Using ecological momentary assessment to validate self-reported areas of potential transmission of latent tuberculosis and estimate annual risk of latent tuberculosis infection in Lubaga Division, Kampala

Abstract

From 2021 to 2023, Tuberculosis (TB) rose to become the primary cause of infectious disease mortality worldwide, surpassing its previous position as the second most prevalent. This shift was particularly significant in sub-Saharan Africa, where TB bears a disproportionately heavy burden, with an annual incidence of 125 per 100,000. In Uganda, the incidence rate was even higher at 200 per 100,000 population, with Lubaga and Kawempe divisions reporting the highest rates. The persistence of community transmission has been attributed to a lack of understanding about where TB exposure occurs in the community. Therefore, this study aimed at estimating the risk of Latent TB Infection (LTBI), finding the best machine learning (ML) models for predicting LTBI as a function of time spent in community avenues and validating quarterly self-reported time spent in avenues using Ecological Momentary Assessment (EMA). This study utilised secondary data from a case-prospective study that aimed to estimate the probability of TB infection and disease according to proximity among index TB cases using social network analysis (R01 AI093856). TB index cases and matched TB-negative community contacts were enrolled. TB-negative participants were followed at 3-month intervals for 12 months, assessing TB risk and time spent in community avenues and tested for LTBI at the 12-month visit using a tuberculin skin test (TST). Of those, a random sub-sample was exposed to EMA. This current sub-study focused on TB-negative participants to achieve its aims. Spearman’s correlation and Bland-Altman's methods validated the time spent in avenues as self-reported quarterly versus reports from EMA. A GEE Poisson model estimated the risk of LTBI as a function of time spent in avenues. ML models were compared using a confusion matrix for accuracy, sensitivity and Area under the Receiver Operating Characteristic Curve (AUC ROC). There were 1083 participants, of whom 632 (58.4%) were females, 514 (47.5%) were living partners or married, and the majority 1065 (98.3%) had attained education; secondary 627 (57.9%), primary 332 (30.7%) and tertiary 106 (9.8%). The median (IQR) age was 24 years (IQR 21, 29). The one-year incidence of LTBI was 18.9% (205/1083). LTBI was associated with female gender (Adjusted Risk Ratio= 0.691, p-value=0.006, 95% CI: 0.530, 0.893), being diabetic (ARR 4.24, <0.001, 95% CI: 2.527, 7.122) and travelling outside Kampala for work (ARR 1.00, 0.04, 95% CI: 1.001, 1.0013). Throughout the one-year period, in each quarter, on average, 50 (24.6%) LTB-positive participants reported visiting taxi parks, 21 (10.1%) visited friends’ homes, 20 (10.0%) visited trading centres, and 17 (8.3%) visited markets. On average, these participants spent more time in a friend’s home 2.1 (±1.3) hours, trading centres 1.9 (±1.3) hours, taxi parks 1.5 (±0.8) hours, and marketplaces 1.0 (±0.2) hours. LTBI was associated with hours spent in school (RR 1.07, 0.001 95% CI: 1.026, 1.114) and saloons (RR 1.15, 0.047, 95% CI: 1.002, 1.329). Through ML feature selection, schools, markets and bars were selected. The best ML models were Decision Tree (81.1%), logistic regression (80.7%) based on accuracy and K-Nearest Neighbor based on AUCROC. EMA validated self-reports for hours spent in worship centres, taxis and trading centres. LTBI transmission was associated with age, travelling, and hours spent in schools and saloons., Recall of time spent in community avenues was accurate for worship and trading centres. Scaling up EMA will strengthen community contact tracing. Additionally, increasing TB community awareness through mass media in crowded avenues and instituting stringent guidelines might promote self-protection and curb community TB transmission