Impact of hypotension prediction index on mean arterial pressure and stroke volume index during pancreatic surgery: a retrospective pre-post observational pilot study

Introduction

Intraoperative hypotension (IOH) is a relatively common event occurring during general anesthesia, associated to increased perioperative morbidity and mortality (1). Importantly, previous studies showed that as little as 5 min of IOH are associated to increased risk of postoperative complications (2), including renal insufficiency and myocardial injury (3). In this setting, hemodynamic optimization using advanced monitoring and goal-directed therapy (GDT) approaches already proved to be effective in reducing complications incidence in surgical patients (4, 5). However, IOH often occurs suddenly and without any warning signs, generally resulting in a reactive approach to hemodynamics derangement rather than a proactive strategy.

Recently, the use of a machine-learning developed algorithm allowed the identification of specific signatures associated with impending IOH (6, 7). An easy-to-interpret Hypotension Prediction Index Software (Acumen^TM HPI - Edwards Lifesciences, Irvine, CA, USA) became available by fitting these signatures in a mathematical model. The HPI is an adimensional number ranging from 0 to 100. The higher the value, the higher the probability of the hypotensive event occurrence and the shorter the onset time. Some studies have already shown that it can predict IOH events up to 15 min before the actual occurrence (6, 8, 9).

Limiting IOH is essential in high-risk patients undergoing major surgery. Among non-cardiac operations, major pancreatic surgery is complex and long-lasting. In our center, roughly 200 patients per year undergo major pancreatic surgery, and in our experience, these patients are often frail, debilitated, and have reduced physiological reserves at the time of surgery (10). In addition, postoperative complications, including acute kidney injury (AKI), bleeding, sepsis, shock, bilio-digestive fistulas, and anastomotic leaks, are common in this setting (11). In these high-risk patients, IOH limitation and optimal hemodynamic management may contribute to risk reduction of postoperative organ dysfunction incidence (12).

To date, most of the published research on the use of HPI has been performed on heterogeneous surgical cohorts. Additionally, to the best of our knowledge, HPI monitoring has not yet been specifically assessed in pancreatic surgery. Some studies have shown that an HPI-GDT protocol results in reduced IOH compared to GDT alone (13–17). No data are available to understand whether this occurs in patients undergoing pancreatic surgery and if it has an impact on cardiac output and/or a reduced incidence of postoperative complications. The objective of this single-center retrospective study was to compare intraoperative hemodynamic parameters and postoperative clinical outcomes before and after the implementation of HPI monitoring in patients undergoing pancreatic surgery. In accordance with its pilot and exploratory design, the study was principally conceived to furnish effect size estimates and feasibility data to inform the design, sample size assumptions, and clinical endpoints of future randomized controlled trials.

Methods

This retrospective study was approved by the institutional independent review board (Ethics Committee approval 15/23), adheres to the principles of the Declaration of Helsinki (The World Medical Association Inc., Ethical Principles for Medical Research Involving Human Subjects), and was registered on the ClinicalTrials.gov database (NCT06097052). All participants gave written informed consent for using perioperative data upon admission. Therefore, given the study's retrospective nature, specific informed consent for the study was waived. Demographic, surgical intra- and postoperative clinical data were retrieved from a prospectively maintained database. A review of the electronic anesthesia records yielded pre-operative anesthesia data. In contrast, intraoperative hemodynamic data from the hemodynamic monitor were downloaded and stored on a password-protected local server at the end of each surgery. These data are accessible upon reasonable request to the authors.

Study population

Patients undergoing pancreaticoduodenectomy, total pancreatectomy, or distal pancreatectomy at IRCCS Humanitas Research Hospital (Rozzano, Italy) between March 2022 and October 2023 were included in this study. Inclusion criteria were as follows: age >18 years, laparotomic major pancreatic surgery, invasive blood pressure monitoring and GDT monitoring, total surgical time >360 min. Exclusion criteria were age >90 years, ASA class >III, urgent or emergency surgery.

Until March 2023, subjects were monitored using the FloTrac sensor for invasive arterial hemodynamic monitoring (Edwards Lifesciences, Irvine, CA, USA). These patients were included in the FloTrac cohort. Conversely, after March 2023, patients received the invasive arterial hemodynamic monitoring by the Acumen IQ sensor (HPI group). Both sensors were used with the hemodynamic platform EV1000 or HemoSphere (Edwards Lifesciences, Irvine, CA, USA). The perioperative anesthetic and surgical management is delineated in Appendix 1.

