A time-series clustering analysis of postinduction blood pressure trajectories

Abstract Induction of general anesthesia is often associated with significant hemodynamic changes, particularly in blood pressure (BP). These early postinduction fluctuations can vary widely among patients and contribute to perioperative complications. Current clinical approaches to managing postinduction BP changes are largely reactive and may not fully account for individual variability. This study aimed to identify distinct patterns of mean arterial pressure (MAP) response during the first 10 min following induction of general anesthesia, using a time-series clustering approach. We conducted a retrospective cohort study of 17,645 adult patients undergoing non-cardiac, non-obstetric inpatient surgery under general anesthesia at a tertiary medical center. BP was measured at 1 min intervals using either invasive arterial lines (8.3% of cases) or standard non-invasive oscillometric cuffs. An unsupervised X-means clustering algorithm with dynamic time warping was applied to identify recurring MAP trajectory patterns. Patient demographics, comorbidities, anesthetic drug doses, and other perioperative characteristics were compared across clusters. Five distinct MAP trajectories were identified: Initial Decline—Plateau (31.8%), Gradual Moderate Decline (18.4%), Initial Decline—Recovery (7.5%), Gradual Severe Decline (29.6%), and Initial Decline—Low Plateau (12.7%). These patterns differed significantly in baseline MAP, comorbidity profiles and antihypertensive use, while differences in anesthetic agent doses were statistically but not clinically meaningful. Distinct postinduction BP trajectories were identified using a time-series clustering approach. These findings provide a framework for future validation in datasets with richer clinical context.