Experimental Lung Research, 40, 186–197, 2014 Copyright © 2014 Informa Healthcare USA, Inc. ISSN: 0190-2148 print / 1521-0499 online DOI: 10.3109/01902148.2014.900156

ORIGINAL ARTICLE

A new adaptive controller for volume-controlled mechanical ventilation in small animals ¨ Robert Huhle, Peter M. Spieth, Andreas Guldner, Thea Koch, and Marcelo Gama de Abreu Department of Anaesthesiology and Intensive Care Therapy, Pulmonary Engineering Group, University Hospital Carl Gustav Carus, Dresden University of Technology, Fetscherstr. 74, Dresden, Germany AB STRACT Background: This study aimed to develop and evaluate an adaptive control system for volume-controlled ventilation (VCV) in small animals to guarantee accurate delivery of tidal volume (VT ) in the presence of changes in lung mechanics. Methods: The adaptive control system to control the Harvard Inspira ventilator was designed and evaluated on a custom-made physical model during step changes of resistance and elastance of the respiratory system assessing difference in minute ventilation (MVc ) during convergence cycles (NC ). The controller was then evaluated during conventional and variable volume VCV in rats with acute respiratory distress syndrome (ARDS) induced by intratracheal HCl (six animals/group), where the difference between desired and applied VT (dVT,d ), its root-mean square error (RMSE) and relative deviation from target minute ventilation (MV) were determined. Results: The controller showed fast convergence NC < 20 cycles with an acceptable MVC < 10% in simulations and nearly abolished dVT,d (VCV: 0.23 ± 0.1 mL to 0.0 ± 0.0 mL, P < .001 and vVCV: 0.05 ± 0.8 mL to 0.0 ± 0.0 mL, P < .001), significantly reduced RMSE (VCV: 0.23 ± 0.1 to 0.04 ± 0.01 mL, P < .001 and vVCV: 0.13 ± 0.04 to 0.08 ± 0.02 mL, P < .001) and MV (VCV: 11.6 ± 4.2 to 0.04 ± 0.15%, P < .001 and vVCV: −3 ± 3.8 to −0.35 ± 1.3 %, P < .001) in animal experiments. In VCV the improvement was more pronounced, due to reduced respiratory system elastance in this group (VCV: 5.6 cmH2 O mL−1 versus vVCV: 3.8 cmH2 O mL−1 , P < .001). Conclusions: The new adaptive controller ensured accurate delivery of VT in VCV and proved valuable for mechanical ventilation of small animals especially in ARDS research. KEYWORDS acute respiratory distress syndrome, closed-loop system, mechanical ventilation, small animals, volume-controlled ventilation

sated by adjustment of ventilator settings, but since lung mechanical properties may vary over time in models of lung disease, differences between VT set at the ventilator (VT,s ) and ultimately measured VT (VT,m ) may be overlooked. Such differences may amount to 50% in small animals, if no compliance compensation is used [1]. In our setup, such differences may exceed 25% for a VT of 6 mL kg−1 in rats during volume-controlled, variable inspiratory flow, ventilation (VCV) with a piston type ventilator, possibly resulting in VT outside the protective range of 4–8 mL kg−1 [2] as suggested for patients with acute respiratory distress syndrome (ARDS) [3]. Furthermore, in acutely injured lungs, delivered VT might thus be reduced to unacceptably low values. In situations where application of high tidal volumes is intended such as during studies of ventilator-induced lung injury, applied VT might actually fall within the protective range.

BACKGROUND Investigations on mechanical ventilation are frequently performed in small animals. Even though dedicated mechanical ventilators are available, target and delivered tidal volume (VT ) may differ importantly. For example, tidal ventilation may be partially lost in the ventilator tubing system due to air compression and/or changes in tubing geometry with pressure. Normally, such losses are easily compen-

Received 3 December 2013; accepted 27 February 2014 We thank Hannes Krause for essential support in conducting the bench tests and Hannes Krause, Maria Lippmann, and Nadine Oppermann for essential support in conducting the animal experiments. Address correspondence to Robert Huhle, Department of Anaesthesiology and Intensive Care Therapy, Pulmonary Engineering Group, University Hospital Carl Gustav Carus, Dresden University of Technology, Fetscherstr. 74, 01307 Dresden, Germany. E-mail: [email protected]

186

Adaptive Controller for Small Animal Mechanical Ventilation

a)

b) -0.15

0

-0.25

-0,5

-0.35

dVT,s in mL

dVT,s in mL

187

-0.45

-1

-1,5 -0.55 1st Order Polyn. Fit R² = 0.9709 -0.65

2

4 6 E in cmH2O / mL

2nd Order Polyn. Fit R² = 0.9986

-2 8

1

2

3

4 5 VT,s in mL

6

7

FIGURE 1. Dependence of tidal volume difference dVT,s on respiratory system elastance E (A) and set tidal

volume VT,s at the ventilator (B). Mean values of dVT,s at different elastance E of the described physical test lung and different set tidal volumes at the inspirator VT,s (E = 7.5 cmH2 O/mL) with their corresponding 1st/2nd order polynomial fits and its coefficients of determination R2 . The dependents of dVT,s on the elastance E of the physical test lung was evaluated with conventional VCV at a respiration rate of 60 bpm and a tidal volume VT,s of 2 mL.