Study protocol

Figure 1 summarizes the intraoperative FloTrac- and HPI-GDT protocols. Following general anesthesia induction, advanced waveform analysis data were acquired from continuous blood pressure waveforms after arterial line catheterization (radial artery). The induction period was excluded from the analysis because anesthesiologists did not initiate hemodynamic monitoring in a standard manner. Data recording was terminated at the end of surgery (skin closure).

After recovery from general anesthesia, patients were monitored in a post-anesthesia care unit and subsequently discharged either to the Intensive Care Unit (ICU) or to the surgical ward, depending on the judgment of the anesthesiologist in charge.

In both groups, hemodynamic management aimed to maintain mean arterial pressure (MAP) ≥ 70 mmHg, according to the institutional GDT protocol historically applied with FloTrac monitoring and consistently adopted in the HPI-guided strategy to ensure comparability. While HPI predicts hypotension, defined as a MAP of less than 65 mmHg, interventions were triggered at a higher threshold as part of a preventive management approach. We defined a fluid challenge as a 4 ml/kg bolus of crystalloid solution administered in <15 min. Vasoconstrictors and inotropic drugs included etilefrine, ephedrine, norepinephrine, and dobutamine. EV1000 and HemoSphere raw hemodynamic data analysis was performed by means of Microsoft Excel® (Microsoft Corporation, Redmond, WA, USA). A linear interpolation function was used whenever continuous arterial pressure recording was interrupted for more than 20 s.

The primary outcome of this study was a difference in the time-weighted average (TWA) of MAP < 65 mmHg between the two cohorts. TWA-MAP < 65 mmHg was calculated as described elsewhere (3). It has previously been used as a representative index of hypotension's duration and severity (13, 15, 18). IOH was defined as a MAP < 65 mmHg for >1 min (3), while severe IOH was MAP < 50 mmHg for >1 min. Hypertension was defined as a MAP > 100 mmHg for >1 min. These indices were calculated as previously described (15).

In the HPI-guided group, the number of HPI alerts was documented as instances in which the HPI value exceeded 85, a threshold associated with a high probability of imminent hypotension within the following minutes. Conversely, hemodynamic reassessment and interventions within the study protocol were initiated at a lower threshold (HPI > 65), reflecting a preventive management approach.

To analyze the possible different impact of the two GDT protocols on hemodynamics, several parameters were considered. In particular, the stroke volume index (SVI) was analyzed to investigate whether the HPI protocol, primarily focused on MAP, could determine a different trend. Inadequate SVI was considered for values <35 ml/m² for >1 min.

Secondary outcomes were: SD and generalized ARV of MAP and SVI were used to describe the variability of these hemodynamic indices. Generalized ARV of MAP was calculated as proposed by Mascha et al. (19). Generalized ARV of SVI was calculated in the same way.

• - Mean MAP
• - Frequency of IOH events (expressed as the number of hypotensive events per hour)
• - Absolute duration of MAP < 65 and >100 mmHg and as a percentage of surgical time
• - Average time of hypotensive and hypertensive (>100 mmHg) events
• - Area under the curve of MAP (AUC-MAP) < 65 and >100 mmHg
• - TWA-MAP < 50 mmHg and TWA-MAP > 100 mmHg
• - Generalized average real variability of MAP (ARV-MAP)
• - Standard deviation of MAP (SD-MAP)
• - Generalized average real variability of SVI (ARV-SVI)
• - Standard deviation of SVI (SD-SVI)
• - Mean SVI and mean SVI < 35 ml/m²
• - TWA-SVI < 35 ml/m²
• - Frequency of SVI < 35 ml/m² per hour
• - Average time of SVI < 35 ml/m²

As an exploratory analysis, the following clinical outcomes were considered: ICU admission rate, length of ICU stay, red blood cell or plasma transfusions rate, Clavien-Dindo classification of postoperative course, rate of postoperative AKI according to the Joint Consensus report of the Acute Disease Qualitive Initiative and PeriOperative Quality Initiative (20), length of hospital stay, in-hospital mortality and 90-day mortality.

We used the STROBE case-control checklist when writing this manuscript (information on the STROBE Initiative is available at www.strobe-statement.org).

