Introduction

Discrepancies between demand (the number of patients requiring mechanical ventilation) and supply (the number of skilled clinicians available to manage these patients) are anticipated to become more frequent due to an aging population and to increasing severity of illness in patients [1, 2]. Costs related to ICU patients represent a major burden for healthcare systems [3] and they are largely driven by costs pertaining to mechanically ventilated patients [4, 5]. In addition, transforming research knowledge into clinical practice has been difficult in the field of mechanical ventilation, specifically in regard to the application of weaning protocols [6] and to tidal volume (TV) reduction [7, 8]. Automated systems of mechanical ventilation provide a potential solution in these problems.

A fully automated mode of ventilation has been recently developed (Intellivent) [9]. This new mode stems from adaptive support ventilation, a pressure-controlled and pressure-assisted mode that continuously adjusts the inspiratory pressure level and inspiratory time based on the least work of breathing concept of Otis et al. [10, 11]. In addition, Intellivent incorporates new features allowing fully automated mechanical ventilation, including a closed loop for ventilation to adjust minute ventilation based on end-tidal CO2 (EtCO2) targets and a closed loop for oxygenation to adjust positive end-expiratory pressure (PEEP) and FiO2 based on SpO2 targets.

The present randomized controlled trial compared this fully automated mode of ventilation and a protocol-based ventilation strategy in patients after cardiac surgery [12]. The postoperative course after cardiac surgery is dynamic. Within 2–4 h, the patient’s temperature increases, resulting in increased CO2 production and the need for adjustment of minute ventilation. Patients wake up after cessation of sedation and are switched from mandatory breath ventilation to spontaneous breath ventilation. Finally, PEEP and FiO2 levels have to be weaned and adjusted to maintain adequate oxygenation. These physiological changes represent an ideal situation to test a fully automated system of mechanical ventilation. The aim of this study was to ascertain if automated adjustments of mechanical ventilator settings would result in safe mechanical ventilation in patients following cardiac surgery.

Materials and methods

We conducted a prospective randomized controlled study comparing automated ventilation (AV) and protocolized ventilation (PV) in patients after cardiac surgery at the Institut Universitaire de Cardiologie et de Pneumologie de Québec. Study approval was obtained from the local ethics committee and signed informed consent was obtained from all patients before surgery.

Patients

Preinclusion criteria were the following: (1) elective cardiac surgery, (2) age between 18 and 90 years, (3) body mass index <40 kg/m2, (4) baseline PaCO2 <50 mmHg, and (5) serum creatinine <200 μmol/l. The patients were included after surgery if by 15 min after their arrival in the ICU they were hemodynamically stable and had a urine output >50 ml/h. Patients were excluded if a bronchopleural fistula was present or if the study ventilator was not available. Figure 1 provides a flow chart of the study.

Fig. 1
figure 1

Study flow chart

Full size image

Study protocol

Patients were either randomized to AV (AV group) and were connected to a modified G5 fully automatic ventilator (Intellivent; Hamilton Medical, Rhazuns, Switzerland) or to protocolized ventilation (PV group) with a G5 ventilator. The concealed randomization was performed by sealed and opaque envelopes. Randomization with 1:1 allocation and a block randomization scheme was obtained from http://www.random.org.

Automated ventilation group

The Intellivent system adjusts TV, respiratory rate (RR), FiO2 and PEEP based on the patient’s respiratory mechanics, EtCO2 and SpO2. The basic principles of this ventilator have been recently described by Arnal et al. [13] and are presented in detail in the Electronic supplementary material. PEEP level was limited to 10 cmH2O and the only manual setting after inclusion in the AV group was the patient’s height and sex to determine initial minute ventilation.

Protocolized ventilation group

PV was administered according to the local written protocol which was based on recommendations in anesthesiology textbooks [14, 15]. On ICU arrival, patients were connected to the ventilator and the initial settings were prescribed by the treating anesthesiologist. TV was set at 10 ml/kg, RR at 10 breaths per minute, PEEP at 5 cmH2O, and FiO2 ranged from 70 to 100 %. The ventilatory protocol for the postoperative period was managed by respiratory therapists for FiO2 weaning (decreased by 10 % every 10 min if SpO2 ≥95 % to reach 40 %) and for the switch to pressure support ventilation (PSV) as soon as patients were deemed to be able to breath spontaneously. PSV was used during the weaning phase. Modifications of RR and TV were managed by intensive care physicians based on arterial blood gas results collected at inclusion and after any adjustments in ventilator settings.