As previously found during bench tests performed in our lab, the observed difference between measured tidal volume VT,m and set tidal volume VT,s (dVT,s ) was dependent on the elastance of the test lung, and VT,s itself Figure 1A. A linear relation between dVT,s and the elastance E was found, indicating an increasing absolute dVT,s up to 30% of VT,s . The relationship between dVT,s and VT,s was found to best fit a second-order polynomial, as shown in Figure 1B. However, the coefficients of such polynomial vary when the elastance of the lung changes. Thus, an adaptive control algorithm is necessary to perform adjustments and minimize the error in desired VT (VT,d ). To overcome those limitations, we implemented and evaluated an adaptive control system (ACS) for VCV in small animals, which can be used with a commercial mechanical ventilator. The performance of the ACS was evaluated at the bench with a custommade physical model of the respiratory system upon two relevant challenges: (I) during cycle-by-cycle random variation of VT in volume controlled ventilation (vVCV) [4,5]; (II) after step changes in elastance (E) and resistance (R). Finally, the ACS was tested in an animal study of experimental ARDS in rats (III).  C

2014 Informa Healthcare USA, Inc.

METHODS Development of the Adaptive Control System The input of the ACS (Figure 2) is the desired VT for the forthcoming cycle (VT,d ), which may be constant (e.g., conventional VCV), or vary randomly cycle-bycycle (e.g., variable VCV). Prior to the beginning of a given respiratory cycle i, the tidal volume to be set at the ventilator (VT,s ) is derived from the desired tidal volume VT,d as shown in Equation (1) (Block “Adaptive Control,” Figure 2), which is an expansion of the standard linear least-square method to incorporate the quadratic relationship between VT,d and VT,s : 2 VT,s (i ) = VT,d (i ) + a(i ) · VT,d (i ) + b(i ) · VT,d (i )

(1)

with VT,d determined a priori by the operator as a single value, or as a sequence of values. Parameters a and b represent adaption step factors that are calculated to minimize dVT,s by the least-squares method [6] (Block “System Identification” in Figure 2). The appropriate number of past cycles to be considered for adaption w was determined by numerical simulations, as described further.

188

R. Huhle et al.

a, b

SYSTEM IDENTIFICATION

VT,m(ΔP)

mesh resistance

chest wall

ΔP

VT,d

ADAPTIVE CONTROL

Ventilator VT,s volume air → E

plethysmographic chamber

perimeter → R

test lung

FIGURE 2. Schematic diagram of the adaptive control system and the bench evaluation setting. The ACS calculates VT,s to be set

at the ventilator for the forthcoming cycle according to desired tidal volume VT,d . The adaptation step factors a and b are derived from past values of measured VT,m and set tidal volume VT,s . The physical model of the respiratory system consisted of an outer and two inner rubber balloons placed within a plethysmographic chamber. The volume of air in the inter-balloons space was varied with a 25 mL syringe to modulate the elastance (E) of the model. The resistance (R) was modulated through changes of the perimeter of the elastic tube connecting the ventilator’s Y-piece to the inner rubber balloons, which led to changes in its cross-sectional area. Tidal volume was measured through differential pressure measurement over the mesh resistance.

Numerical Simulations Numerical simulations were performed with routines written for Matlab (Mathworks, MA, USA) using the relationship between volume and pressure of the respiratory system known as equation of movement, which is shown in Equation (2) (see ref. [7]): Paw = R · V˙ + E1 · V + E2 · V 2 + EEP

(2)

where Paw is the airway pressure, R the resistance of the respiratory system, V˙ is air flow, V is the air volume, and EEP is the airway pressure (Paw ) at zero V˙ and V, which is close to the end-expiratory pressure (set at zero in the numerical simulations). E1 and E2 represent the linear and the nonlinear components of the elastance of the respiratory system, respectively. The volume difference dVT results from air compression in the system, and can be approximated from a model of compliance of air volume in a rigid vessel, as shown in Equation (3): dVT =

V0 · Paw = VT,s − VT,m , κ · p0

(3)

where V0 represents the sum of air volumes in lungs, connection tubing and piston at the reference atmospheric pressure p0 , and κ the adiabatic exponent of air. Hence, the relationship between the input and output of the controlled system (VT,s and VT,m , respectively) is of quadratic nature and depends on E1 , E2 , R, V0 , and p0 , where the latter two are assumed to be constant. Therefore, the stability and speed of con-

vergence of the control system for a given combination of values depends both on w and VT,d sequence. The following scenarios, which are summarized in Table 1, were used to evaluate the ACS in numerical simulations: (N1) transition from conventional VCV to variable VCV with 30% coefficient of variation (CV) in VT and Gaussian distribution of randomly generated values. The CV of 30% was used because it was associated with an optimal response in gas exchange and respiratory system mechanics in an animal model of ARDS [8]; (N2) step increase in R by the factor of 10 (N2low ) and 100 (N2high ); (N3) step increase in E1 and E2 by a factor of 10 and 100, respectively, in N3low and N3high , at N3high being 10 times the range of values between healthy and severely injured lungs [9]; (N4) two consecutive cycles of corrupted tidal volume measurement VT,m , a problem specific for the open chamber configured plethysmograph, with random errors of 25/50% of VT,s , in order to estimate the robustness of the control system; (N5) simultaneous step increase of R and E1 /E2 as a combination of N2 and N3, which reflect changes in respiratory mechanics in the HCl aspiration model of ARDS in rats [7]. Simulations N1 to N4 were performed in an explorative fashion, and w was increased from 2 to 100 in incremental steps of one. For each value of w one set of simulations was performed. An appropriate value of w was obtained as a compromise between convergence speed and magnitude of overshooting in simulations N1 to N3. In simulation N4, the influence of short-term flow measurement corruption of two cycles on the consecutive control cycles was studied. We Experimental Lung Research

Adaptive Controller for Small Animal Mechanical Ventilation TABLE 1.

