Closed-loop assisted versus manual goal-directed fluid therapy during high-risk abdominal surgery: a case–control study with propensity matching

This study was approved by the University of California Irvine (UCI) Institutional Review Board (IRB) (HS 2011-8554) and registered with ClinicalTrials.gov (protocol ID NCT02020863). The study was conducted at UCI Medical Center in Orange, CA, USA, from September 2013 through February 2014. Patients eligible for the study were all those over 18 years of age scheduled for hepatobiliary, pancreatic or splenic surgery and expected to receive intravascular arterial blood pressure monitoring as part of their anesthetic care. Exclusion criteria were under 18 years of age, pregnant, body mass index >35 kg/m², presence of moderate to severe valvular disease, cardiac arrhythmias, left ventricular ejection fraction <40% and right ventricular failure. Additionally, any patients who were found to have metastatic disease during a laparoscopic examination and who had their primary procedures aborted as a result were excluded. Patients were recruited on days when a member of the study group was available to consent and the research staff was available to set up the closed-loop system. In general, this was about 30% of the time during the study period.

As a stipulation of the IRB approval, none of the investigators were permitted to be part of the primary anesthesia team for study cases. All patients included in the study were given details of the project verbally and in writing by a member of the study group, and then they provided written informed consent if willing to participate. The results of this study are reported according to the Strengthening the Reporting of Observational Studies in Epidemiology guidelines [22].

Anesthesia was induced with fentanyl (2 μg/kg) and propofol (2 to 3 mg/kg). Following induction, rocuronium was administered to facilitate intubation. Anesthesia was maintained with sevoflurane and fentanyl boluses at the discretion of the anesthesia team. All patients were mechanically ventilated using a volume control mode with tidal volume at 8 ml/kg of ideal body weight and respiratory rate adjusted to achieve an end-tidal CO₂ between 32 and 36 cmH₂O. Any necessary adjustments to anesthetic delivery were made at the discretion of the primary anesthesia providers.

All subjects had invasive arterial pressure monitoring via the radial artery planned as part of their anesthetic care. Standard practice in our institution for moderate- to high-risk hepatobiliary surgery patients is to use flow-based monitoring that includes a dynamic predictor of fluid responsiveness such as SVV to provide GDFT. For these cases (both study and control group), all patients were monitored with one of three dedicated EV-1000 monitors and a FloTrac sensor (Edwards Lifesciences, Irvine, CA, USA).

Finally, all patients except those undergoing liver resection had a thoracic epidural placed by the pain service in the preoperative holding area prior to coming to the operating room. The epidural was not activated (save for a 3-ml test dose of 2% lidocaine with 1:200,000 epinephrine) until the surgical case was closing and emergence of the patient was beginning (and data collection completed).

The closed-loop software (Sironis, Newport Beach, CA, USA) was run on a Shuttle X50 Touchscreen PC (Shuttle Computer Group, City of Industry, CA, USA) running Windows 7 (Microsoft, Redmond, WA, USA). The system was connected via a USB to serial adapter to the serial output port of the EV-1000 for real-time capture of data.

A Q Core Sapphire Multi-Therapy Infusion Pump (Q Core Medical, Netanya, Israel) was used by the closed loop to deliver fluid. The Sapphire pump is a single-channel volumetric pump capable of flow rates from 0.1 to 999 ml per hour. The pump was controlled by the closed-loop system using software provided by Q Core via serial connection (Commands Server R.00).

Following placement of the arterial line (and central line if applicable), but before surgical start time, the closed-loop system target was selected by the primary anesthesia team (scaled stroke volume (SV) increase of 7.5% to 22.5% over 500 ml, standard setting 15%) and started. The primary team was responsible for selection of therapy targets throughout the case (see below in Closed-Loop System Description).

The closed-loop system has been described extensively in previous publications [17-21]. Briefly, the system monitors SV, heart rate (HR), mean arterial pressure (MAP) and a dynamic predictor of fluid responsiveness such as pulse pressure variation or SVV and uses this information to optimize SV in line with a GDFT protocol. The controller uses both a model layer to formulate a predicted response to a fluid bolus and an adaptive layer for bolus-based error correction during direct fluid management to correct for changes induced by surgical and anesthetic conditions. The final action to be taken by the controller is then determined by a rule-based layer based on data provided by the previous layers [19]. The system is ultimately a slope-seeking controller whose aim is to optimize patients’ fluid status and SV to near the plateau of the Frank–Starling curve. The system delivers 100-ml fluid boluses over the course of 6 minutes when indicated (maximum rate =1,000 ml/hour) and is therefore designed not for high-volume resuscitation, but rather for hemodynamic optimization in line with GDFT protocols.

The exact point on the Frank–Starling curve to be targeted by the controller can be adjusted by the primary anesthesiologist by adjusting the target SV increase, allowing providers to make fluid therapy more liberal or more conservative as appropriate while still following a GDFT protocol. Additionally, the primary anesthesia team can put the system on standby (monitor-only mode), deliver a fluid challenge manually or deliver continuous fluid in the case of an emergency.

The primary anesthesiologist was aware of each bolus of fluid administered by the closed-loop system throughout the surgical procedure. This was done through visual and audio signals alerting the clinician that a fluid bolus was initiated. Additionally, the closed-loop system requires the primary care provider to input the amount of fluid available for delivery, and this volume is limited to 500 ml at a time. This feature was specifically designed as another safety check, as it requires the supervising anesthesiologist to “reauthorize” the system to deliver fluid in 500-ml increments, ensuring the system cannot deliver more than this volume at any time without the supervisor being aware of it. Lower volumes could be chosen at the supervisor’s discretion (200 ml or even 100 ml).