Features common to both groups

Patients were managed using fast-track extubation procedures [16]. Patients were extubated if they had stable respiratory parameters with a PSV level of 10 to 12 cmH2O, PEEP ≤5 cmH2O and FiO2 ≤40 %, and adequate neurological and hemodynamic parameters.

Data recorded

A research assistant was at each patient’s bedside during the entire duration of the study (4 h) to assess patient safety, to record the times that the patient passed in the different predefined zones of ventilation (Table 1) and to record the number and duration of interventions required to set the ventilator. All ventilator data were stored in the ventilators for both groups. In the PV group the plateau pressure (Pplateau, end-inspiratory pause of 0.3 s) was recorded continuously and in the AV group the peak pressure was assumed to be equivalent to Pplateau (pressure mode).

Table 1 Predefined zones of ventilation
Full size table

Endpoints

The study was designed to assess feasibility and safety. The primary endpoint was the number of episodes and the time in the predefined “not acceptable” zone of ventilation defined as a TV greater than 12 ml/kg of predicted body weight (PBW), Pplateau above 35 cmH2O, EtCO2 below 25 or above 51 mmHg and a SpO2 below 85 % for a minimum of 30 s (Table 1). These targets were based on the concordance between the above criteria and a multicenter survey involving 53 physicians and respiratory therapists [17]. The secondary endpoints were time spent by patients with safe ventilation (predefined zones of optimal and acceptable ventilation; Table 1), the number of manipulations required to set MV, FiO2 and PEEP, and the time needed for these manipulations.

Statistical analysis

This was a prospective study and no information was available to estimate the sample size. We analyzed results from 30 subjects per group to get information about the primary end-point. Based on previous data on high TVs in this population [18], our sample size was sufficient to detect a reduction for this parameter from 20 to 1 % [19, 20], with a power of 80 % and alpha risk of 5 %. Only 50 patients would be required to demonstrate a difference in a time in the “not acceptable” zone of ventilation of 15 min and a time of 1 min with a standard deviation of 20 min with a power of 90 % and alpha risk of 5 %.

Continuous variables are expressed as means (SD) or median (min–max), depending on variable distribution. Group comparisons were analyzed using Student’s t test or Wilcoxon’s rank-sum test for continuous variables. Categorical variables are expressed in percentage and were analyzed using the chi-squared or Fisher’s exact tests. A mixed model was used to analyze TV, Pplateau, PaCO2 and PaO2 with one experimental fixed factor associated with the comparison between PV and AV at inclusion and hourly for four consecutive hours. This level was analyzed as a repeated measure. A heterogeneous autoregressive structure was used to measure the dependence among repeated measurements. The multivariate normality assumptions were verified with Shapiro–Wilk tests after Cholesky factorization. The results were declared significant with P values <0.05. The data were analyzed using the statistical package program SAS v9.2 (SAS Institute, Cary, NC).

Results

Patient characteristics at inclusion

From among the 94 patients assessed for eligibility, 60 were randomized, 30 to the AV group and 30 to the PV group. All 60 patients completed the study, except one in the AV group who needed reoperation for massive bleeding. The two groups were comparable at inclusion in terms of demographic data, severity scores, and type and duration of surgery (Table 2), as well as for physiological data at study entry (Table 3). At inclusion, all patients were hypothermic with a mean temperature of 35.6 ± 0.7 °C, most had stable hemodynamic parameters (11/60, 18.3 %, had a cardiac index below 2 l/min/m2) and were deeply sedated at study inclusion with a Richmond assessment sedation scale score of −5. They were ventilated with a mean TV of 10.4 ± 1.4 ml/kg of PBW and RR of 10.5 ± 1.2 breaths/min with no differences between groups. At ICU arrival, 60 % of patients were ventilated with a TV above 10 ml/kg, as previously reported [18]. FiO2 at ICU arrival was set at 69 ± 12 % and PEEP at 5.1 ± 1.4 cmH2O (Table 3).

Table 2 Patient characteristics at inclusion in the PV group and the AV group
Full size table
Table 3 Physiological data at inclusion and at 1 h in the PV group and the AV group
Full size table

Primary outcome

In the PV group, 13/30 patients (43 %) had at least one episode in the “not acceptable” zone of ventilation (1.4 ± 2.1 times per patient), compared to 4/30 patients (13 %) in the AV group (0.4 ± 1.0 times per patient; P = 0.01). Patients were within the “not acceptable” zone of ventilation for 15 ± 38 min in the PV group and for 1 ± 4 min in the AV group (P = 0.048; Fig. 2). These times correspond to 7.3 and 0.5 % of the total duration of the mechanical ventilation, respectively.