189

Characteristics of the Numerical Simulations

Parameter undergoing step change

Low magnitude

High magnitude

Constant parameter

0 → 30

0 → 60

0.03 → 0.3

0.03 → 3

0.11/0.011 → 1.09/0.11 0 → 25

0.11/0.011→ 10.9/1.09 0 → 50

VT,d = 2.5; E1 = 1.09; E2 = 0.11; R = 0.3 VT,d = 2.5; E1 = 1.09; E2 = 0.11; CV = 30 VT,d = 2.5; R = 0.3; CV = 30

see N2 and N3

see N2 and N3

(N1) VCV -> vVCV with step CV in % (N2) step change in R (N3) step change in E1 /E2 (N4) measurement corruption in % of VT,s (N5) step change in R and E1 /E2

VT,d = 2.5; E1 = 1.09; E2 = 0.11; R = 0.3; CV = 30 VT,d = 2.5; CV = 30

The values for the parameters are adapted to the target system’s goal setting of lung protective ventilation (VT,d = 6 mL·kg−1 ) of animals (body weight of 0.25 to 0.5 kg) and taken from literature [30]; desired tidal volume VT,d in mL; resistance R in cmH2 O·s·mL−1 ; volume independent component of elastance E1 in cmH2 O·mL−1 ; volume dependent component of elastance E2 in cmH2 O·mL−2 ; CV of VT,d in %; mean measurement corruption of five consecutive cycles in % of VT,m . Ventilation mode simulated was volume controlled ventilation with a respiratory rate of 60 bpm.

arbitrarily defined an average 10% of VT,s to be acceptable during the convergence phase. In simulation N5, the performance of the controller during onset of acute ARDS was evaluated. For each set of simulations with the numerical model, 100 individual runs were performed using different patterns of randomly generated, Gaussian distributed VT,d as input of the control system for each run. In each run, 600 iterations were performed representing 600 respiratory cycles.

Bench Evaluation of the New Adaptive Controller The ACS was implemented in Labview (National Instruments, Austin, Texas, USA) on a laptop connected to an Inspira Advanced Safety Ventilator (ASVp, Harvard Apparatus, Holliston, MS, USA) through a serial interface RS232. Tidal volume was ˙ was measet once per respiratory cycle. Air flow (V) sured by a resistive mesh in the wall of a full body plethysmograph (V = 2740 mL, Type 853, Harvard Apparatus) connected to a difference pressure sensor (DLP 2.5, Type 381, Harvard Apparatus) and analogue-to-digital converted with a sampling frequency of 1 kHz. Thus the plethysmographic chamber was used in an open chamber configuration to limit low frequency drift of air vapor content and temperature, as shown in Figure 2 [10–12]. The performance of the ACS was evaluated with a custom-made physical model of lungs and chest wall of rats, which was placed in the plethysmographic chamber as shown in Figure 2. The lung model consisted of three rubber balloons, where the outer balloon mimicked the chest-wall, and the two smaller balloons mimicked the right and left lung. The outer balloon was connected to a 25 mL syringe and air was  C

2014 Informa Healthcare USA, Inc.

injected into the inter-balloon space to modulate the total elastance (E) of the test lung in the range of 2 to 7 cmH2 O·mL−1 . The test lung was connected to the Y-piece of the ventilator through an elastic tube, whose cross-sectional area could be changed to modulate the test lungs resistance (R) in the range of 0.2 to 1.2 cmH2 O·s·mL−1 . E and R values were determined after ventilation with VCV using least-mean squares fitting. The ventilator tubing system, including the Ypiece, was virtually the same used for mechanical ventilation in rats in our lab. Accordingly, the dead space was 0.2 mL, which represents ≈10% of the protective VT = 6 mL·kg−1 [3] in rats with 330 g body weight. The protocol for evaluation of the ACS consisted of two tasks with desired mean VT = 2.5 mL, a value chosen in analogy to the protective tidal volume range in humans for rats with an average body weight of 330 g: (P1) switching from conventional to variable VCV with randomly generated VT (Gaussian distribution, coefficient of variation, CV = 30%); (P2) step changes from ERlow (defined by low elastance E = 2 cmH2 O·mL−1 and low resistance R = 0.26 cmH2 O·s·mL−1 ) to ERhigh (defined by high elastance E = 7 cmH2 O·mL−1 and high resistance R = 1.3 cmH2 O·s·mL−1 ), which were derived from the HCl aspiration model of ARDS in rats [13]. Positive end-expiratory pressure, respiratory rate, and inspiratory/expiratory ratio were maintained at 5 cmH2 O, 60 min−1 and 1:1, respectively, for reasons of comparability to the physiological measurements. For each task, five repetitions were performed with different patterns of VT,d as input of the control system.

Physiological Measurements Following approval by the local animal care committee and the government of the State of Saxony (249168.11-1/2011-34), the ACS was evaluated also in

190

R. Huhle et al.

a model of the ARDS in 12 male Sprague-Dawley rats (312.5 g (265–360 g), age 51 days (48–62 days), Janvier France). Sample size calculation was based on our previous observations in this injury model. Animals were housed in groups of five under a 12 h/12 h light/dark cycle and had free access to a standard rodent diet and water until the beginning of experiments. Animals were anesthetized with ketamine (200 mg kg−1 i.p. bolus for induction and 100 mg (kg h)−1 i.v. for maintenance) and midazolam (2 mg kg−1 i.p. bolus for induction and 1 mg (kg h)−1 i.v. for maintenance). Muscle paralysis was achieved by intra-muscular application of atracurium (20 mg kg−1 bolus and subsequently 10 mg kg−1 every hour). Anesthetic agents for maintenance were administered in 7.5 mL kg−1 of balanced electrolyte solution (E153). Arterial blood pressure and ECG were used for hemodynamic monitoring. A closed-loop system was used to maintain physiological body temperatures during the procedure. After surgical preparation of the airway by insertion of a 14G metal cannula and induction of mechanical ventilation, lung injury was induced by intra-tracheal instillation of HCl (1.2 mL kg−1 , 0.2 N) [2]. Animals were then randomly assigned to 4 hours of mechanical ventilation with either VCV or vVCV (n = 6 in each group). Controlled ventilation was performed with a respiratory rate of 80 min−1 to keep arterial CO2 partial pressure at physiological levels. Inspiratory fraction of O2 (0.35), PEEP (5 cmH2 O) and a tidal volume (6 mL kg−1 ) were set to values deemed as “protective” in human clinical practice. CV was 0% (conventional VCV) and CV = 30% (vVCV), respectively. Measurements of respiratory airway pressure and flow were taken after preparation (baseline measurement—BL1), induction of lung injury, randomization (baseline 2—BL2) and every hour for 4 hours thereafter (T1 –T4 ). VT,s , VT,m , and VT,d and their differences dVT,s = VT,m − VT,s and dVT,d = VT,m − VT,d were recorded. At the end of experiments, animals were killed under deep anesthesia by exsanguination via arterial catheter (common carotid art). Lung mechanics, e.g., E and R, were determined offline from airway pressure and flow recordings. Finally, it was assumed that VT,s equals desired tidal volume VT,d = VT,s as a surrogate for the case that these experiments are done without the controller in place. Thereby, evaluation of the performance of the controller was possible without additional animals.

