Computer-assisted Individualized Hemodynamic Management Reduces Intraoperative Hypotension in Intermediate- and High-risk Surgery: A Randomized Controlled Trial

This single-center, prospective, two-arm, parallel, randomized controlled superiority study was approved by the institutional review board of Bordeaux (Comité de Protection des Personnes Sud-Ouest et Outre Mer III, Bordeaux, France No. DC2015/117), and the study protocol was published on clinicaltrials.gov (NCT03965793) on May 29, 2019, before patient enrollment began (Principal Investigator: Alexandre Joosten). Importantly, since 2018, clinical research protocols are not reviewed by the local institutional review board but rather are randomly directed to a different institution’s review board in France to reduce bias in reviews. The study was conducted at Bicêtre Hospital from October 28, 2019, through June 26, 2020. All patients were approached by the principal investigator, and after presentation of the study purposes, written informed consent was obtained before inclusion.

Adult patients scheduled for an elective intermediate- to high-risk abdominal or orthopedic surgical procedure¹⁹  who were expected to be managed according to an individualized hemodynamic protocol using a radial catheter coupled to an advanced hemodynamic monitoring device (EV1000, Edwards Lifesciences, USA) as part of their anesthetic care were eligible for this protocol. Noninclusion criteria were minor patients (less than 18 years old), pregnancy, cardiac arrhythmias, and refusal to participate. The trial was conducted in accordance to the original protocol, and as a result, no change has been made in the protocol after trial commencement.

Randomization assignments were generated without restriction and were generated on October 24, 2019, by our research nurse using an internet-based software (http://www.randomization.com), to either closed-loop vasopressor and decision-support–guided fluid therapy (computer-assisted group) or manually adjusted goal-directed therapy group. Opaque envelopes in recruitment sequence were then created by an unaffiliated research nurse. The morning of the surgery, a sealed opaque envelope containing the assigned patient number was opened. The envelopes were kept in the research office of Bicêtre Hospital. Patients were blinded to group allocation, but anesthesia providers were not. However, outcome data were collected by collaborators blinded to both the study group allocation and the reasoning for the research protocol. Importantly, as stipulated by the institutional review board, the principal investigator could not be the primary anesthesiologist but should be present to supervise the computer-assisted systems. Patients in the manually adjusted goal-directed therapy group were managed by an anesthesiologist not involved in the current study who applied the individualized hemodynamic protocol (manually) that is routine in our institution for these surgical procedures (fig. A1).

Preoperatively, all patients stopped taking angiotensin-converting enzyme inhibitor drugs and/or angiotensin receptor blocker drugs 48 h before surgery per institutional standard of care. Standard monitoring was applied to all patients and included a three-lead electrocardiogram, noninvasive pulse oximetry, standard arm arterial pressure cuff, capnography, central temperature assessment (esophageal probe) and a depth of anesthesia monitor (BIS, Medtronic, France). A radial artery catheter was also inserted during induction and linked via a Flotrac sensor to a pulse contour analysis hemodynamic monitor (EV1000). Anesthesia was induced with sufentanil (2 µg/kg) and propofol (2 mg/kg). Atracurium (0.6 mg/kg) was administered for intubation, and 10-mg boluses were added as needed to maintain the train-of-four ratio less than 2 (TOF Scan Technology, Idmed, France). Anesthesia was maintained with sevoflurane to keep a BIS value between 40 and 60. Sufentanil boluses (0.1 to 0.2 µg/kg) could be administered at the discretion of the primary anesthesia provider. All patients received mechanical ventilation using a volume control mode with tidal volumes of 7 to 8 ml/kg of predicted body weight and a respiratory rate adjusted to achieve an end tidal carbon dioxide between 34 and 38 cm H₂O. Recruitment maneuvers were done at the discretion of the anesthesiologist in charge of the patient. Prophylactic antibiotics and antiemetics were administered 30 min before surgical incision.

