Machine Learning to Predict Individualized Treatment Effects of Sodium Bicarbonate for Patients With Out-of-Hospital Cardiac Arrest.

Abstract

Objectives: Current evidence regarding the effect of sodium bicarbonate (SB) on patients with out-of-hospital cardiac arrest (OHCA) remains unclear. This study aimed to develop a machine-learning model to predict the individualized treatment effect (ITE) of SB use in OHCA patients.

Design: An eXtreme Gradient Boosting-based causal forest model was developed using an 8-year retrospective OHCA database after propensity score matching (PSM) for age, serum potassium, pH, bicarbonate, and Pco2 level.

Setting: Multicenter study across three hospitals affiliated with the National Taiwan University Hospital system.

Patients: Adult patients with nontraumatic OHCA.

Interventions: None.

Measurements and main results: The main outcome was any return of spontaneous circulation (ROSC) following resuscitation in the emergency department. Covariates included age, sex, witness status, bystander cardiopulmonary resuscitation, arrest location, response time, scene-to-hospital time, defibrillation using automatic external defibrillators, prehospital advanced airway type and epinephrine administration, initial cardiac rhythm, and laboratory data. The PSM cohort included 2368 patients. The ROSC rate was not different between the SB-treated and untreated groups. The predicted ITE ranged from a 24.7% absolute increase to a 28.3% absolute reduction in ROSC when SB was administered. The tertile of the predicted ITE significantly modified the effect of the original clinician treatment assignment on outcome (p < 0.001), and it can discriminate patients who benefit from SB better than random allocation when assessed by the Qini curve and C-for-benefit (0.61). Factors associated with higher predicted benefit from SB administration included older age, poorer renal function, longer scene-to-hospital time, metabolic acidosis, and hyperkalemia.

Conclusions: This study suggests the heterogeneous effects of SB on ROSC rates in patients with OHCA. The developed model may help identify specific subgroups more likely to benefit or be harmed by treatment. Further external validations and clinical trials are still needed to evaluate the model.