Performance Indices The convergence number (NC ) is defined as the number of respiratory cycles for which the RMSE be-

tween VT,d and VT,m of the previous five cycles was < 0.2 mL. The mean of absolute differences (MoD) of all cycles until convergence was used to measure instability/overshooting. Also, the relative difference in minute ventilation between measured VT,m and desired VT,d pattern was computed for the number of cycles until convergence (MVC in%). Additionally, the RMSE and the relative difference between measured and desired minute ventilation (MV in%) were calculated from the 60 preceding and subsequent cycles of a step change after convergence. RMSE and MV were obtained for each recording for the physiological animal experiments and compared between groups as well as between ventilation with and without closed-loop controller.

Statistics Values are given as median and ranges. Differences between experimental groups were tested using the Wilcoxon signed rank test. Significance was accepted at P < .05.

RESULTS Numerical Simulations Results for N1 are not shown, since NC was equal to zero and independent of number past cycles w. For tasks N2 to N4, the appropriate value for w was selected based on the two performance measures cycles to convergence NC and mean of absolute differences MoD, depicted in Figures 3 and 4, respectively. While NC is increasing linearly with window length w a biphasic relation between MoD and w becomes obvious. For w < 10 the range of MoD decreases exponentially in all experiments except for simulation N1, where it is independent of w. This range of values for w is considered to be unfavorable as the control system is adapting too fast and hence depending on the sequence of VT,d values possibly producing additional overshoot. For simulation N2 and N3 the range of MoD reaches an asymptotic value for w > 10, while in simulation N4 this asymptotic value is only reached for w > 70, however independent on the magnitude of measurement corruption. Conversely the median of MoD over all 100 simulations in task 4 is below 10% of mean VT,d for w > 16 and hence we selected a window length w = 17 to be appropriate for the selected control system. For the selected control system with a window length of w = 17 the performance measures showed a clear dependence on the magnitude of the step change in simulation task N5 (Table 2). NC and MoD Experimental Lung Research

Adaptive Controller for Small Animal Mechanical Ventilation

increased significantly with the magnitude of the step change more than doubling their respective median values. RMSE, the measure of the quality of adaption of the control system to the numerical model, was significantly different between the three different numerical models ERlow , ERmed , and ERhigh , however, their magnitude still remained well below 0.01 mL.

25 N2low N3low

20

NC in cycles

N4low 15

10

Bench Evaluation

5

0

5

10

15

20

25 30 w in cycles

35

40

45

50

FIGURE 3. Dependence of cycles to convergence NC to

window length w in simulation task N2 and N3. Expressed as mean of NC of 100 simulations per value of w—for each of the simulation task: N2 . . . StepRlow (defined as resistance step change 0.03 → 0.3 cmH2 O·s·mL−1 ), N3 . . . StepElow (defined as elastance step change 0.11/0.011 → 1.09/0.11 cmH2 O·mL−1 / cmH2 O·mL−2 ) and N4 . . . StepNoiselow (defined as five consecutive cycles of random measurement corruption of 25%) (cmp. Table 1). Results of experiment N1 are not displayed as the convergence time was equal to zero and independent of the window length w. Slash-dotted line indicates the value w = 17 chosen for bench test and physiological measurements.

0.5 N2low

MoD in ml

N4noise

Physiological Measurements

0.3 0.25 0.2

0.1

5

10

15

20

25 30 w in cycles

35

40

45

50

FIGURE 4. Mean of difference (MoD) in milliliter during the convergence of the control system after a step change in the respective simulation task. N2 . . . StepRlow (defined as resistance step change 0.03 → 0.3 cmH2 O·s·mL−1 ), N3 . . . StepElow (defined as elastance step change 0.11/0.011 → 1.09/0.11 cmH2 O·mL−1 / cmH2 O·mL−2 ) and N4 . . . StepNoiselow (defined as five consecutive cycles of random measurement corruption of 25%) (cmp. Table 1). Results of experiment N1 are not displayed as the convergence time was equal to zero and independent of the window length w. Slash-dotted line indicates the value w = 17 chosen for bench test and physiological measurements.

