Risk factors for tuberculosis treatment outcomes: a statistical learning-based exploration using the SINAN database...

Area of Science:

• Medical research
• Public health
• Epidemiology

Background:

• Understanding early predictors of treatment outcomes is crucial for efficient tuberculosis (TB) management and resource allocation.
• Predicting TB treatment success requires addressing challenges like incomplete health records and data imbalance.

Purpose of the Study:

• To predict TB treatment outcomes using real-world health data.
• To identify risk factors associated with mortality during TB treatment.
• To evaluate the effectiveness of multiple imputation analysis (MIA) in handling missing data for outcome prediction.

Main Methods:

• Employed upweighting and multiple imputation analysis (MIA) to manage missing data and response imbalance.
• Utilized logistic regression (LOGIT), random forest, and stochastic gradient boosting for outcome prediction.
• Conducted interpretation of LOGIT models, comparing MIA with complete case analysis (CCA).

Main Results:

• Multiple imputation analysis (MIA) proved effective for handling missing data in TB patient records.
• MIA-derived LOGIT models identified more statistically significant covariates associated with TB treatment outcomes than CCA.
• Factors increasing odds of death included TB clinical form (pulmonary and extrapulmonary), retreatment after abandonment, and absence of isoniazid or rifampicin in the treatment regimen.

Conclusions:

• MIA is a suitable method for addressing missing data in TB treatment outcome prediction.
• Identified key risk factors for mortality during TB treatment, informing clinical practice.
• Recommended the use of interpretable LOGIT models for predicting TB treatment outcomes.