(367) Predictive Nature of a Hemodynamic Indicator by Duration of Instability in Critically Ill Patients

Introduction: While hemodynamic instability (HI) is generally a precursor to adverse events, its duration and severity can be variable across patient populations. Therefore, a hemodynamic monitoring analytic should be effective across any duration of HI. Here, we investigate the Analytic for Hemodynamic Instability Predictive Indicator (AHI-PI) for predicting HI episodes of varying durations. AHI-PI uses only a single lead of ECG (II) as input and produces an output of future hemodynamic status once every two minutes.

Methods: The cohort analyzed in this retrospective observational study consisted of 1,129 consecutive arterial line patients monitored from 2019-2020 at Michigan Medicine. HI was defined as simultaneous tachycardia (HR>=100bpm) and hypotension (MAP < 70mmHg/Systolic < 90mmHg) that persisted at least 10 minutes. Episodes lasting up to 60 minutes were analyzed. Low or at-risk outputs were generated by AHI-PI at two-minute intervals while ECG was available. The prediction time was calculated as the duration of continuous at-risk AHI-PI outputs to the onset of each episode of HI. Only episodes that were deemed independent, such that AHI-PI prediction time does not overlap with another episode, were examined. Prediction times were then binned by episode duration to evaluate AHI-PI's performance over various HI episode durations.

Results: Of the 1,129 patients analyzed, 883 independent episodes in 427 patients were identified with durations between 10 and 60 minutes. AHI-PI had positive prediction times in 826 (93.5%) episodes. Most commonly episodes lasted 10-20 minutes (461 episodes). AHI-PI's median prediction times for episodes durations of 10-20, 20-30, 30-40, 40-50, and 50-60 were 50, 48, 54, 48, and 51 minutes respectively prior to episode onsets. The median prediction time was 50 minutes (mean: 200 minutes) across all event durations.

Conclusions: These results suggest that AHI-PI is an effective prediction tool for future episodes of HI regardless of their duration. While we primarily focused on ICU patients due to the requirement of arterial line to continuously evaluate HI, a tool such as this would be potentially most effective in predicting instability in lower acuity units where hemodynamic changes go unnoticed for lack of frequent monitoring.