 C

On average, the ACS converged within seven cycles with MoD of 0.2 mL and a MVC of 2.5% in bench test P1 (Table 3). In its worst performance, the ACS converged within 17 cycles. RMSE significantly increased in variable compared to constant VCV, with a median value of 0.18 mL. Absolute MV did not differ significantly between variable and conventional VCV. The results of the bench test P2 are also depicted in Table 3 and representative pattern of convergence of ACS during the bench evaluation P2 is shown in Figure 5. R and E differed significantly from ERlow to ERhigh , with comparably high accuracy, namely 10% in injured animals (E ∼ 7 cmH2 O mL−1 ) compared to VT,s (Figure 6A). dVT,d depended on elastance E but not on resistance R (Figure 6B). With the controller active dVT,d (VCV: 0.23 ± 0.1 mL to 0.0 ± 0.0 mL, P < .001 and vVCV: 0.05 ± 0.8 mL to 0.0 ± 0.0 mL, P < .001) was nearly abolished. Furthermore RMSE (VCV: 0.23 ± 0.1 to 0.04 ± 0.01 mL, P < .001 and vVCV: 0.13 ± 0.04 to 0.08 ± 0.02 mL, P < .001) and MV (VCV: 11.6 ± 4.2 to 0.04 ± 0.15%, P < .001 and vVCV: −3 ± 3.8 to −0.35 ± 1.3 %, P < .001) were significantly reduced during closed-loop controlled ventilation regardless of experiment group. However, both performance indices were reduced to a higher degree in VCV

192

R. Huhle et al.

TABLE 2. Results of Numerical Simulation N5–Step Change of Elastance and Resistance for the Selected Control System with Window Length w = 17

NC in cycles MoD in mL MVC in %

RMSE in mL MV in %

ERlow → ERmed

ERlow → ERhigh

3 (2–7) 0.14 (0.064–0.36) 0.14 (0.04–0.25)

12 (9–15) 6.86 (3.85–10.1) 4.5 (2.5–6.6)

Sgn xxx xxx xxx

ERlow

ERmed

ERhigh

0.27 (0.19–0.39) × 10−7 21 (1.5 . . . 38) × 10−8

0.30 (0.22–0.46) × 10−5 0.25 (0.08–0.48) × 10−4

1.8 (1.3–2.7) ×10−3 1.5 (0.8–2.9) × 10−1

+++∗∗∗ +++∗∗∗

Numbers are medians (range) of all 100 simulation runs. Number of cycles until convergence NC ; mean of differences MoD; difference in minute ventilation during convergence MVC and after convergence MV; root-mean square error RMSE; ERlow (defined by elastance E1 /E2 = 0.11/0.011 cmH2 O·mL−1 / cmH2 O·mL−2 and resistance R = 0.03 cmH2 O·s·mL−1 ); ERmed (defined by elastance E1 /E2 = 1.09/0.11 cmH2 O·mL−1 / cmH2 O·mL−2 and resistance R = 0.3 cmH2 O·s·mL−1 ) and ERhigh (defined by elastance E1 /E2 = 10.9/1.09 cmH2 O·mL−1 / cmH2O·mL−2 and resistance R = 3 cmH2 O·s·mL−1 ). Significance was tested using non-parametric Wilcoxon rank-sum test (xxx/+++/∗∗∗ P < .001 between stepERlow→med − stepERlow→high / ERlow − ERmed / ERmed − ERhigh ).

compared to vVCV (∼90% versus ∼50% reduction during VCV and vVCV, respectively).

DISCUSSION The main results of this work were that, in a physical model of the respiratory system of rats, the novel ACS for VCV: (I) adapted VT to the mechanical properties of the physical model of the respiratory system independent from the magnitude of mechanical properties of the lungs; (II) showed fast convergence within TABLE 3.

20 cycles without overshoot/instability, with acceptable loss of minute ventilation during the convergence period (MVC < 10%); (III) resulted in increased accuracy of applied tidal volume in a model of ARDS in rats. To our knowledge, this is the first work describing the implementation and evaluation of an adaptive control algorithm for VCV for small animals. We tested the ACS in physical simulations in three relevant scenarios: (I) random variation of VT ; (II) sudden changes in R and E, and (III) in an experimental study comparing ventilation with and without ACS of intra-tracheal HCl injured as well as healthy rats.

Evaluation of the Adaptive Control System Performance During Volume-Controlled Ventilation in the Physical Lung Model CV: 0 → 30

P1 NC in cycles MoD in mL MVC in %

RMSE in mL MV in %

6 (5–17) 0.21 (0.18–0.41) 2.5 (−7–9.8) CV = 0

CV = 30

Sign.

0.07 (0.04–0.12) −0.05 (−0.18–0.63)

0.18 (0.09–0.22) −0.66 (−1.9–0.29)

n.s.

ERlow

ERhigh

Sign.

0.26 (0.24–0.28) 1.89 (1.86–1.92) 0.07 (0.06–0.09) −0.31 (−0.89–−0.09)

1.26 (1.08–1.40) 7.05 (6.60–7.50) 0.07 (0.04–0.10) −0.35 (−0.83–0.12)

∗∗∗

ERlow → ERhigh

P2 NC in cycles MoD in mL MV in %

R in cmH2 O·s·mL−1 E in cmH2 O·mL−1 RMSE in mL MV in %

∗

14 (11–20) 0.46 (0.33–0.56) 9.65 (5.64–19.61)

∗∗∗

n.s. n.s.

Values are given as median (range) of all simulation runs with the physical model of the respiratory system. Significance was tested using non-parametric Wilcoxon rank-sum test (n.s. P > .05, ∗∗∗ P < .001) with CV in %; Number of cycles until convergence, NC ; Mean of differences until convergence, MoD; Relative change in minute ventilation during convergence, MVC ; Relative change in MV and RMSE when adapted.

