Automatic protective ventilation using the ARDSNet protocol with the additional monitoring of electrical impedance tomography

The system consists of a panel PC (PPC-154 T; Advantech Co., Ltd, Taipei, Taiwan), a mechanical ventilator (SERVO 300; Marquet Critical Care AB, Solna, Sweden), and other equipment including a capnography device with pulse oximetry (CO₂SMO+; Philips Respironics, Best, The Netherlands), a spectrophotometry device (CeVOX; Pulsion Medical Systems SE, Feldkirchen, Germany) to measure arterial oxygen saturation (SaO₂), a patient monitor (Sirecust 960; Siemens AG, Munich, Germany), and an EIT device (GOE-MF II; Dräger AG, Lübeck, Germany). All measured signals are transmitted directly to the panel PC including parameters from the mechanical ventilator, by which airway pressure and airway flow are converted by a 12-bit analog-to-digital converter (KPCMCIA-12AI-C; Keithley Instruments Inc., Cleveland, OH, USA). The computed commands of ventilatory variables are transmitted from the panel PC to the mechanical ventilator by a 12-bit digital-to-analog converter (PCMDA12B; SuperLogics Inc., Waltham, MA, USA). Automatic adjustment of ventilatory settings can be made by this setup.

The protocols from the different commercial devices are graphically programmed using Labview software (version 7.1; National Instruments Inc., USA). The specific binary codes for each device are transmitted from the panel PC to the devices. The interfaces are based on RS-232 standard. To obtain the up-to-date measured parameters, the requested commands must be repeatedly sent to all devices in every sampling period of 100 milliseconds. Once the panel PC receives a response from a device, the data are decoded and saved on a regular basis within the sampling time.

After approval from the Department of Health and Social Services Berlin (reference number IC 113-G0151/10), all animal procedures were conducted complying with national regulations and institutional animal care committee guidelines. Seven female domestic pigs (29 ± 3 kg) received premedication and general anesthesia with thiopental, fentanyl and pancuronium, and were then tracheotomized in a supine position. A spectrophotometry catheter (CeVOX; Pulsion Medical Systems AG) was inserted into the carotid artery for measuring SaO₂. Noninvasive measurements of peripheral oxygen saturation (SpO₂) from capnography device and from the patient monitor were placed at the left ear and at the tail, respectively. In addition, a central venous line and a pulmonary artery catheter were placed into the internal jugular vein for continuous monitoring of pulmonary artery pressure and central venous pressure and for drug and crystalloid fluid infusion. Subsequently, the pigs underwent surfactant depletion with repetitive lavages by warm (37 to 38°C) saline solution (0.9% NaCl, 40 ml/kg body weight) to induce ARDS (PaO₂/FiO₂ < 200 mmHg) at FiO₂ of 1.0 [19]. The lavages were carried out between two and four times (average three times) within 5 minutes. The pigs were then ventilated in volume-controlled mode with V_T of 6 ml/kg body weight and static PEEP of 5 cmH₂O. After 30 minutes, the closed-loop ventilation was started. However, after 2 hours of ventilation, a disconnection from the ventilator was performed for 10 seconds, simulating involuntary patient–ventilator disconnection during patient handling.

To some extent, our algorithm of the ARDSNet protocol is based on earlier work of our group [20, 21]. The protocol is a ventilation strategy using low tidal volume at 6 ml/kg of the PBW based on the formulae given in Equations (1) and (2) for male and female subjects [17]:

In the present study, PBW was replaced with the actual measured body weight for the ventilation setting on V_T. The following goals should be fulfilled.

The protocol can be effectively developed by goal-oriented structural programming. The overall complexity of the protocol is simplified by a task-based programming structure, presented in Figure 1. This structure increases efficiency in coding the program.

Once the automatic ventilation is started, the initialization activates all initial settings of ventilation variables, such as PEEP, FiO₂, V_T, RR, and I:E ratio. Regarding the oxygenation goal, PEEP of 14 cmH₂O and FiO₂ of 0.7 from the middle of the Table 1 are initially selected. These settings should be given for a number of breaths and this is represented by a 30-second waiting block or the ventilation task in Figure 1. Thereafter, SaO₂ is evaluated for further adjustment of the PEEP and FiO₂ combination.

