# Sedative exposure and mortality in intracranial hypertensive tuberculous meningitis: a cohort study with propensity-score matching and machine learning analysis

## TL;DR

Prolonged sedation in patients with severe tuberculous meningitis is linked to higher mortality, suggesting a need for better sedation strategies.

## Contribution

This study uses machine learning and propensity-score matching to analyze sedative impact on mortality in tuberculous meningitis patients.

## Key findings

- Prolonged sedation (>3 days) is associated with reduced survival in intracranial hypertensive TBM.
- Machine learning identified gender, age, and hospital length of stay as top mortality predictors.
- Propensity-score matching showed higher crude mortality in sedated patients compared to non-sedated ones.

## Abstract

Tuberculous meningitis (TBM) complicated by intracranial hypertension requires aggressive neurocritical care, yet the mortality impact of sedative and antipsychotic exposure remains controversial. This study investigates the association between sedative exposure and mortality while identifying modifiable risk factors in this vulnerable population.

In this retrospective cohort study, we analyzed 1,875 intracranial hypertensive TBM patients from the MIMIC-IV database (v2.0). Exposure was stratified by cumulative sedative days (>3 vs. ≤3). Primary outcomes included 200-day mortality assessed using multivariable logistic regression and Cox proportional hazards models. Propensity score matching (PSM) was performed to adjust for confounding, and machine learning (XGBoost) was used to predict mortality and evaluate feature importance.

Unadjusted analyses identified age (odds ratio [OR] = 1.03 per year; 95% confidence interval [CI]: 1.01–1.05), sedative duration (OR = 1.13 per day; 95%CI: 1.04–1.22), and hospital length of stay (LOS; OR = 1.02 per day; 95%CI: 1.00–1.03) as significant mortality predictors. In the PSM cohort (n = 160 matched pairs), crude mortality rates were 16% in sedated versus 2.6% in non-sedated patients (p < 0.001), though the adjusted hazard ratio was non-significant (hazard ratio [HR] = 1.12; 95%CI: 0.83–1.50). Survival curves showed 200-day survival rates of 82% (95%CI: 79%–85%) for non-sedated and 47% (95%CI: 39%–55%) for sedated patients. The XGBoost model achieved an AUC-ROC of 0.79, identifying gender (SHAP value = 0.41), age (0.38), and LOS (0.29) as top predictors of mortality.

Prolonged sedation (>3 days) is associated with substantially reduced survival in intracranial hypertensive TBM, potentially reflecting both underlying disease severity and iatrogenic effects. Although residual confounding remains, machine learning analysis highlights the critical influence of gender and LOS on outcomes. These findings demonstrate the need for randomized trials evaluating targeted sedation minimization strategies to improve neurotuberculosis care.