Personalizing Prediction of High Opioid Use in the Neurointensive Care Unit: Development and Validation of a...

Background/Objectives: This study aimed to develop and validate a stratified risk model for predicting high opioid use in patients with acute brain injury due to stroke or traumatic brain injury (TBI) admitted to a neurocritical care intensive care unit. Methods: We examined the factors associated with the use of high-opioids (≥75th quartile, ≥17.5 oral morphine equivalent/ICU day) in a retrospective cohort study including patients with acute ischemic stroke, spontaneous intracerebral hemorrhage, spontaneous subarachnoid hemorrhage, and TBI. We then developed, trained, and validated a risk model to predict high-dose opioids. Results: Among 2490 patients aged 45-64 years (β = -0.25), aged 65-80 years (β = -0.97), and aged ≥80 years (β = -1.17), a history of anxiety/depression (β = 0.57), a history of illicit drug use (β = 0.79), admission diagnosis (β = 1.21), lowest Glasgow Coma Scale Score (GCSL) [GCSL 3-8 (β = -0.90], {GCS L 9-12 ((β = -0.34)], mechanical ventilation (β = 1.21), intracranial pressure monitoring (β = 0.69), craniotomy/craniectomy (β = 0.6), and paroxysmal sympathetic hyperactivity (β = 1.12) were found to be significant predictors of high-dose opioid use. When validated, the model demonstrated an area under the curve ranging from 0.72 to 0.82, accuracy ranging from 0.68 to 0.91, precision ranging from 0.71 to 0.94, recall ranging from 0.75 to 1, and F1 ranging from 0.74 to 0.95. Conclusions: A personalized stratified risk model may allow clinicians to predict the risk of high opioid use in patients with acute brain injury due to stroke or TBI. Findings need validation in multi-center cohorts.