During the experiments, a number of choices for oxygenation goal can be made either by invasive measurement of SaO₂ from the spectrophotometry device (CeVOX; Pulsion Medical Systems SE), or by noninvasive measurement of SpO₂ from the capnography device (CO₂SMO+; Philips Respironics), or from the patient monitor (Sirecust 960; Siemens AG). In our experiments, SaO₂ measured from CeVOX was chosen for control of oxygenation. If SaO₂ falls below 88%, a higher combination of PEEP and FiO₂ should be given. In contrast, if SaO₂ rises above 95%, a lower combination of PEEP and FiO₂ should be applied to minimize hemodynamic effects of PEEP and to reduce the risk of oxygen toxicity.

After achieving the oxygenation goal, the next step is to check for the pH goal. If there is no new pH value, a further evaluation of P_plat should be made. However, if a new pH value is manually given, proper adjustment of RR and V_T should be performed for the pH goal. After fulfilling the pH goal, the plateau pressure goal is carried out every 10 minutes or after a change in PEEP or a change in V_T. All new variables are applied to the subject for a number of breaths. This repeated process is continuously performed. Using this goal-oriented structure, all goals of the ARDSNet protocol will be accomplished.

Seven cases of porcine dynamics were studied with the protocol. As an example of the system performance, we present 4 hours of ventilation from one of the pigs (27 kg). During the lavage, the ventilation settings were set by manual operation and the automatic mode was subsequently turned on for ventilation therapy after PaO₂/FiO₂ < 200 mmHg for 15 minutes. The oxygenation was kept within the range between 88 and 95%, hence satisfying the oxygenation criterion. Using Table 1, the knowledge-based controller was able to stabilize and regulate the SaO₂ value.

Figure 2 shows the response of lung lavage in the first 30 minutes and the automatic ventilation for stabilization and regulation of SaO₂ by adjusting PEEP and FiO₂ referred to in Table 1. At 2.5 hours, or 2 hours after automatic ventilation, a disconnection of ventilation was made for 10 seconds, simulating a clinical scenario of airway suction or accidental disconnection. The controller was able to recover the critical situation of low oxygenation by step-by-step change for the values of PEEP and FiO₂, until PEEP of 24 cmH₂O and FiO₂ of 1.0. Subsequently, an automatic titration of suitable PEEP and FiO₂ was carried out again to fulfill the oxygenation goal.

To minimize ventilator-induced lung injury, the P_plat goal should be kept below 30 cmH₂O by the adjustment of V_T[13]. During the time 0.5 to 2.5 hours shown in Figure 3, P_plat was definitely less than 25 cmH₂O while V_TPW was maintained at 6 ml/kg.

During the time between 2.5 and 3.25 hours (hypoxia as presented in Figure 2 due to disconnection at the patient–ventilator interface at 2.5 hours), P_plat was >30 cmH₂O and V_TPW was reset from 6 to 5 ml/kg and from 5 to 4 ml/kg, respectively, while the PEEP and FiO₂ combination was increased to possible maximum values. At 3.25 hours, when P_plat < 25 cmH₂O and V_TPW < 6 ml/kg, V_TPW was automatically increased stepwise by 1 ml/kg increments to 5 ml/kg. Using this approach, the goal of P_plat is satisfied with the main objective to minimize P_plat ≤ 30 cmH₂O.

Figure 4 shows the result of pH control for the 27 kg pig. The pH values were regularly measured every 30 minutes. At 0.5, 1 and 1.5 hours, the pH value was <7.30, and RR was increased by 5 bpm after entering the pH value into the system. At 2 and 2.5 hours, the pH goal was satisfied and RR remained unchanged. At 2.5 hours, ABG was measured before the 10-second disconnection at the patient–ventilator interface. At the next ABG (3 hours), the pH value falls below 7.15. RR was immediately set at 35 bpm to treat severe acidosis due to disconnection at the patient–ventilator interface. With the maximum limit of RR at 35 bpm, this resulted in an increase of the pH value to >7.15. At 3.5 and 4 hours, the pH value was 7.15 to 7.30 and RR should be increased. However, RR was already at its limit set at 35 bpm.

Figure 5 shows arterial carbon dioxide tension (PaCO₂) from ABG and end-tidal carbon dioxide (etCO₂) during the 4 hours of ventilation. PaCO₂ significantly increased after lung lavage, which indicates poor gas exchange or partial lung collapse (atelectasis). This is also confirmed by the SaO₂ curve and the EIT images in the next subsection. After turning on the protocol for 2 hours during 0.5 to 2.5 hours, gas exchange was gradually improved due to the ventilation therapy. Again, at 2.5 hours, poor gas exchange recurred during hypoxia because of disconnection at the patient–ventilator interface for 10 seconds. Since PaCO₂ was taken before disconnection at the patient–ventilator interface, severe hypercapnia was later detected (at 3 hours). However, the automatic ventilation improved gas exchange and severe hypercapnia was relieved in a timely manner.

