Abstract T MP78: Real Time Prediction of Early Neurologic Deterioration using A Blood Pressure Volatility Model in Acute Ischemic Stroke

Abstract

Objectives: We aimed to develop a blood pressure (BP) embedded real time prediction model for early neurologic deterioration (END) in patients with acute ischemic stroke Methods: We identified consecutive ischemic stroke patients hospitalized within 48 hours of symptom onset from a prospective stroke registry database. BP data during hospitalization were obtained from the electric medical records. Probability of END at each time point of BP measurement was estimated using logistic model with covariates derived from two models for clinical information and BP parameters. A model for clinical information was fitted by using logistic model with patients’ clinical characteristics to predict END. A model for BP was fitted by random-effects models with temporal correlations at each time point of BP measurement with irregular intervals for mean as well as dispersion. Prediction performance was evaluated by calculating receiver operating characteristic (ROC) curve, and the cut-off value of high probability of END was determined at each time point. An alarm criterion for a proportion of high probability of END at each time point was defined as more than 50% of point probabilities being above the cut-off during the prior 24 hours. Predictive values of the prediction model were analyzed. Result: Of 1805 subjects, 18.3% experienced END. The predicted model for END within 24h hours from each time point of BP measurement was fitted by the model for clinical information of age, sex, history of stroke, time to arrival, baseline NIHSS score, diabetes mellitus, initial glucose level, atrial fibrillation, leukocyte count, stroke subtypes, recanalization therapy, and location of symptomatic vessel and by estimated mean of systolic BP and dispersion of diastolic BP from the temporal model for BP. Prediction performance was determined (Area Under Curve of ROC = 0.72) and the cut-off probability of END at each time point was set as > 0.01 (sensitivity = 50% and specificity = 81%). Using these criteria, about 70% of patients with END could be alarmed within 24 hours before the occurrence of END although 20% of those without END were falsely alarmed. Conclusion: This BP-embedded real time prediction model would helpful to predict and give a warning of the following END.