In the standard GDFT patients treated at UCI Medical Center, a 3 ml/kg/hr baseline crystalloid infusion is run, and providers follow the protocol shown in Figure 1 by hand [23]. For the closed-loop cases, patients received a restrictive baseline isotonic crystalloid infusion of 3 ml/kg/hr on a pump throughout the duration of the case as maintenance fluid. In addition to the baseline crystalloid, the closed-loop system administered colloids chosen by the primary anesthesia team. Available fluids were 6% hydroxyethyl starch 130/0.4 (Voluven; Fresenius Kabi, Frankfurt, Germany) or 5% serum albumin (Grifols, Sant Cugat del Vallès, Spain). The anesthesia provider had the option of stopping a bolus (or initiating one) if he or she disagreed with the closed-loop fluid management system, and these events were recorded by the system. If the primary anesthesia team felt the patient was fluid-optimized but the patient’s systolic blood pressure was still 20% or more below baseline value (despite appropriate anesthetic depth), they were to consider an inotrope or vasopressor per the lower portion of the manual protocol (Figure 1).

Following completion of enrollment, each study patient was matched to a non–closed-loop assisted patient treated during the same time period using a propensity score match to reduce bias [24,25] (Figure 2). The propensity score was estimated using logistic regression to regress group assignment on the predictors in Table 1 (age, height, weight, sex, HR, Stroke Volume Index (SVI), MAP, SVV, specific scheduled procedure and surgeon). Each study patient was then matched 1:1 to a control case using calipers of 0.04% width (20% of the standard deviation (SD) of the propensity score), creating a control manual GDFT (M) group to compare with the closed-loop assisted (CL) group. The complete propensity match process, data table and matched pairs are detailed in Additional file 1.

The following data were collected continuously at 60-second intervals for each patient by the anesthesia information management system (AIMS) (Surgical Information Systems, Alpharetta, GA, USA): HR, MAP, SV, CO and SVV. Fluid volume, urine output, blood loss, and vasopressors were also recorded in the AIMS by the primary anesthesia team. In study cases, the closed-loop controller also recorded continuous hemodynamic data (HR, MAP, SV and SVV) provided by the EV-1000 at 2-second intervals, as well as fluid administration and provider interactions (set point changes, bolus stops and test boluses administered).

Baseline values for HR and MAP were taken from the preoperative anesthesia assessment. Baseline values for SVI and Cardiac Index (CI) were taken as the mean of the first minute’s recording 5 minutes after placement of the arterial line and calibration of the hemodynamic monitoring system. Mean case value for each hemodynamic variable was calculated as the average value of all recordings between baseline and the beginning of surgical closure. Final case value for each hemodynamic variable was defined as the average value of each parameter over the last minute before the beginning of emergence (lowering of anesthetic agent in preparation for extubation). Finally, a Postoperative Morbidity Survey (POMS) score [26] was calculated for each patient, and length of Surgical Intensive Care Unit (SICU) and hospital stay were recorded.

The primary outcome between groups was GDFT compliance, defined as percentage of case time where SVV was ≤12%. Upon study initiation in September 2012, the original endpoint was cardiac SV. When our case series in Europe was finished, informal comparison with other standard case data showed strong effects on SVV for the CL group, but weaker effects on SV itself [21]. Moreover, we observed that SVI as recorded by the minimally invasive CO monitoring systems has a very wide distribution between and within patients [27,28], meaning a large sample size would be needed to detect true effects on this variable. Second, Pearse and colleagues reported the full “Optimisation of peri-operative cardiovascular management to improve surgical outcome” (OPTIMISE; ISRCTN04386758) study in 2014 [5]. One of the findings of that study was the importance of protocol compliance to the beneficial effects of the GDT:

In the prespecified adherence-adjusted analysis conducted using established methods [citation omitted], the observed treatment effect was strengthened when the 65 patients whose care was nonadherent [internal cross-reference omitted] were assumed to experience the same outcome as if they had been allocated to the alternative group (RR, 0.80; 95% CI, 0.61-0.99; P = .04). (p 2186 [5])

On the basis of the pilot study data, the findings derived from the OPTIMISE trial and the fact that it would be the same endpoint as the pilot study (GDFT protocol compliance), we decided to treat protocol compliance measured as SVV as the primary outcome measure in the present study.

Secondary endpoints were case mean hemodynamic values, final case hemodynamic values, fluid volumes, urine output, postoperative complications, SICU length of stay and hospital length of stay. Additional measures of controller performance recorded during the study were distribution of fluid boluses given across cases (minimum, maximum, mean and SD), percentage of the boluses that resulted in an SV increase and predictive specificity as number of fluid boluses given to a patient increased. Provider interactions with the system were also recorded (start and stop of boluses and changes in controller set point).

Based on local data, case time with SVV ≤12 in the case types to be studied was approximately 80 ± 10% over the previous 1-year period, or about 20 ± 10% non-compliance time. Cutting non-compliance time in half (or more) was considered a clinically relevant endpoint that, assuming a power of 0.8 and a significance level of 0.05, would require 21 patients in each group to be adequately powered.

Continuous scalar data were tested for normality using a Kolmogorov–Smirnov test. If normal, the data are reported as mean ± SD and comparisons were made using Student’s t-test. Non-normally distributed scalar data and ordinal data (namely, POMS scores) are reported as median (25th, 75th percentile), and comparisons were made with the Mann–Whitney U test. The significance level for the primary outcome was set at 0.05. All other comparisons were made at the 0.025 level to compensate for the multiple secondary endpoints.