Experimental Lung Research

VT,d, VT,m in ml

Adaptive Controller for Small Animal Mechanical Ventilation

4 3 2 1 -10

-5

0

5

10

15

20

25

30

-5

0

5 10 15 respiratory cycle #

20

25

30

dVT in ml

1.5 1 0.5 0 -0.5 -10

FIGURE 5. Representative example of the transient state of the adaptive control system after sudden change in mechanical properties of the physical model of the respiratory system. During random variable tidal volumes, a step change of elastance and resistance was applied at cycle zero (vertical arrow). Top: open circles: desired tidal volume VT,d , full circles: measured tidal volume VT,m ; Bottom: dVT = VT,d − VT,m from bench tests and numerical simulations with the same VT sequence indicated by x and open diamonds, respectively. The horizontal arrow indicates the median of the number of cycles needed for the system to converge (NC = 17).

dVT,d = VT,m - VT,d in ml

a)

b) 0.25

0.25

0.2

0.2

0.15

0.15

0.1

0.1

0.05

0.05

0

0

-0.05

-0.05

-0.1

-0.1

-0.15

-0.15

-0.2

-0.2

-0.25

-0.25 1

2 3 (VT,m - VT,d) / 2 in ml

4

2

3

4 5 6 E in cmH2O / ml

7

FIGURE 6. (A) Bland-Altman-Plot of measured VT,m − VT,d tidal volume versus their mean value

without the controller during vVCV and (B) dependence of difference dVT,d on respiratory system elastance E with (cross) and without (filled circles) closed-loop control. With upper and lower limits of agreement (dotted lines) and mean (dash-dotted line) in (A); difference dVT,d = VT,m − VT,d with (cross) and without (closed circles) the controller in (B). Straight lines in (B) are linear fits for performance without (full line) and with (dotted line) active controller with respective RMSE of 0.1252 and 0.0030.  C

2014 Informa Healthcare USA, Inc.

193

194

R. Huhle et al.

TABLE 4

Results of Physiological Measurements with (IndexCLC ) and Without Closed-Loop Control (IndexnoCLC )

E in cmH2 O·mL−1 R in cmH2 O·s·mL−1 VT,d in mL VT,s in mL VT,m in mL dVT,noCLC in mL dVT,CLC in mL RMSEnoCLC in mL RMSECLC in mL MVnoCLC in % MVCLC in %

VCV vVCV VCV vVCV VCV vVCV VCV vVCV VCV vVCV VCV vVCV VCV vVCV VCV vVCV VCV vVCV VCV vVCV VCV vVCV

BL1

BL2

T1

T2

T3

T4

2.4 2.6 0.16 0.17 1.97 1.81 1.89 1.73 1.97 1.80 −0.09 0.08 0 0 0.21 0.21 0.04 0.05 −6.15 −5.45 −0.29 0.10

6.2 4.2 0.28 0.26 1.97 1.94 2.30 1.92 1.95 1.92 −0.33 0.03 0 0.01 0.31 0.14 0.04 0.08 15.5 −3.13 0.15 −0.77

5.6 3.7 0.23 0.22 1.97 1.92 2.19 1.89 1.97 1.90 −0.21 0.03 0 0.01 0.20 0.14 0.04 0.09 10.3 −1.99 −0.06 0.65

5.3 3.7 0.20 0.20 1.97 1.95 2.19 1.92 1.97 1.94 −0.22 0.04 0 0.01 0.21 0.12 0.04 0.08 10.2 −3.14 0.01 −0.27

5.5 3.7 0.20 0.21 1.97 1.95 2.20 1.89 1.97 1.95 −0.22 0.06 0 0.01 0.21 0.14 0.03 0.08 10.6 −4.52 0.04 −0.72

5.6 3.7 0.21 0.20 1.97 1.93 2.22 1.92 1.97 1.93 −0.25 0.01 0 0 0.23 0.15 0.03 0.08 11.5 −3.73 0.09 −0.69

Sign. ∗

n.s. n.s. ∗

n.s. ∗∗∗

+++ n.s.

n.s. ∗∗∗

+++ n.s.

n.s. ∗∗∗

+++ n.s.

n.s.

Values were obtained during ventilation with closed-loop control (given as mean of all experiments). Significance was tested using GLM with factors group and time with Tukeys post-hoc test and correction for multiple comparisons according to Holms method (n.s. P > .05, and P < .01 with ∗/∗∗/∗∗∗ for group comparison and + /++ /+++ for comparison between with (CLC ) and without (noCLC ) controller). With respiratory system elastance E and resistance R; desired, set and measured tidal volume VT,d , VT,s , and VT,m , respectively, and their differences dVT,noCLC = dVT,s = VT,m –VT,s and dVT,CLC = dVT,d = VT,m − VT,d ; BL1 = baseline, BL2 . . . After injury and randomization, Time of mechanical ventilation after injury T1 = 1 h up to T4 = 4 h.

The number of past cycles w was chosen as a compromise between fast convergence and robustness against single cycle flow measurement corruption, an issue specific for the plethysmographic chamber based flow measurement. Targeting variable volume controlled ventilation only, a reasonable mean of choosing w would have been to consider the length of the repetitive pattern used to draw cycle by cycle VT values. However this was not considered primarily as the implemented ACS is also useful for conventional volume controlled ventilation. Since VT is a major determinant of dynamic stress and strain in lungs [14], investigations on the functional and biological impact of mechanical ventilation on the respiratory system require accurate delivery of VT . Accordingly, tight control of VT represents a precondition for studies on mechanical ventilation in small animals. However, in our experience VT delivered by mechanical ventilators may diverge as much as 30% from setting values when the mechanical properties of the respiratory system worsen (elastance value > 7 cmH2 O / mL) which is actually even less than the value of 50% reported by other authors [1]. When the elastance of the respiratory system changes, air compression within the system also changes and this in turn affects the amount of lung expansion.

