Ability of the hypotension prediction index to predict hypotension in patients with septic shock in the intensive care unit.

The hypotension prediction index (HPI) is a machine learning-based model for predicting hypotension. It provides good performance for predicting intraoperative hypotension but has rarely been studied in critically ill patients admitted to the intensive care unit (ICU). This prospective study aimed to determine the ability of HPI to predict hypotension in ICU-admitted patients with septic shock. Hypotension was defined as a mean arterial pressure (MAP) of < 65 mmHg ≥ 1 min. Non-hypotensive events included MAP > 75 and ≥ 65 mmHg for the primary and secondary objectives, respectively. Positive events were selected at 5, 10, and 15-min before hypotension. Thirty patients with septic shock (293,315 data points) were included; 645 hypotensive events occurred, with a median of 14.5 [8-31] events/patient and a median time of 3.2 [2.5-4] min. At 5 and 10 min before hypotension, the area under the receiver operating characteristic curve (AUC) of HPI was significantly higher than at 15 min (0.953 vs. 0.919, p = 0.002 and 0.934 vs. 0.919, p = 0.01, respectively). HPI at 5, 10, and 15 min before the events showed better performance than other hemodynamic variables. However, after including MAP 65-75 mmHg as a non-hypotensive event, the ability of HPI declined and was similar to that of current MAP for predicting hypotension at all time points (AUC 0.817 vs. 0.786, p = 0.28 for 5 min, 0.787 vs. 0.761, p = 0.4 for 10 min, and 0.753 vs. 0.720, p = 0.31 for 15 min before the event). HPI showed excellent performance in predicting hypotension in septic shock patients in the ICU. However, its ability decreased when MAP 65-75 mmHg was considered non-hypotension. The HPI algorithm requires revisions to improve its predictive ability.