By monitoring carbon dioxide parameters, physiological dead space can be estimated by Bohr–Enghoff’s equation [22]:

where V_D denotes physiological dead space. Based on the dataset of 4-hour ventilation, the average fraction of physiological dead space for this pig was 0.39. In the other words, approximately 60% of the tidal volume took part in the gas exchange.

EIT allows non-invasive monitoring of electrical impedance within the thoracic cavity in a two-dimensional and cross-sectional plane in order to assess regional ventilation [23]. Pathophysiological changes of the lung can be observed from the EIT images in real time at the bedside. Sixteen electrodes were used for the voltage measurement and the backprojection algorithm [24] was implemented for image reconstruction. A 32 pixel × 32 pixel EIT image is captured at the end of inspiration, as shown in Figure 6. Based on the attachment of the EIT belt in the predefined arrangement of the electrodes shown in Figure 6 (left image), ventral and dorsal parts of the animal are situated at the top (electrode position 1) and at the bottom (electrode position 9) of the EIT image, respectively. The position of the left and right lungs can therefore be determine in the specified position as shown and similar to the standard interpretation, obtained from a computed tomography scan image.All seven female pigs (weighing 29 ± 3 kg) were ventilated using the autoARDSNet protocol; the results of their EIT images are summarized in Figure 7. These results show the EIT images before and after lavage, and after 2 hours and 4 hours of ventilation using the protocol. The area of high electrical impedance corresponds to the movement of air, which is designated by tones of orange and yellow. After lavage, a loss of lung volume and poor dorsal ventilation can be observed by the images in all cases. After 2 hours and 4 hours of ventilation, a progressive improvement of dorsal ventilation can be seen compared with the EIT images after lavage.

Regarding pig #7, pneumothorax was observed by EIT image after 2 hours of ventilation demonstrating that only the left lung was ventilated. A corrective action was made at 2.75 hours to release excess pressure at the right lung, which improved lung compliance, oxygenation, hemodynamics and carbon dioxide exchange. Based on this experience, we believe that the EIT device is useful for practical decision-making at the bedside.

Regional analysis of ventilation [25] was carried out for six pigs (excluding pig #7 due to the pneumothorax), as shown in Figure 8. Horizontal bars represent the median of regional ventilation in percent at each specific pixel, while the whiskers are the outliers of extreme regional ventilation. Before lavage, median regional ventilation at the 15th pixel contributed the most to ventilation (55%). After lavage, the 13th pixel occupied the leading median regional ventilation of 48%, reflecting atelectasis in dorsal lung sections. After 2 hours and 4 hours of automated ventilation, the 14th pixel contributed the most to median regional ventilation of 45% and 50%, respectively, signifying the recruitment of previously atelectatic surface.

Box-and-whisker plots indicating the median (25th to 75th percentiles) are shown in Figure 9. These plots quantitatively describe various significant parameters for all seven pigs. The parameters are presented before lavage, after lavage and every 0.5 hours. During the process of lavage inducing ARDS, PaO₂/FiO₂ was evaluated by ABG. The median of PaO₂/FiO₂ was 70 mmHg and all cases were below 100 mmHg, representing severe ARDS. By regulating SaO₂ at 88 to 95% in Figure 9a, PaO₂/FiO₂ values were improved for all cases by the protocol as shown in Figure 9b. At 2.5 hours, disconnection at the patient–ventilator interface was carried out and the median of PaO₂/FiO₂ was 94 mmHg. The 4 hours of ventilation using the protocol increased PaO₂/FiO₂ and the ARDS condition generally improved from severe ARDS to moderate ARDS based on the Berlin definition [15].

In Figure 9c, the P_plat goal ≤30 cmH₂O was satisfied in most cases. However, the pH value in Figure 9d was slightly lower than the pH goal between 7.30 and 7.45 in the first 1.5 hours of ventilation period for the protocol. Initial minute ventilation or baseline minute ventilation may increase from 4 to 5 l/minute to improve the pH value at the beginning of automatic ventilation.

Based on PaCO₂ measured from ABG every 0.5 hours, Figure 9e represents permissive hypercapnia with an approximate value of 60 mmHg. Whilst Figure 9f shows etCO₂ of an average 43 mmHg during ventilation therapy, etCO₂ differed significantly from PaCO₂, indicating a diffusion problem.