Fig. 2
figure 2

Time within the different zones of ventilation in the AV group and the PV group. The time in the “not acceptable” zone of ventilation was reduced by AV, when expressed as percentage of time, total duration or number of episodes per patient

Full size image

In the AV group, “not acceptable” ventilation was related to a TV greater than 12 ml/kg PBW in one patient for 207 s, a Pplateau >35 cmH2O in two patients for a mean of 306 s, a SpO2 below 85 % in two patients for a mean of 99 s, an EtCO2 <25 mmHg in one patient for 30 s, and an EtCO2 >51 mmHg in three patients for a mean of 445 s.

Secondary outcomes

Patients were in the “optimal” and “acceptable zones” of ventilation for 25 ± 124 and 174 ± 133 min in the PV group compared to 192 ± 52 and 21 ± 31 min in the AV group (P < 0.001 for both comparisons; Fig. 2). The number of patients requiring manual changes of ventilator settings during the 4-h study period was 30/30 (100 %) in the PV group as compared to 4/30 (13 %) in the AV group (P < 0.001), corresponding to 148 and 5 interventions, respectively (P < 0.001). Over the study period safe ranges of TV, Pplateau, PaCO2 and PaO2 were maintained with AV (Fig. 3). The TV/PBW was automatically reduced in the AV group (Table 3). The TV/PBW distribution demonstrates that in the AV group, no patients received a TV above 10 ml/kg PBW after inclusion (Fig. E1).

Fig. 3
figure 3

Evolution of TV, Pplateau, PaCO2 and PaO2 during the study from baseline (H0) to the fourth hour of the study (study end, H4) in the AV group (black squares) and the PV group (white circles, dotted line)

Full size image

At inclusion, mean initial PaCO2 was 40 ± 5 mmHg. After the first hour, PaCO2 was lower in the PV group resulting in a significantly higher pH (Table 3). Hyperoxia was less frequent in the AV group (Table 3; Fig. 3). There were no hemodynamic differences between the study groups during the 4-h study period in terms of cardiac index, heart rate and systolic blood pressure.

The median duration of mechanical ventilation was 6.6 h (interquartile range 4.8–8.7 h) in the PV group and 5.2 h (4.4–7.4 h) in the AV group (P = 0.29). In both groups more than 90 % of patients were successfully liberated from the ventilator and extubated on the day of surgery. All patients were ultimately extubated without reintubation or noninvasive ventilator support and all were discharged alive from the ICU within 48 h.

Discussion

This is the first randomized controlled study of fully automated mechanical ventilation using the Intellivent system in a population of adult cardiac surgery patients. We found that AV with this system was safe in hemodynamically stable patients following cardiac surgery and required fewer mechanical ventilator interventions compared to a PV strategy. TVs below 10 ml/kg PBW were provided automatically with the automated system. Oxygenation parameters were also automatically managed, allowing automated weaning of FiO2 and avoiding hyperoxia. These results were obtained with a reduction in the number of manual interventions as compared to PV managed by respiratory therapists, anesthesiologists and critical care physicians.

The new mode of ventilation evaluated in the present study stems from adaptive support ventilation that was previously evaluated in patients after cardiac surgery [21]. However, this system is more advanced because it includes closed-loop algorithms that titrate both oxygenation and ventilation, and is therefore a fully automated mechanical ventilation system [22, 23]. Intellivent is now considered the most advanced form of closed-loop ventilation with automated delivery of mandatory and spontaneous breaths with automated settings for inspiratory and expiratory pressure, inspiratory time and FiO2 and automated weaning [24]. This new generation of closed-loop ventilators will require sufficient clinical evaluations before clinicians gain confidence and use them routinely.

Automated weaning (SmartCare) is the closed-loop system that has been most evaluated to date. This system was at least equivalent in weaning patients compared to usual care or protocol-directed weaning [25–28]. More recently, Intellivent has been developed, and only one pilot evaluation is available [13]. Arnal et al. [13] compared Intellivent to adaptive support ventilation in a cross-over short-term study in a mixed medical population including patients with acute respiratory distress syndrome (ARDS). This study demonstrated that the system could be used safely for short periods even in critically ill patients. In the present study, we demonstrated that the system is safe and has the potential to reduce the clinician workload after cardiac surgery. All these promising evaluations were conducted with first-generation of closed-loop ventilators, and improvements in the initial algorithms are ongoing.