In addition, because of the system compliance of the ventilator and the connected tubing, the volume “lost” before reaching the airways, is also influenced by changes in the respiratory system elastance. Thus, an ACS is necessary to guarantee delivery of desired VT during mechanical ventilation in small animal models of ARDS without the need of measurement of the system compliance of the respiratory set-up. The ACS was technically easy to implement and founded on sound principles of closed-loop control theory. Although there are several possible combinations for the values of parameters used in the adaptation steps, we opted for a fixed combination derived from numerical simulations. Such combination resulted in adequate speed of convergence over a broad range of conditions of the respiratory system. The performance of the ACS was independent of the mechanical properties of the test lungs, even during situations that occur only during the onset of injury in animal experiments, for example, airway hyper-responsiveness [15] and bronchoconstriction [16]. In fact, the controller showed acceptable adaptation for use in a respiratory system with compromised resistive and elastic components of mechanics, overcoming the need for periodic manual Experimental Lung Research

Adaptive Controller for Small Animal Mechanical Ventilation

resetting of VT,s during mechanical ventilation and paving the way to study different patterns of tidal volume in small animals. The limitation of the adaptive approach for system compliance compensation is the necessity of reliable air flow measurement which is in general prone to long-term drift and also short-term measurement corruptions, depending on the sensor method. Other methods for system compliance compensation documented in the literature rely primarily on the more robust measurement of airway pressure. However these methods depend on the determination of system compliance after every modification of the experimental setup, which may not be trivial and may not guarantee correct delivery of tidal volume [17–21]. In contrast, the controller described in the present study requires solely a mechanical ventilator that can be remotely controlled and reliable measurement of air flow, for example, with pneumotachography or plethysmography. In the physiological measurements the controller ensured accurate delivery of tidal volume independent of tidal volume profile. The average differences between set and measured tidal volume were different between groups and almost null in the vVCV group, which might indicate a lack of necessity of the controller for variable ventilation. However this apparently logical conclusion can be falsified by looking at the strong dependence of dVT,s and respiratory system elastance and considering the fact that in the vVCV group elastance was consistently lower compared to VCV, as reported in the literature [22–28]. A likely explanation for the improvement of lung mechanics during variable VCV is dynamic recruitment of different lung areas with beneficial effects on stress and strain, and increased secretion of surfactant. vVCV thus prevents VILI and de-recruitment, promotes recruitment and improves lung mechanics and gas exchange [4,29]. The animal experiments confirmed that set and measured tidal volumes differ importantly during mechanical ventilation of small animals. These differences are likely caused by the system compliance of the ventilator and the corresponding tubing including the Y piece, which led to an increase/decrease of measured tidal volume >10% compared to set tidal volume for lower/higher values of elastance. In experimental ARDS, such deviations can be too high and interfere with the interpretation of results. Our experiments suggest that the present closed-loop control system is useful to guarantee the application of desired mechanical ventilation, particularly in injured lungs. This controller may be particularly useful when large changes in respiratory mechanics are expected

 C

2014 Informa Healthcare USA, Inc.

195

and especially for small animal studies in models of ARDS.

CONCLUSION The new ACS allowed accurate delivery of VT during VCV with a commercially available mechanical ventilator even during random variation of VT and sudden changes in the mechanical properties of the respiratory system. The controller may be useful for respiratory research in mechanically ventilated small animals.

Funding Research was funded partly by the Department of Anaesthesiology and Intensive Care Therapy, University Hospital Carl Gustav Carus, Dresden University of Technology and the German Research Council (Grant #: GA1256/1-1).

ABBREVIATIONS ACS

adaptive control system described in this manuscript ARDS acute respiratory distress syndrome CV coefficient of variation of cycle-to-cycle variable VT dVT,s difference between desired and applied VT : VT,m – VT,s dVT,d difference between desired and applied VT : VT,m – VT,d E1 volume independent elastance of the respiratory system volume-dependent elastance of the respiraE2 tory system airway pressure measured between endoPaw tracheal tube and Y-piece R (pneumatic) Resistance of the respiratory system RMSE root-mean-square error between two pairwise quantities MoD mean of absolute differences between two pairwise quantities number of cycles until converged NC airway pressure measured at the tracheal Paw tube (P)EEP (Positive) end-expiratory airway pressure VCV volume-controlled ventilation with cycle invariant tidal volume vVCV VCV with variable VT drawn from Gaussian white noise set VILI ventilator induced lung injury

196

V V˙ VT VT,s VT,m VT,d w

R. Huhle et al.

thorax volume expansion air flow, the first derivative with respect to time of V maximum V during one respiratory cycle (tidal volume) tidal volume set at the ventilator tidal volume measured with a plethysmographic chamber tidal volume desired or targeted by the physician number of past cycles to be considered for least-mean squares adaption

Declaration of interest: The authors report no conflict of interest. The authors alone are responsible for the content and writing of the paper.

REFERENCES [1]

Thammanomai A, Majumdar A, Bartol´ak-Suki E, Suki B: Effects of reduced tidal volume ventilation on pulmonary function in mice before and after acute lung injury. J Appl Physiol. 2007;103(5):1551–1559. [2] Matute-Bello G, Frevert CW, Martin TR: Animal models of acute lung injury. Am J Physiol Lung Cell Mol Physiol. 2008;295(3):L379–L399. [3] ARDS Network T: Ventilation with lower tidal volumes as compared with traditional tidal volumes for acute lung injury and the acute respiratory distress syndrome. The Acute Respiratory Distress Syndrome Network. N Engl J Med. 2000;342(18):1301–1308. [4] Suki B, Alencar AM, Sujeer MK, Lutchen KR, Collins JJ, Andrade JS Jr, Ingenito EP Zapperi S, Stanley HE: Life-support system benefits from noise. Nature. 1998;393(6681):127–128. [5] Gama de Abreu M, Spieth PM, Pelosi P, Carvalho AR, Walter ¨ C, Schreiber-Ferstl A, Aikele P, Neykova B, Hubler M, Koch T: Noisy pressure support ventilation: a pilot study on a new assisted ventilation mode in experimental lung injury. Crit Care Med. 2008;36(3):818–827. [6] Widrow B, Stearns SD: Adaptive Signal Processing. New Jersey: Prentice-Hall; 1985:504. [7] Carvalho A, Zin W: Respiratory system dynamical mechanical properties: modeling in time and frequency domain. Biophysical Reviews. 2011;3(2):71–84. ¨ [8] Spieth PM, Carvalho AR, Guldner A, Pelosi P, Kirichuk O, Koch T, Gama de Abreu M: Effects of different levels of pressure support variability in experimental lung injury. Anesthesiology. 2009;110(2):342–350. [9] Boker A, Graham MR, Walley KR, McManus BM, Girling LG, Walker E, Lefevre GR, Mutch WAC: Improved arterial oxygenation with biologically variable or fractal ventilation using low tidal volumes in a porcine model of acute respiratory distress syndrome. Am J Respir Crit Care Med. 2002;165(4):456–462. [10] Borg BM, Thompson BR: The Measurement of Lung Volumes Using Body Plethysmography: A Comparison of Methodologies. Respiratory Care [Internet]. 2012; Available from: http://www.ncbi.nlm.nih.gov/pubmed/22272788 [11] Chaui-Berlinck JG, Bicudo JEPW: The signal in total-body plethysmography: errors due to adiabatic-isothermic difference. Respiration Physiology. 1998;113(3):259–270.

