OBJECTIVES: Obtaining initial medication history at admission can significantly influence clinical outcomes. Medication regimen complexity index (MRCI) scores highlight the differential clinical burden among races exacted by inaccurate medication errors. In this study, we evaluated racial disparities that stem from using MRCI for the prediction of inpatient outcomes.

METHODS: This was a retrospective cohort study using electronic medical record from 322 patients enrolled into the intensive care unit (ICU) in a 220-bed community hospital in Providence, Rhode Island, between 1 February 2020 and 30 August 2020. Outcomes included ICU mortality, inpatient length of stay, and the need for mechanical ventilation. Covariates of interest included Data included age, Charlson score, gender, race (Whites and Non-Whites), BMI, health insurance, and MRCI scores at admission, the Acute Physiology And Chronic Health Evaluation (APACHE) II score, and the Sequential Organ Failure Assessment (SOFA) score. Five models were designed from the following predictors and combinations: MRCI, APACHEII, SOFA, MRCI&APACHEII and MRCI&SOFA, and fitted to five classification machine learning algorithms (logistic regression, naïve Bayes, random forest, extreme gradient boosting). The Synthetic Minority Oversampling Technique was used to readjust sample sizes.

RESULTS: Although 65% of all patients were White, Non-Whites had higher representation among those with high MRCI (31.9% vs. 21.4%). Across all outcomes, the MRCI model performed poorer for minorities than it did for Whites (Accuracy: ~ 70% vs. 80% ). For minorities, the SOFA and MRCI&SOFA models were the best in terms of sensitivity and positive predictive value, respectively. The best models among minorities were obtained using logistic regression as opposed to extreme gradient boosting for Whites.

CONCLUSIONS: These findings provide justification to include home medication histories to the list of existing patient equity scoring systems such as APACHE II, SOFA.