Sample size calculation and statistical analysis

This pilot study was designed to test the feasibility of a large randomized controlled trial comparing two GDT protocols: a standard protocol based on the use of FloTrac and a protocol based on the HPI technology. The objectives of the randomized trial will be to test whether an HPI-GDT protocol improves hemodynamic stability in the intraoperative phase (less hypotension, lower cardiac output, less hemodynamic variability) and improves outcomes in the postoperative phase (fewer complications, shorter hospital stay). For sample size calculation, in line with the published literature, we hypothesized that the HPI-GDT protocol could reduce the mean TWA-MAP < 65 mmHg by approximately 50%. We considered 1 mmHg as a reference value of TWA-MAP < 65 mmHg for patients who underwent pancreatic surgery and were managed with the FloTrac-GDT protocol in our center (unpublished raw data extrapolated from a cohort of 10 patients). To detect a significant difference in TWA-MAP < 65 mmHg between the HPI and the FloTrac group of 0.5 mmHg with a standard deviation of 0.4 mmHg, the study required a sample size of 14 cases for each group to achieve a 90% power and 5% significance level (two-tailed). The sample size calculation was performed with Statulator statistical calculator (Dhand K and Khatkar M, 2023—freely available at www.statulator.com). Given the retrospective design of the study, the protocol allowed for more patients to be enrolled than the sample size calculation to improve the estimation of the variables under study.

Continuous variables are expressed as mean (standard deviation, SD) or median [interquartile range, IQR]. Categorical variables are expressed as absolute value and percentage. Statistical differences in primary and secondary outcomes were analyzed using unpaired Student's t-test or Mann–Whitney test, depending on the respective data distribution. Chi-square or Fisher exact tests were used for the comparison of discrete variables. The P critical value was set at 0.05. Statistical analysis was performed with PRISM 8 (Graphpad Software, Boston, MA, USA).

Results

In this study we enrolled 66 subjects. After an appropriate review of medical records, 18 did not meet the inclusion criteria or had at least one exclusion criterion. The final study population included 48 subjects. Among these, 20 were monitored using the FloTrac sensor and 28 using the Acumen IQ. The two groups were comparable for age, sex, anthropometric characteristics and ASA class score (Table 1).

Table 1. Demographic and intra-operative characteristics of the study populations.

BMI indicates body mass index; BSA, body surface area; ASA, American Society of Anesthesia classification; PD, pancreaticoduodenectomy; CVD, cardiovascular disease. Continuous variables are reported as mean (SD) or median [IQR]. Continuous variables were analyzed using student t-test or Mann–Whitney test. Categorical variables using Fisher's exact test.

Table 2 summarizes the results for primary and secondary hemodynamic outcomes.

Table 2. Primary and secondary outcomes.

TWA, indicates time-weighted average; MAP, mean arterial pressure; AUC, area under the curve; ARV, general average real variability; SVI, stroke volume index; SD, standard deviation. Values are reported as median [IQR]. Continuous variables were analyzed using Mann–Whitney test.

HPI alerts during anesthesia monitoring occurred on average 2.6 ± 1.1 times per hour. The HPI group was characterized by a reduction of median TWA-MAP < 65 mmHg (p = 0.036), of absolute time spent with MAP < 65 mmHg (p = 0.031) and median AUC-MAP < 65 mmHg (p = 0.017) (Figure 2). The HPI also displayed increased intraoperative median MAP (p = 0.016), absolute time spent with MAP > 100 mmHg and AUC-MAP > 100 mmHg compared to the FloTrac group.

The TWA-MAP < 50 mmHg, the number of hypotensive events normalized per hour of anesthesia monitoring, and the SD-MAP were comparable between the two groups. The HPI group was characterized by higher ARV-MAP (p = 0.001).

The mean SVI, the number of events characterized by SVI < 35 ml/m² and the TWA-SVI < 35 ml/m² were similar between the FloTrac and HPI groups. Figure 3 shows the plot of TWA-MAP < 65 mmHg and TWA-SVI < 35 ml/m² for both study groups.

Table 3 summarizes the main intraoperative and postoperative outcomes that were analyzed. Intraoperative blood loss was similar between the two groups, as the rate of packed red blood cells or fresh frozen plasma transfusion. In the postoperative period blood components were transfused with similar occurrences. Norepinephrine or dobutamine administration was more frequent in the HPI group although this difference was not statistically significant. Postoperative ICU admissions were similar between FloTrac and HPI group (10% vs. 29% respectively, p = 0.160). One patient died in the FloTrac group before hospital discharge. Postoperative surgical and non-surgical complications were similar in both groups. According to the Clavien-Dindo classification, most of the complications were graded as I or II. Following hospital discharge, the mortality rate at 90 days was found to be 0% in both study groups.

Table 3. Intraoperative and postoperative clinical outcomes.

