIV-TB co-infection, most of the population living with TB are not HIV co-infected. The disease is endemic across the country, with 80-90% exposure by adulthood. We investigated epidemiological risk factors for (TB) in the Northern Cape Province, South Africa: an understudied TB endemic region with extreme TB incidence (926/100,000). We leveraged the population's high TB incidence and community transmission to design a case-control study with similar mechanisms of exposure between the groups. We recruited 1,126 participants with suspected TB from 12 community health clinics and generated a cohort of 774 individuals (cases = 374, controls = 400) after implementing our enrollment criteria. All participants were GeneXpert Ultra tested for active TB by a local clinic. We assessed important risk factors for active TB using logistic regression and random forest modeling. We find that factors commonly identified in other global populations tend to replicate in our study, e.g. male gender and residence in a town had significant effects on TB risk (OR: 3.02 [95% CI: 2.30-4.71]; OR: 3.20 [95% CI: 2.26-4.55]). We also tested for demographic factors that may uniquely reflect historical changes in health conditions in South Africa. We find that socioeconomic status (SES) significantly interacts with an individual's age (p = 0.0005) indicating that protective effect of higher SES changed across age cohorts. We further find that being born in a rural area and moving to a town strongly increases TB risk, while town birthplace and current rural residence is protective. These interaction effects reflect rapid demographic changes, specifically SES over recent generations and mobility, in South Africa. Our models show that such risk factors combined explain 19-21% of the variance (r2) in TB case/control status.

Fig 1. Case-control decision tree.

Study participants were categorized as cases or controls based on medical record information and self-reported data. All participants were GeneXpert tested for active TB infection at the time of enrollment. Past TB episodes were self-reported and cross-referenced with medical records when available.

Fig 2

Density plots of continuous variables A) Age by case-control status B) SES by case-control status.

Fig 3. Khoe-san ancestry is the primary genetic ancestry in clinics from the northern cape, South Africa.

A subset of participants (n = 159) was genotyped to obtain the average genome-wide ancestry proportions across all individuals for each clinic. A) Khoe-San ancestry is the largest proportion of ancestry in our sample, it varies significantly across study sites. The boxplots show the median, the 25^th and 75^th percentile and 1.5 times said percentile, and all outliers as dots. B) The study population is admixed with 5 distinct ancestries with the southern African indigenous Khoe-San ancestry being the highest proportion of ancestry at all study sites.

Fig 4. Logistic regression interaction plots.

A) The odds of active TB by education level vary across age groups (shown above by the different color lines). More years of education decreases the odds of active TB in younger age groups, but this pattern reverses in the oldest age groups. B) Effect plot from the residence model visualizing an interaction term between birthplace residence and current residence. Regardless of birthplace, the odds of active TB is highest in individuals who currently reside in towns. Individuals born in towns and currently residing in rural areas have the lowest odds of active TB.

Fig 5. Random forest model with common risk factor variables (n = 774).

Random subsets of all 6 variables on the y-axis were used to grow 5000 trees to classify participants into cases and controls. The model had an overall “out-of-bag” misclassification rate of 23%. Variables with higher variable importance are most crucial for case-control classification. Predictor variables with negative variable importance values worsen the ability of the model to classify TB status.