[12] Cri´ee CP, Sorichter S, Smith HJ, Kardos P, Merget R, Heise D, Berdel D: Body plethysmography–its principles and clinical use. Respir Med. 2011;105(7):959–971. [13] Schulz H, Muhle H: Respiration. In: Krinke GJ, ed. The Laboratory Rat [Internet]. London: Academic Press; 2000 [cited 2012 Oct 16]. p. 323–44. Available from: http://www. sciencedirect.com/science/article/pii/B9780124264007500558 [14] Caironi P, Langer T, Carlesso E, Protti A, Gattinoni L: Time to generate ventilator-induced lung injury among mammals with healthy lungs: a unifying hypothesis. Intensive Care Medicine. 2011;37(12):1913–1920. [15] Allen GB, Leclair TR, Reyn J von, Larrabee YC, Cloutier ME, Irvin CG, Bates JHT: Acid aspiration-induced airways hyperresponsiveness in mice. J Appl Physiol. 2009;107(6):1763– 1770. [16] Labont´e I, Hassan M, Risse P-A, Tsuchiya K, Laviolette M, Lauzon A-M, Martin JG: The effects of repeated allergen challenge on airway smooth muscle structural and molecular remodeling in a rat model of allergic asthma. Am J Physiol—Lung Cell Mol Physiol. 2009;297(4):L698–L705. [17] Castle RA, Dunne CJ, Mok Q, Wade AM, Stocks J: Accuracy of displayed values of tidal volume in the pediatric intensive care unit. Crit Care Med. 2002;30(11):2566–2574. [18] Chow LC, Vanderhal A, Raber J, Sola A: Are tidal volume measurements in neonatal pressure-controlled ventilation accurate? Pediatr Pulmonol. 2002;34(3):196–202. [19] Al-Majed SI, Thompson JE, Watson KF, Randolph AG: Effect of lung compliance and endotracheal tube leakage on measurement of tidal volume. Crit Care. 2004;8(6):R398– R402. [20] Silvestri S: The influence of flow rate on breathing circuit compliance and tidal volume delivered to patients in mechanical ventilation. Physiol Meas. 2006;27(1):23–33. [21] Bachiller PR, McDonough JM, Feldman JM: Do new anesthesia ventilators deliver small tidal volumes accurately during volumecontrolled ventilation? Anesth Analg. 2008;106(5):1392–1400, table of contents. [22] Mutch WAC, Eschun GM, Kowalski SE, Graham MR, Girling LG, Lefevre GR: Biologically variable ventilation prevents deterioration of gas exchange during prolonged anaesthesia. Br J Anaesth. 2000;84(2):197–203. [23] Mutch WAC, Harms S, Ruth Graham M, Kowalski SE, Girling LG, Lefevre GR: Biologically variable or naturally noisy mechanical ventilation recruits atelectatic lung. Am J Respir Crit Care Med. 2000;162(1):319–323. [24] Arold SP, Mora R, Lutchen KR, Ingenito EP, Suki B: Variable tidal volume ventilation improves lung mechanics and gas exchange in a rodent model of acute lung injury. Am J Respir Crit Care Med. 2002;165(3):366–371. [25] Graham MR, Haberman CJ, Brewster JF, Girling LG, McManus BM, Mutch WAC: Mathematical modelling to centre low tidal volumes following acute lung injury: a study with biologically variable ventilation. Respir Res. 2005;6:64–74. doi:10.1186/1465-9921-6-64. [26] Spieth PM, Carvalho AR, Pelosi P, Hoehn C, Meissner C, ¨ Kasper M, Hubler M, Neindorff M von, Dassow C, Barrenschee M, Uhlig S, Koch T, Gama de Abreu M: Variable tidal volumes improve lung protective ventilation strategies in experimental lung injury. Am J Respir Crit Care Med. 2009;179(8):684– 693. [27] Graham MR, Goertzen AL, Girling LG, Friedman T, Pauls RJ, Dickson T, Espenell AEG, Mutch WAC: Quantitative computed tomography in porcine lung injury with variable versus conventional ventilation: recruitment and surfactant replacement. Crit Care Med. 2011;39(7):1721–1730.

Experimental Lung Research

Adaptive Controller for Small Animal Mechanical Ventilation [28] Thammanomai A, Hamakawa H, Bartol´ak-Suki E, Suki B: Combined effects of ventilation mode and positive endexpiratory pressure on mechanics, gas exchange and the epithelium in mice with acute lung injury. PLoS ONE. 2013;8(1):e53934. doi:10.1371/journal.pone.0053934.

 C

2014 Informa Healthcare USA, Inc.

197

[29] Brewster JF, Graham MR, Mutch WAC: Convexity, Jensen’s inequality and benefits of noisy mechanical ventilation. J R Soc Interf. 2005;2(4):393–396. [30] Krinke GJ. The Laboratory Rat. 1st ed. New York: Academic Press; 2000, ISBN: 012426400X.

Copyright of Experimental Lung Research is the property of Taylor & Francis Ltd and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.