Several arguments favor the development of automated closed-loop ventilators. First, the healthcare system is expected to be challenged by increasing numbers of mechanically ventilated patients which will not be accompanied by an equivalent increase in clinicians [1, 4], and data suggest that the healthcare system is already stressed [29, 30]. Second, the costs pertaining to mechanical ventilation and ICU stays are increasing, and interventions that facilitate a decrease in weaning time and ICU length of stay are valuable [5]. Third, knowledge transfer of results from research into clinical practice is notoriously difficult [6–8]. Computerized knowledge-based systems may help implement and homogenize clinical practice. Initial evaluations of ventilators using advanced closed loops demonstrate that for specific and repetitive tasks, the systems may outperform clinicians. It has been previously shown that adaptive support ventilation could reduce the workload in comparison with synchronized intermittent mandatory ventilation (volume-controlled intermittent mandatory ventilation) after cardiac surgery [21]. In the present study, the number of ventilator manipulations after inclusion was reduced by a factor of almost 30, demonstrating the potential for workload reduction with such a system. However, limitations of these systems exist and must be appreciated by clinicians to avoid complications [9].

Recent data demonstrate deleterious effects of high TVs even in patients without acute lung injury (ALI) or ARDS [31, 32]. In patients following cardiac surgery, several physiological studies have demonstrated that lower TVs decrease the inflammatory state after cardiopulmonary bypass [33, 34]. A recent study based on 3,434 consecutive patients after cardiac surgery showed that TVs above 10 ml/kg of PBW are an independent risk factor for organ dysfunction and prolonged ICU stay [18]. However, the application of a TV reduction strategy in clinical practice faces several barriers [7, 8], and a recent longitudinal survey has shown that a significant number of patients are still not ventilated with a low TV strategy [35]. In the present study, TV was reduced automatically and all patients received TVs below 10 ml/kg of PBW after inclusion in the AV group. Of note, in patients without ALI or ARDS at initiation of mechanical ventilation, TVs below 10 ml/kg have been recommended by several authors [31, 32, 36].

This study had several limitations. By its nature, the study could not be blinded, but the associated potential biases were limited by the a priori defined zones of ventilation. The patients included in the study were hemodynamically stable and recovering from cardiac surgery, and therefore conclusions cannot be generalized to all patients. Evaluation of more challenging patients such as those suffering from ARDS or sepsis, or those with hemodynamic instability, is required. Owing to the study design, the duration of this study was limited to 4 h and could not fully evaluate outcome data such as duration of mechanical ventilation. The duration of intubation was reduced from 6.6 h (4.8–8.7 h) to 5.2 h (4.4–7.4 h) with AV (P = 0.29). This difference was not statistically significant, but the sample size calculation was not based on this parameter. Clinicians must understand the limitations of automated systems. Closed loops in these systems rely on the availability and quality of the EtCO2 and SpO2 signals. If the signals are inadequate, an automated switch to manual ventilation is performed. More importantly, EtCO2 can be affected by sudden changes in cardiac index and a drop, due to a decrease in cardiac index, may result in an automatic decrease in minute ventilation. In the present study, most patients were hemodynamically stable, but 18 % had a cardiac index below 2 l/min/m2. None of these patients had an inadequate minute ventilation, but this theoretical issue must be kept in mind in very unstable patients. Finally, in the PV group in this study our local postoperative weaning protocol that follows recommendations in anesthesiology textbooks was used [14, 15] and was conducted by well-trained respiratory therapists, and consequently the results may not be generalizable to other centers. The TV used in the PV group is in line with that used in an observational study involving 3,434 patients in the same center [18], and with other published data in cardiac surgery patients [37] and current textbook recommendations [14, 15]. Nonetheless, the finding of high TVs delivered during PV, which in the present study was an index of non-optimal ventilation, may have been due to incorrect manual ventilator settings. It is likely that lower TVs will be used more frequently in patients without ALI or ARDS [31], and this may lead to a reduction in the differences between AV and manual settings.

Conclusion

We demonstrated that a fully automated system of ventilation can be safely used in stable patients following cardiac surgery. AV optimized ventilator settings while reducing the number of manual interventions. Advanced closed-loop systems are anticipated to assume a larger role in the future owing to the growing number of patients requiring mechanical ventilation with an anticipated persistence of limited physician resources. Preliminary evaluations of such automated systems have provided encouraging results and further improvements are expected in the next generation of closed-loop technology. For repetitive and relatively simple tasks, computerized systems may be safer and more efficient in knowledge transfer than the individual physician even in a well-staffed unit. However, we should be aware of the limitations of these devices and remember that technology can help with, but not replace, clinical judgment, which will always be required to manage complex patients in the ICU environment.