Application of Predictive Models for Tuberculous Meningitis Outcome: A Comparative Analysis of Decision Tree and Logistic Regression Approaches

Background: Tuberculous meningitis remains one of the most severe complications of tuberculosis infection. This study evaluated critical factors influencing mortality among tuberculous meningitis (TBM) patients and compared the predictive efficacies of logistic regression and decision tree models. Methods: A retrospective cohort analysis using medical records from 65 TBM patients at R.D. Kandou Hospital from January 2018 to July 2021. Patient outcomes were assessed with the Glasgow outcome scale (GOS), and the mortality risk was calculated. Key predictors of mortality identified by both multivariate logistic regression and the decision tree were compared using the receiving operating characteristic (ROC) curve. Result: Multivariate logistic regression analysis identified SGOT levels at admission (aOR: 1.06; CI95% 1.02-1.09; p=0.001), length of stay (aOR: 0.81; CI95% 0.71-0.92; p=0.002), and positive nuchal rigidity (aOR: 41.78; CI95% 3.41-512.27; p=0.004) as significant predictors of mortality. Decision tree analysis highlighted the British Medical Research Council (BMRC) stage, temperature, and potassium levels below 4.3 as critical predictors. Both models showed comparable predictive performance on the ROC curve, with no significant difference (0.85 vs. 0.95; p = 0.074). Conclusion: These results suggest that decision tree analysis is a viable alternative to logistic regression for predicting mortality in TBM patients, providing complementary insights into outcome-related factors. Further research is needed to refine these predictive models.