It is common practice in our hospital that fluid management for intermediate- to high-risk abdominal and orthopedic surgical patients be standardized This is achieved with a baseline maintenance infusion of Ringer’s lactate solution (Fresenius Kabi, France) set at a rate of 2 to 4 ml · kg^–1 · h^–1 (laparoscopic vs. open) and additional fluid boluses based on an advanced uncalibrated hemodynamic device (EV1000). This monitor provides flow-based variables (SV index and cardiac index) and SV variation, a dynamic predictor of fluid responsiveness. To avoid intraoperative hypothermia, a forced-air warming system (Bair Hugger, SEBAC, France) and a blood-fluid warming system (Fluido compact, SEBAC, Flaxlanden) were utilized during the surgical procedure. Packed red blood cells were infused to maintain the hemoglobin concentration between 7 and 9 g/dl, depending on each patient’s status and comorbidities.

For postoperative pain management, all patients who had a laparotomy had a thoracic epidural catheter inserted by the anesthesiologist before induction. The epidural was activated after a test dose (4-ml lidocaine 2% with 1:200,000 epinephrine) and was infused with ropivacaine 2 mg/ml in a bag of 200 ml of normal saline into which 10 mg of morphine was added. The mixture was infused at a rate of 5 to 6 ml/h from skin incision until postoperative day 3.

In the manually adjusted goal-directed therapy group, vasopressor and fluid administration was adjusted manually by anesthesia providers and/or nurse anesthetists using our institutional individualized hemodynamic protocol. This protocol consists of two main components: (1) individualized MAP control starting at anesthesia induction and continued until the end of the surgery and (2) SV index maximization using mini-fluid challenges of 100-ml fluid boluses. Details regarding this individualized hemodynamic protocol are shown in figure A1.

In the computer-assisted group, the same strategy was applied but using a computer-assisted system including the following:

1. A closed-loop vasopressor system was used to titrate vasopressor administration. Details on this system are described in our previous publications^{16,17,20–23}  and in the online Supplemental Digital Content 1 (https://links.lww.com/ALN/C620). As review, MAP values from an EV1000 monitor were collected by the closed-loop vasopressor, and the proportional, integral, and derivative errors (if any) were converted into a dose titration of a norepinephrine infusion. The closed-loop vasopressor controller itself is a hybrid proportional integral derivative and rules-based system. The algorithm was coded in Visual C (Microsoft, USA), and software version 2.93 of the closed-loop system was used exclusively for patients enrolled in the computer-assisted group. The system was run on an Acer (Taiwan) laptop using Windows 7 (Microsoft). It collected MAP data from the EV1000 monitor via a serial RS232 connection and controlled a Chemyx Fusion 100 syringe pump (Chemyx Inc., USA) via USB. Importantly, a second norepinephrine syringe not associated with the system was also available to the primary team if the system experienced errors or for emergency use. Per protocol, the primary team was only to give “rescue” boluses of vasopressors in the event the system performance was inadequate.

2. A real-time clinical decision support system called “assisted fluid management” (AFM, Edwards Lifesciences) was used to guide mini-fluid challenges. Briefly, this system suggests to the anesthesia provider that a fluid bolus should be administered, analyzes the effects of the bolus, and continually reassesses the patient for further fluid requirements.²⁴  The AFM algorithm core is the same as the closed-loop algorithm we used in previous publications; 100% compliance with the AFM prompts would result in identical treatment that achieved by closed-loop control, although with minimal time delays for acceptance of recommendations. Figure 1 presents the schematic for the computer-assisted group. In both groups, the norepinephrine infusion was administered through an intravenous line not used for any other purpose. Figure A2 shows the computer-assisted set-up in the operating room at Bicêtre Hospital. Importantly, in both groups, no other vasopressor than norepinephrine was allowed.

