Pulmonary TB survivors face a high burden of post-TB lung disease (PTLD) after TB treatment completion. In this secondary data analysis we investigate the performance of parameters measured at TB treatment completion in predicting morbidity over the subsequent year, to inform programmatic approaches to PTLD screening in low-resource settings. Cohort data from urban Blantyre, Malawi were used to construct regression models for five morbidity outcomes (chronic respiratory symptoms or functional limitation, ongoing health seeking, spirometry decline, self-reported financial impact of TB disease, and death) in the year after PTB treatment, using three modelling approaches: logistic regression; penalised regression with pre-selected predictors; elastic net penalised regression using the full parent dataset. Predictors included demographic, clinical, symptom, spirometry and chest x-ray variables. The predictive performance of models were examined using the area under the receiver-operator curve (ROC AUC) values. Key predictors were identified, and their positive and negative predictive values (NPV) determined. The presence of respiratory symptoms at TB treatment completion was the strongest predictor of morbidity outcomes. TB survivors reporting breathlessness had higher odds of spirometry decline (aOR 20.5, 95%CI:3-199.1), health seeking (aOR 10.2, 2.4-50), and symptoms or functional limitation at 1-year (aOR 16.7, 3.3-133.4). Those reporting activity limitation were more likely to report symptoms or functional limitation at 1-year (aOR 4.2, 1.8-10.3), or severe financial impact of TB disease (aOR2.3, 1.0-5.0). Models were not significantly improved by including spirometry or imaging parameters. ROC AUCs were between 0.65-0.77 for the morbidity outcomes. Activity limitation at treatment completion had a NPV value of 78-98% for adverse outcomes. Our data suggest that whilst challenging to predict the development of post-TB morbidity, the use of symptom screening tools at TB treatment completion to prioritise post-TB care should be explored. We identified little benefit from the additional use of spirometry or CXR imaging.
Read full abstract