Medication Regimen Complexity Index Score at Admission as a Predictor of Inpatient Outcomes: A Machine Learning Approach

Background: In the intensive care unit, traditional scoring systems use illness severity and/or organ failure to determine prognosis, and this usually rests on the patient’s condition at admission. In spite of the importance of medication reconciliation, the usefulness of home medication histories as predictors of clinical outcomes remains unexplored. Methods: A retrospective cohort study was conducted using the medical records of 322 intensive care unit (ICU) patients. The predictors of interest included the medication regimen complexity index (MRCI) at admission, the Acute Physiology and Chronic Health Evaluation (APACHE) II, the Sequential Organ Failure Assessment (SOFA) score, or a combination thereof. Outcomes included mortality, length of stay, and the need for mechanical ventilation. Machine learning algorithms were used for outcome classification after correcting for class imbalances in the general population and across the racial continuum. Results: The home medication model could predict all clinical outcomes accurately 70% of the time. Among Whites, it improved to 80%, whereas among non-Whites it remained at 70%. The addition of SOFA and APACHE II yielded the best models among non-Whites and Whites, respectively. SHapley Additive exPlanations (SHAP) values showed that low MRCI scores were associated with reduced mortality and LOS, yet an increased need for mechanical ventilation. Conclusion: Home medication histories represent a viable addition to traditional predictors of health outcomes.