A Novel Clinical Nomogram for Predicting Unfavorable Tuberculosis Treatment Outcomes: A Logistic Regression Risk Model

TL;DR

This study created a clinical tool to predict poor tuberculosis treatment outcomes in Mozambique, which could help improve patient care and resource use.

Contribution

A novel clinical nomogram was developed to predict unfavorable TB treatment outcomes using logistic regression.

Findings

The nomogram showed good performance with an AUC of 83.2% and accuracy of 84.9%.

Key predictors of unfavorable outcomes included prior TB treatment, lack of DOT, and positive smear results.

The model demonstrated strong calibration and potential clinical utility in decision-making.

Abstract

Communicable diseases remain one of the major public health challenges in Sub-Saharan Africa, with tuberculosis (TB) ranking among the leading causes of morbidity, mortality, and significant economic impact. Mozambique is among the countries with the highest TB burden in the region. This study aimed to develop a clinical prediction model, in the form of a nomogram, to predict the probability of unfavorable treatment outcomes (UTO) among TB patients treated at a district health center in Nacarôa, Nampula Province, Mozambique.

A retrospective cohort study was conducted using secondary data from patients diagnosed and treated for TB between 2021 and 2023. A multivariable logistic regression analysis was performed to identify factors associated with UTO, and a predictive nomogram was subsequently constructed. Model performance was assessed using the receiver operating characteristic (ROC) curve, accuracy, Brier Score (BS), calibration plot, and the Hosmer–Lemeshow goodness-of-fit test. Clinical utility was evaluated through decision curve analysis (DCA) and clinical impact curves.

UTO were observed in 26.8% of patients (55/205). The multivariable analysis identified as significant predictors of UTO being previously treated for TB, not receiving directly observed therapy (DOT), having a clinical or radiological diagnosis, and having a positive smear microscopy result. The nomogram showed good performance, with an AUC of 83.2% and an accuracy of 84.9%. The Hosmer–Lemeshow test indicated good model fit (p = 0.132), and the calibration plot demonstrated strong agreement between predicted and observed outcomes (BS = 0.119). DCA and clinical impact analyses confirmed the model’s potential to support and optimize clinical decision-making in TB management.

The nomogram developed in this study represents a promising and practical tool for estimating the individual risk of UTO in tuberculosis care and may contribute to improved clinical management and resource allocation in high-burden settings.

The online version contains supplementary material available at 10.1007/s44197-026-00532-z.

Figures