Background

The Hypotension Prediction Index is designed to timely predict intraoperative hypotension and is based on arterial waveform analysis using machine learning. It has recently been suggested that this algorithm is highly correlated with the mean arterial pressure (MAP) itself. Therefore, the aim of this study was to compare the Index with MAP based prediction methods and it is hypothesized that their ability to predict hypotension is comparable.

Methods

In this observational study, the Hypotension Prediction Index was used in addition to routine intraoperative monitoring during moderate- to high-risk elective non-cardiac surgery. The agreement in time between the default Hypotension Prediction Index alarm (>85) and different concurrent MAP thresholds was evaluated. Additionally, the predictive performance of the Index and different MAP based methods were assessed within five, ten and fifteen minutes before hypotension occurred.

Results

A total of 100 patients were included. A MAP threshold of 73 mmHg agreed 97% of the time with the default Index alarm, while a MAP threshold of 72 mmHg had the most comparable predictive performance. The areas under the receiver operating characteristic curve of the Hypotension Prediction Index (0.89 (0.88-0.89)) and concurrent MAP (0.88 (0.88-0.89)) were almost identical for predicting hypotension within five minutes, outperforming both linearly extrapolated MAP (0.85 (0.84-0.85)) and delta MAP (0.66 (0.65-0.67)). The positive predictive value was 31.9 (31.3–32.6)% for the default Index alarm and 32.9 (32.2–33.6)% for a MAP threshold of 72 mmHg.

Conclusion

In clinical practice, the Hypotension Prediction Index alarms are highly similar to those derived from MAP, which implies that the machine learning algorithm could be substituted by an alarm based on a MAP threshold set at 72 or 73 mmHg. Further research on intraoperative hypotension prediction should therefore include comparison with MAP based alarms and related effects on patient outcome.