The primary outcome measure was intraoperative hypotension defined as the percentage of intraoperative case time patients spent with a MAP less than 90% of the patient’s baseline value, measured during the preoperative screening. This cutoff has been chosen based on the implementation of our individualized hemodynamic protocol already in place in our institution a few months before the beginning of the current study. The primary outcome was calculated on a per-patient basis as follows:

It was assumed that with the modest sample size, we would be unlikely to see differences in major complications, so our secondary outcome measure was the incidence of minor postoperative complications measured at postoperative day 30 (including postoperative nausea and vomiting, delirium, wound infection, urinary infection, pneumonia, acute kidney injury, paralytic ileus, other infections, and readmission to hospital within 30 days postsurgery). These complications have been defined in our previous publications.^25,26 

Other hypothesis-generating outcome measures included major complications (at postoperative day 30), percentage of case time with a MAP less than 65 mmHg, percentage of case time “in target” (MAP ± 10 mmHg of the baseline MAP), percentage of case time above target (MAP greater than 10 mmHg), mean SV variation, cardiac index, SV variation, SV index during the first and last 30 min of the case, and the percentages of case time with a SV variation less than 13%, with a cardiac index less than 2 l · min⁻¹ · m⁻², and with a SV index less than 30 ml · m⁻². We also recorded total intraoperative volumes of fluids administered, net fluid balance, total doses of norepinephrine given, the number of norepinephrine rate modifications during surgery, lactate values measured before skin incision and at arrival in the postanesthesia care unit, and lengths of stay in the postanesthesia care unit, intensive care unit, and the hospital. No change in the outcomes has been made after trial commencement.

All hemodynamic variables were collected at 20-s intervals via the EV1000 monitor. The study started once the radial artery catheter had been inserted and connected to the EV1000 monitor, approximately 5 min after anesthesia induction.

Power calculation was performed based on the primary objective. Our data showed that patients spent 12 ± 8% of intraoperative case time with a MAP of less than 90% of their preoperative value when norepinephrine was titrated manually.²⁷  To detect a 50% reduction in the time spent in hypotension in the computer-assisted group, the study needed 19 patients/group (38 patients total) to achieve 80% power with Welch’s unequal variances t test and bilateral α risk fixed at 5%. No dropout was taken into account.

The data were analyzed using an intention-to-treat approach and in a blinded fashion. Normality of data was evaluated using the Kolmogorov–Smirnov test. Variables normally distributed were compared with independent samples t test and are expressed as means ± SD. Nonnormally distributed variables were compared with a Mann–Whitney U test and were expressed as median [25% to 75%] percentiles. Discrete data are expressed as numbers and percentages and were compared using a chi-square or Fisher’s exact test when indicated.

The primary outcome was evaluated using a Mann–Whitney U test; differences between groups with 95% CI were also calculated. The secondary outcomes were evaluated using a chi-square test. The analyses were not adjusted for additional variables. No interim analysis was planned on the data. Statistical significance was set at P < 0.05, and all tests were two-tailed. Analyses were performed using Minitab (France).

Between October 28, 2019, and June 26, 2020, 133 patients were screened for eligibility, and 38 patients were enrolled and randomized. As a result, the trial was not stopped before obtaining the sample size goal. Reasons for noninclusion are shown in figure 2. All patients were included in this intention to treat analysis.

Patients in both groups had similar baseline MAP. Importantly, there were no differences between groups in the numbers of patients undergoing a laparoscopic surgery or having a thoracic epidural analgesia (table 1).

There were no significant differences between groups in anesthesia duration, baseline fluid infusion, or the total volume of fluid boluses received (table 2). However, estimated blood loss and urine output were significantly higher in the computer-assisted group than in the manually adjusted goal-directed therapy group. Consequently, the net fluid balance at the end of the surgery was significantly lower in the computer-assisted group than in the manually adjusted goal-directed therapy group (table 2). The total dose of norepinephrine was more than 40% lower in the computer-assisted group than in the manually adjusted goal-directed therapy group. The computerized closed-loop vasopressor system made more than 1,000 modifications of the infusion rate per case compared to a median of 15 in the manually adjusted goal-directed therapy group (table 2).

