# Exploring the association between dexmedetomidine and all-cause mortality in mechanically ventilated patients with sepsis through propensity score matching analysis and machine learning algorithms: a MIMIC-IV retrospective study

## TL;DR

Dexmedetomidine reduces 28-day mortality in ventilated sepsis patients but increases hospital and ICU stays.

## Contribution

This study combines propensity score matching and machine learning to evaluate dexmedetomidine's impact on sepsis mortality and ventilation outcomes.

## Key findings

- Dexmedetomidine use is associated with reduced 28-day and 180-day mortality in ventilated sepsis patients.
- Dexmedetomidine increases hospital and ICU length of stay and duration of mechanical ventilation.
- Random Forest machine learning model outperformed others in predicting mortality with dexmedetomidine as a significant feature.

## Abstract

Sepsis carries high ICU mortality globally, often requiring sedated mechanical ventilation. While some studies suggest dexmedetomidine improves survival in these patients, others contradict this finding. This study evaluates dexmedetomidine’s survival benefit and sedation value for ventilated sepsis cases.

This retrospective cohort study utilized the MIMIC-IV database and eICU-CRD to analyze mechanically ventilated septic patients. Propensity score matching was employed to balance covariates. Machine learning algorithms were applied to validate dexmedetomidine’s role in predicting mortality.

A propensity score matching analysis was performed for 5176 pairs of patients. The use of dexmedetomidine was associated with a reduced risk of 28-day mortality (13.39% vs. 19.84%, HR: 0.595, P < 0.001) and of 180-day all-cause mortality (17.45% vs. 23.18%, HR: 0.632, P < 0.001). However, dexmedetomidine use was also associated with longer hospital (median 15.08 days vs. 10.2 days, P < 0.001) and ICU stays (median 6.81 days vs. 4.0 days, P < 0.001). Moreover, the duration of mechanical ventilation was significantly longer in the dexmedetomidine group (median 78 h vs. 51.00 h, P < 0.001). Dexmedetomidine was included among the significant features identified with the Boruta algorithm, and of the five machine learning models built using the 20 most important features (including dexmedetomidine), the model constructed on the basis of the Random Forest algorithm performed the best (training set: AUC = 0.781; test set: AUC = 0.811; eICU-CRD set: AUC = 0.820). SHapley Additive exPlanations (SHAP) revealed that comorbid acute kidney injury (AKI) was the most important predictor of mortality among mechanically ventilated septic patients. This was followed by the use of opioids, PaO2, and the SOFA score, with the use of dexmedetomidine relatively closely behind.

Dexmedetomidine use significantly reduces short-term mortality in mechanically ventilated patients with sepsis but prolongs the hospital and ICU length of stay (LOS) and duration of mechanical ventilation. Administering dexmedetomidine within 48 hours and maintaining an infusion rate at or below 0.6 μg/kg/h appears to be more beneficial. Moreover, dexmedetomidine use strongly influences mortality in these patients.