Continuous variables are reported as median [IQR]; RBC indicates packed red blood cells; FFP, fresh frozen plasma; ICU, intensive care unit. P values were obtained using Mann–Whitney test or Fisher's exact test, whenever appropriate.

Discussion

This pilot study found that the introduction of an HPI-triggered GDT protocol reduced the TWA-MAP < 65 mmHg, increased median intraoperative MAP, TWA-MAP > 100 mmHg, and ARV-MAP compared to a MAP-triggered GDT protocol in patients who underwent pancreatic surgery. Notably, these effects were not accompanied by a notable increase in the utilization of vasopressors and inotropic infusion. Further, no significant difference in SVI values were found between the two GDT protocols.

Our investigation focused on a specific patient population. Pancreatic surgery patients are a high-risk surgical population, as they are exposed to long-lasting and highly complex and extensive surgery. Prolonged surgical time increases the risk of IOH, which has been linked to perioperative cardiovascular morbidity and mortality (2). In our study, both cohorts were characterized by advanced age and a high prevalence of comorbidities, with a substantial proportion of ASA III patients. This reflects the intrinsic risk profile of major pancreatic surgery and further emphasizes the importance of strategies aimed at minimizing intraoperative hypotension in this vulnerable population, in whom even transient hemodynamic instability may have clinically relevant consequences.

Machine-learning approaches, such as the HPI system, are setting conditions for a new approach to hemodynamic monitoring, promoting a transition from a reactive to a preventive management of IOH. Studies from different investigators showed that the use of HPI was associated with reduced number of IOH events (14, 21), time and severity of hypotension (13, 15, 18) in non-cardiac surgery. In line with these findings, we similarly found that the HPI group had lower TWA-MAP < 65 mmHg and absolute hypotension time, along with increased median intraoperative MAP compared to the FloTrac group. We also observed a trend towards fewer hypotensive events, consistent with previous literature (16, 17). We did not find differences in TWA-MAP < 50 mmHg, considered a severe hypotension measure. This is a reasonable finding, as we routinely use a FloTrac-GDT approach that already prompts the anesthesiologist intervention for MAP values <65 mmHg, limiting more severe hypotensive events. Furthermore, both groups exhibited a shared objective of maintaining a mean arterial pressure (MAP) level of at least 70 mmHg, a target that aligns with the institution's goal-directed therapy protocol. This consistent target across groups supports the comparability of the two strategies and suggests that the observed differences are unlikely to be driven by different therapeutic goals, but rather by differences in monitoring approach and timing of interventions.

In our study, the median MAP and the TWA-MAP > 100 mmHg were significantly higher in the HPI group. Similar findings were previously reported (9, 15) and some authors already raised concerns about the risk of overtreatment of patients (22). A recent randomized controlled trial found that HPI use was associated to increased rate of norepinephrine continuous infusion (23). To date, we cannot exclude that increased utilization of vasoactive drugs or crystalloids infusion in similar settings may have side effects. In our study we did not observe any difference in utilization of vasopressors or inotropes between the two groups suggesting that the aforementioned effects may be attributed to a better timing in IOH management. It should be noted, however, that we recorded a trend toward greater use of these drugs in the HPI group that might be more significant in a larger study. Indeed, the use of an HPI threshold of 65/100, which is below 85/100, which was used in other studies (14, 15, 17), may have promoted early reassessment of the patient's hemodynamic state. Despite the absence of systematic objective timing metrics due to the retrospective design, this mechanistic interpretation aligns with the conceptual framework of predictive hemodynamic monitoring and warrants further investigation. Future prospective studies could incorporate time-to-intervention metrics to better characterize this effect and confirm whether earlier intervention represents the primary driver of improved hemodynamic outcomes. It is worth mentioning that, to the best of our knowledge, there are no data showing that a TWA-MAP > 100 mmHg is associated with an increase in perioperative complications. We believe that it may have a bearing from this point of view to the fact that most of the patients included in the study were hypertensive. For such patients, MAP values above 100 mmHg might therefore be particularly close to those of daily life and therefore not harmful in the intraoperative period. Indeed, we did not observe an increase in the occurrence of postoperative complications.

To better estimate the hemodynamic impact of using the HPI-guided protocol, we analyzed the hemodynamic variability in the two study groups. We adopted the generalized ARV parameter for this purpose because it is considered more reliable than SD for variability estimation of time-series data (19). The generalized ARV-MAP was higher in the HPI group, but median values for both cohorts were in a range that was not related to increased complications and mortality, referring to the study by Mascha et al. which included data from more than 100,000 patients (19). The higher generalized ARV-MAP value in the HPI group could be related to a more aggressive hemodynamic management that may have frequently altered the MAP trend to higher values, as evidenced by the mean MAP and TWA-MAP > 100 mmHg values of that group. SD-MAP values, on the other hand, were not different between the two groups, but this could be expected as this parameter does not consider timing of measurements.