The primary outcome, i.e., the percentage of intraoperative case time a patient had hypotension (defined as a MAP less than 90% of their baseline MAP) was 1.2% [0.4 to 2.0] in the computer-assisted group compared to 21.5% [14.5 to 31.8] in the manually adjusted goal-directed therapy group (difference, −21.1 [95% CI, −15.9 to −27.6]; P < 0.001; table 3). The percentage of intraoperative case time with a MAP less than 65 mmHg was also lower in the computer-assisted group than in the manually adjusted goal-directed therapy group (0.0% [0.0 to 0.0] vs. 1.9% [1.2 to 5.0]; P < 0.001). Patients in the computer-assisted group were within the target MAP range (± 10 mmHg of their baseline MAP value) for a greater percentage of time than those in the manually adjusted goal-directed therapy group (97.2% [95.0 to 97.7] vs. 58.8% [48.3 to 70.5]; P < 0.001). The percentage of time with hypertension (defined as a MAP greater than10 mmHg of the MAP target) was also lower in the computer-assisted group than in the manually adjusted goal-directed therapy group (2.5% [1.5 to 4.9] vs. 12.9% [5.7 to 22.8]; P = 0.001; table 4). A heat map of MAP error and stroke volume index by group for all patients is shown in figure 3. The MAP error relative to target MAP in all 38 cases is given in figure A3.

All flow-based variables were better maintained within optimal targets during surgery in the computer-assisted group than in the manually adjusted goal-directed therapy group despite a greater estimated blood loss during surgery in the computer-assisted group. There was no significant difference in baseline SV index between groups, but the average SV index during the last 30 min of the case was significantly higher in the computer-assisted group than in the manually adjusted goal-directed therapy group (P = 0.001). Figure A4 shows SV index in all patients. Importantly, we did not have any missing data among the primary and secondary outcomes.

Lactate concentration at the beginning of the surgery was similar in the two groups, but on arrival in the postanesthesia care unit, patients in the computer-assisted group had a lower lactate concentration than those in the manually adjusted goal-directed therapy group (1.2 [0.9 to 2.3] µmol/L vs. 2.7 [1.6 to 3.2] µmol/L; P = 0.011; table 2). The incidence of postoperative minor complications (secondary outcome) was not different between groups (computer-assisted: 42%; control: 58%; mean difference 16% [95% CI, −16 to 48]; P = 0.330; table 3). Major complications, as well as lengths of stay (postanesthesia care unit, intensive care unit, or hospital), were similar between groups (table 4). No patient died within 30 days postsurgery.

The closed-loop vasopressor system did not experience any technical failures during use, and the anesthesiologists in the computer-assisted group never used backup vasopressor options throughout the surgery. Lastly, the second norepinephrine infusion was never used in the computer-assisted group.

The current study demonstrates that use of computer-assisted hemodynamic management clearly outperforms standard hemodynamic management in intermediate-and high-risk abdominal and orthopedic surgery. Patients in the computer-assisted group had less hypotension, less hypertension, higher mean cardiac index and SV index, and lower lactate concentrations on arrival in the PACU. When using the usual “population” definition of hypotension (MAP less than 65 mmHg), patients in the computer-assisted group had no detectable hypotension. Moreover, these results occurred despite twice the amount of blood loss in the computer-assisted group compared to the manually adjusted goal-directed therapy group. In addition to all the differences noted above, the computer-assisted group also had lower positive net fluid balance at the end of surgery and a higher end-case SV index.

One concern with targeting an individualized MAP during surgery may be that cardiac output (flow) may be sacrificed at the expense of pressure and that hypovolemia may not be corrected if it is hidden by a blood pressure considered as acceptable. However, patients in the computer-assisted group exhibited statistically superior flow-based measures than those in the manually adjusted goal-directed therapy group. The lower postoperative lactate concentrations in the computer-assisted group compared to the manually adjusted goal-directed therapy group suggest that the balance between oxygen transport and oxygen consumption was maintained better in the former group.

