Introduction: Opioids are commonly used to treat pain and agitation in the cardiac intensive care unit. However, responsiveness to opioids in pediatric patients is poorly understood, and few guidelines exist for personalized management of sedation. This leads to probable overprescription of opioids in the pediatric ICU, which can extend length of stay and have negative long-term health consequences. In this work, we build a decision support tool to predict a patient’s response to an opioid bolus measured by the state behavior scale (SBS), using available patient data at the time of administration. Methods: We built a dataset including all individual doses of morphine administered to patients admitted to the CICU between 2011-2020. Patients missing critical hemodynamic or SBS data were excluded from the analysis. The change in SBS score before and after morphine administration was converted to a binary outcome, where a decrease in SBS score was considered to be favorable (A). A LASSO regression model was trained to predict SBS response. The predictors utilized in the model included demographics, hemodynamic data, SBS score measurements before morphine, concomitant drug infusions/boluses and morphine response to prior doses. The model was trained on 80% of data and evaluated on 20%. Results: Of 197,242 intravenous boluses, 66,195 boluses administered to 6,021 patients (median age 7.2 mo) were included. The median change in SBS score was 0 (IQR -1 - 0) with 33% of administrations classified as favorable responses. The model coefficients are shown in (B). The model predicts 74% of unfavorable morphine responses with 95% confidence, AUC = 0.899 (C). Conclusions: Nearly three-quarters of ineffective responses to morphine can be predicted with 95% confidence using the prediction models developed. Such models could substantially reduce morphine over-administration for critically ill patients in the pediatric cardiac ICU.
Read full abstract