In this study protocol, the attention of the anesthesiologist was drawn to rapid and effective treatment of IOH using an HPI-triggered protocol. Since an increase in systemic vascular resistances and therefore cardiac afterload can negatively impact cardiac output and blood flow, this study also analyzed the performance of the SVI. We found that the TWA-SVI < 35 ml/m² values and median time of SVI < 35 ml/m², which are both markers of relatively low flow state, displayed higher values in the HPI group, although they were not significantly different. Also, the variability of SVI, measured by means of ARV-SVI and SD-SVI, was not different between the two study groups. These findings suggest that our FloTrac- and HPI-GDT protocols, with different trigger for hemodynamic reassessment, do not lead the anesthesiologist to prioritize blood pressure despite blood flow, although future larger studies will better clarify this aspect.

Regarding the postoperative complications rate, we did not find any significant difference between the two groups. This is not surprising because the study was not powered to explore this outcome. It will be necessary a large sample size study to compare this variable between the HPI-triggered group and a well-established FloTrac GDT protocol. Conversely, the study of Solares et al. conducted on 104 patients who underwent mainly trauma/orthopedics surgery, showed that the employment of HPI was associated to reduced post-operative complications occurrence and a shorter hospital length-of-stay (24). Apart from this study, there is currently a lack of data demonstrating the association between HPI algorithm use and a reduction of postoperative complications, as reported by recent meta-analyses (25–27). We believe that future studies should investigate this important clinical outcome.

Our study has several limitations. This is a single-center retrospective study including a relatively small number of subjects. These aspects limit the generalizability of the results. Further, the small sample size limited the possibility to perform age- and comorbidities-based analyses, which is an interesting line of research that should be pursued in future larger studies. We did not report the amount of fluid administered in the two groups because the retrospective design of the study would limit the reliability of the data too much. The substantial variability in intraoperative fluid balance, including negative values, may be indicative of the interplay between fluid administration, blood loss, and intraoperative diuresis. However, given the retrospective design and the potential variability in data collection and recording, these values should be interpreted with caution and considered as limitations of the study. The pre-post study design, the increased familiarity with the HPI platform over time and the standardized intervention protocol may have helped managing more effectively the hemodynamics of all our patients. Finally, given the retrospective and pilot design, the sample size calculation was based on preliminary institutional assumptions that differed from the observed data. Specifically, the expected TWA-MAP < 65 mmHg in the control group was estimated at 1 mmHg, whereas the observed value was substantially lower (0.55 mmHg), with a further reduction in the HPI group (0.32 mmHg). This discrepancy is indicative of the inherent uncertainty of a priori estimates in retrospective settings. Consequently, the study is best interpreted as hypothesis-generating, providing effect size estimates to inform the design and sample size assumptions of future randomized controlled trials. Despite these limitations, our data indicate a decrease of TWA-MAP < 65 mmHg from 0.55 mmHg with standard FloTrac-guided therapy to 0.32 mmHg under HPI-guided therapy. This represents a mean absolute reduction of 0.23 mmHg. The pooled standard deviation was approximately 0.36 mmHg, indicating a moderate effect size (Cohen's d ≈ 0.63). Utilizing a two-sided alpha level of 0.05 and aiming for 80% power, a parallel 1:1 randomized controlled trial would necessitate approximately 40 patients per arm (80 total) to detect this difference. Allowing for a 10% attrition rate, the enrollment target is approximately 88 participants, with 44 allocated for each group. For a higher power of 90%, the sample would increase to approximately 53 patients per arm, resulting in a total of 106 patients. These figures provide a concrete basis for designing a definitive randomized controlled trial (RCT) to confirm the clinical benefit of HPI-guided goal-directed therapy in pancreatic surgery.

In conclusion, hemodynamic management based on the HPI algorithm in patients undergoing long-term major surgery such as pancreatic surgery seems to reduce intraoperative hypotension compared to patients managed with a standard GDT protocol without significant impact on stroke volume index. Further studies are necessary to verify these results and whether this may also have an impact on postoperative complications occurrence.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Plinio Cirillo, University of Naples Federico II, Italy

Reviewed by: Reza Eshraghi, Isfahan University of Medical Sciences, Iran

Marcell Szabó, Semmelweis University, Hungary