Although the incidence of postoperative complications was not significantly different in the two groups, the higher incidence of superficial wound infection in the manually adjusted goal-directed therapy group could reflect less optimal tissue oxygenation in these patients. A larger study is necessary to evaluate the effect of this strategy on postoperative complications.

Individualized hemodynamic management (previously called “goal-directed hemodynamic therapy”) has been associated with improved patient outcomes compared to routine care during major surgery, as recently confirmed by two meta-analysis.^28,29  Futier et al.¹¹  also reported that “individualized” arterial pressure management resulted in less postoperative organ dysfunction. More recently, Nicklas et al.¹²  demonstrated that individualizing hemodynamic management by using patient baseline cardiac index value as a target to guide fluid administration resulted in fewer postoperative complications. However, adoption of these strategies by clinicians has been slow, and even if used during surgery, results may be limited because of poor protocol compliance. Computer-assisted systems may present a bridge between implementing the evidence-based intervention and ensuring compliance, reducing provider workloads while delivering consistent high-compliance therapy that is still directed by the physician. Moreover, the “packaging” of goal-directed strategies into a device reduces the barrier to implementation. Indeed, computer systems are designed for highly repetitive and attention-dependent tasks but thankfully do not suffer with problems associated with vigilance fatigue. Therefore, computational systems are consistently more accurate at maintaining a target set point than anesthesia providers. Moreover, the choice of target set points, and indeed whether a given patient should be placed on a protocol in the first place, are decisions that appropriately remain in the hands of the clinicians. As a result, we believe that automated systems are the best option for automation of noncognitive tasks moving forward.^30,31 

Our study has several limitations that should be considered. First, the principal investigator supervised the computer systems for each patient in the computer-assisted group (in addition to the primary anesthesia provider). This is an impractical setup for widespread adoption, so assurance that the system can be used effectively and safely by a wide range of providers will be necessary for dispersed use. Second, the presence of this additional investigator in the operating room also has the potential to confound the results collected (Hawthorne effect), although the interventions in the “computer-assisted group” are mostly decided by the automated systems in place and will unlikely succumb to any significant observation bias. Similarly, because the primary anesthesia care provider in the manually adjusted group was not involved in the current study, it cannot be ruled out that they were less focused at optimizing hemodynamic status than if they were aware of the study purpose. The specific providers present and level of training (i.e. nurse anesthetists, residents of different years, faculty of different levels of experience) in individual cases was not controlled in the current study. It is possible that some effect caused by group differences in these dimensions may have been overlooked, although the randomization of assignment should have mitigated this. Third, we were unable to record the amount of surgical time each patient was hypotensive during induction because our system required post induction arterial line placement. This is unfortunate, because recording these data would have enabled us to determine whether computer-assisted management functions well with hypotension after induction. Maheshwari et al.³²  demonstrated that approximately 30% of all hypotensive events occur during induction, so management of this period is an important consideration. Additionally, the MAP target chosen to define hypotension (MAP taken during the preoperative screening) could of course be challenged especially in view of the recent literature,³³  but it is the best and the most standardized we have in our institution. Fourth, our protocol was limited to intermediate- to high-risk abdominal and orthopedic surgical procedures and the current findings may not be broadly applicable to other surgeries or clinical settings (e.g., cardiac surgery or intensive care unit). Fifth, the difference in blood loss between the groups is presumed to be random chance; bleeding episodes were surgical in nature and not caused by coagulopathy for example. The design of our study does not rule out the possibility that bleeding might actually be a result of our intervention and not random, however, so this is a question that will also need future study. There is a theoretical risk of increased blood pressure possibly leading to increased bleeding at the surgery site, although the benefits of increased perfusion may outweigh this potential risk. Of note, patients in the manually adjusted goal-directed therapy group spent more time during the procedure with a MAP greater than 10 mmHg above the individualized target. We also cannot rule out an observer effect in the computer-assisted group impacting estimated blood loss estimates, although this difference in estimated blood loss might have been related to the fact that more patients in the computer-assisted group were treated with aspirin.

