Discrimination and calibration properties of the hypotension probability indicator during cardiac and vascular surgery

Abstract

Background: Hypotension during surgery is linked to postoperative complications. Recently, a new hemodynamic algorithm intended to predict hypotensive events (hypotension probability indicator [HPI]) has been developed. The aim of the present study is to test the discrimination and calibration properties of the HPI.

Methods: The intraoperative files of 23 patients undergoing cardiac and major vascular surgery receiving the HPI-based hemodynamic monitoring were retrospectively investigated for prediction of hypotensive events (mean arterial pressure <65 mmHg). The HPI was available at 20 seconds intervals; the values of HPI five to seven minutes before a hypotensive event (HPI5-7) were tested for discrimination and calibration.

Results: The HPI5-7 has a fair level of discrimination (area under the curve 0.768) and a poor calibration, due to overestimation of the hypotensive risk. At the observed prevalence, a cut-off value of 85% carries a sensitivity of 62.4% and a specificity of 77.7%, a negative predictive value (NPV) of 97.8% and a positive predictive value (PPV) of 12.6%; a value of 98% has a PPV of 64% and an NPV of 95.3%.

Conclusions: The HPI5-7 may offer some useful insights. Values ≤85% carry a clinically acceptable NPV for hypotensive events at the observed prevalence and may represent a "safe zone" during surgery. Values >85% do not carry enough PPV to trigger hemodynamic interventions, but represent a warning signal. Values >98% are highly suggesting a hypotensive event after 5-7 minutes. Further studies exploring the predictive ability of the HPI at different times are needed.