Hypotension Prediction Index (HPI) in Lung Resections

The Hypotension Prediction Index (HPI) is a hemodynamic score designed specifically for prediction of the intraoperative hypotension (IOH) episodes. It is based on an algorithm programmed into Edwards Lifesciences HemoSphere monitor clinical platform (Irvine, CA, USA). The HPI is based on a continuous analysis of an arterial pressure waveform. It is processed in addition to the FloTrac algorithm via proprietary Acumen IQ Sensor and uses an artificial intelligence technology. After internal validation, the algorithm was prospectively, externally and clinically validated in general surgical, perioperative patients, cardiovascular surgical patients, and mechanically ventilated COVID-19 ICU patients.

As opposed to conventional monitoring systems, which display physiological parameters in real life, an HPI algorithm detects the earliest changes, multivariate variability and interactions in the physiologic inter-related data on preload, afterload, and contractility to deliver an index predicting an upcoming hypotensive event. Variables used by the patent-protected algorithm to calculate HPI are as follows: heart rate variability (changes in heart rate/changes in MAP); arterial pressure waveform complexity (approximate waveform entropy, sample waveform entropy, frequency domain measure of higher order harmonics); preload parameters (pulse pressure variation PPV, stroke volume variation SVV); contractility parameters (slope of the ascending part of the pressure waveform above time, dP/dt); and afterload parameters (SVR, dynamic arterial elastance Eadyn), but their relative contribution to final index is not revealed.

Final index values of HPI range from 1 to 100, with increasing numbers representing a greater likelihood of an impending hypotensive event. These events are defined as mean arterial pressure (MAP) <65 mmHg occurring for over one minute. HPI values predict the occurrence of hypotension five to fifteen minutes before the event, with sensitivity and specificity in both time-frames of greater than 80%. In most studies, a value of 85 HPI predicts a hypotensive episode, and this value is arbitrarily preprogrammed into the HemoSphere monitor to alert the clinician and allow proactive responses to minimize or even entirely prevent intraoperative hypotension.

Parameters used and incorporated into the HemoSphere monitor can guide a clinician in the optimal management of IOH. These "secondary screen" variables include the left ventricular contractility parameter (dP/dt), dynamic preload parameter (SVV) and afterload parameter dynamic arterial elastance Eadyn.

Maximal left ventricular (LV) pressure rise (LV dP/dt max) is a classical marker of LV performance and systolic function. It is conventionally defined as the change in pressure in the left ventricular cavity over the isovolumetric contraction period and it originally requires LV catheterization. In clinical practice a surrogate peripheral arterial pressure waveform is used to estimate dP/dt value and to predict the need for inotropic support.

SVV is a dynamic preload parameter and represents the difference in the left ventricular stroke volume secondary to changes in intrathoracic pressure induced by mechanical ventilation.

The dynamic arterial elastance Eadyn represents the proportion of pulse-pressure variation (PPV) to SVV. It can be used to assess vascular tone, which can predict arterial pressure response after volume loading and/or potential response to vasopressor administration.

Both PPV and SVV are considered superior to static indices to predict fluid responsiveness. They are both based on heart-lung interactions and reflect hemodynamic cyclic changes induced by mechanical ventilation in the closed-chest condition. Their values are significantly correlated with the magnitude of VT. The current low-tidal volume intraoperative ventilatory strategy protects the lungs, but at the same time lowers the reliability of dynamic indices, particularly in open-chest conditions. Due to limited changes in intrathoracic pressure during the respiratory cycle in open lung conditions, there is a risk of receiving false negative parameter values. PPV and SVV seem to be inaccurate in predicting fluid responsiveness in an open-chest setting during cardiothoracic surgery.

The HPI was validated in general surgery and ICU cases, but not in thoracic surgery one-sided open chest procedures. These procedures include not only significantly abnormal physiologic conditions (open pleura and one-lung protective ventilation) but also a high incidence of sudden manual surgical interventions. All these factors can significantly influence and compromise the HPI performance.

The aim of this study is to validate the HPI technology in open-chest lung resection procedures with the use of one-lung ventilation. The study group will comprise 60 consecutive adult patients qualified for lung resection procedures under general anesthesia with open-chest and one-lung ventilation.

The patients will be monitored during the operation using standard invasive hemodynamic monitoring with arterial pressure transducer and concomitantly with HemoSphere monitor with the HPI software attached to the Acumen IQ transducer (Edwards LifeSciences, Irvine, CA, USA). The clinicians will be blinded to the output of the HemoSphere monitor. Hemodynamic waveforms and HPI prediction data including hypotensive events (IOH) will be recorded from the time of arterial cannula insertion until leaving the operation room. Recorded data will be divided into seven cohorts, represented by separate time frames:

1. Pre-induction baseline, supine, spontaneous breathing (if available and arterial cannula inserted pre-induction)
2. Supine, closed-chest anaesthetized, intubated, two-lung ventilation
3. Lateral decubitus, closed chest, two-lung ventilation
4. Lateral decubitus, closed chest, one-lung ventilation (OLV)
5. Lateral decubitus, open chest, one-lung ventilation (OLV)
6. Lateral decubitus, closed chest, two-lung ventilation post-resection
7. Supine, closed-chest, two-lung ventilation

We will estimate the sensitivity (recall) and positive predictive value (precision) of the HPI algorithm and describe the number of false alarms as well as missed events without explicitly referring to specificity or negative predictive value.

Study conduct and reporting will be performed under the STARD guidelines.