A key consideration for fully automated hemodynamic management is that multiple factors can influence MAP. An ideal computer-assisted hemodynamic management system should be informed not just of current pressure, but also intravascular volume, anesthetic depth, some measure of sympathetic inhibition, heart rate, and cardiac function. Although our closed-loop system can tightly control MAP, it is not sufficient to rely solely on such a system to ensure adequate hemodynamic management. Looking forward, a holistic hemodynamic management system should be able to monitor and modify all the factors implicated in hemodynamic status. Implementation of such systems in clinical practice will still take time; there are many technological, practical, and regulatory considerations. That said, independently operating automated systems have been reported that manage hypnosis, analgesics, and fluid and vasopressor administration. There are even recent experiments using several systems simultaneously.^34–36  Given all these technological advances in computing power, it is evident that we will see much more of this work in the years to come.³⁷  Specifically regarding fully automated systems for both fluid and vasopressor administration, there is only one team (to our knowledge) that has designed such a system, although it is still experimental.^38,39  Our team is also actively working on the development of such a system.

In patients undergoing intermediate- and high-risk abdominal and orthopedic surgery, use of computer-assisted individualized hemodynamic management significantly reduced intraoperative hypotension compared to the manually adjusted goal-directed therapy group. Computer-assisted systems can help anesthesia providers maintain adequate hemodynamic targets during surgery.

The authors thank all the anesthesiology and the surgical teams of Bicêtre Hospital for their support in this study and the sponsor, which was Assistance Publique–Hôpitaux de Paris (Délégation à la Recherche Clinique et à l’Innovation).

Supported in part by the National Heart, Lung, and Blood Institute of the National Institutes of Health (Bethesda, Maryland) under grant No. U54HL119893 and by National Institutes of Health National Center for Advancing Translational Sciences grant No. UL1 TR00001414 (to Dr. Rinehart). Also supported in part by National Institutes of Health grant No. R01 HL144692 and R01 EB029751 (to Dr. Cannesson).

Dr. Cannesson, Dr. Joosten, and Dr. Rinehart are consultants for Edwards Lifesciences (Irvine, California) and have ownership interest in Perceptive Medical Inc. (Newport Beach, California), which is developing closed-loop physiologic management systems. In addition, Dr. Cannesson and Dr. Rinehart have ownership interest in Sironis (Newport Beach, California), which has developed a fluid closed-loop system that has been licensed to Edwards Lifesciences (Irvine, California) and was used in this study as a decision support system (assisted fluid management). The closed-loop system for vasopressor administration used in this study is new and is the sole creation of three of the authors (Dr. Cannesson, Dr. Rinehart, and Dr. Joosten). A provisional patent has been submitted through the University of California Irvine covering aspects of closed-loop vasopressor administration but does not cover any of the processes discussed in the current article. Dr. Van der Linden reports a financial relationship with Fresenius Kabi GMbH (Bad Homburg, Germany). Dr. Cannesson reports a financial relationship with Masimo (Irvine, California). Dr. Vicaut reports financial relationships with the following: Abbott (Issy-les-Moulineaux, France), Bristol Myers Squibb (Rueil-Malmaison, France), Lilly (Neuilly-sur-Seine, France), Novartis (Rueil-Malmaison, France), Pierre Fabre (Boulogne Billancourt, France), Roche (Boulogne Billancourt, France), Astra Zeneca (Courbevoie, France), Bayer (Pontcarré, France), and Pfizer (Montrouge, France).The other authors declare no competing interests.

Full protocol available at: joosten-alexandre@hotmail.com or alexandre.joosten@aphp.fr. Raw data available at: joosten-alexandre@hotmail.com or alexandre.joosten@aphp.fr.