Computational Modeling of Primary Blast Lung Injury: Implications for Ventilator Management

Abstract

Primary blast lung injury (PBLI) caused by exposure to high-intensity pressure waves is associated with parenchymal tissue injury and severe ventilation-perfusion mismatch. Although supportive ventilation is often required in patients with PBLI, maldistribution of gas flow in mechanically heterogeneous lungs may lead to further injury due to increased parenchymal strain and strain rate, which are difficult to predict in vivo. In this study, we developed a computational lung model with mechanical properties consistent with healthy and PBLI conditions. PBLI conditions were simulated with bilateral derecruitment and increased perihilar tissue stiffness. As a result of these tissue abnormalities, airway flow was heterogeneously distributed in the model under PBLI conditions, during both conventional mechanical ventilation (CMV) and high-frequency oscillatory ventilation. PBLI conditions resulted in over three-fold higher parenchymal strains compared to the healthy condition during CMV, with flow distributed according to regional tissue stiffness. During high-frequency oscillatory ventilation, flow distribution became increasingly heterogeneous and frequency-dependent. We conclude that the distribution and rate of parenchymal distension during mechanical ventilation depend on PBLI severity as well as ventilatory modality. These simulations may allow realistic assessment of the risks associated with ventilator-induced lung injury following PBLI, and facilitate the development of alternative lung-protective ventilation modalities.

INTRODUCTION

High-intensity pressure waves generated by explosive blasts may result in injurious shearing and rupture of biological tissues, especially in thin-walled, gas-filled organs such as the lungs and bowels.^1,2 Blast-related injuries may also affect brain function, potentially resulting in traumatic brain injury, concussion-like symptoms, and other epigenetic alterations.^3–5 Primary blast injury caused by extreme pressure waves can occur in victims with or without secondary, tertiary, and quarternary injuries associated with fragmentation, whole-body displacement, burns, and toxin inhalation.^6,7 The severity and mortality of primary blast injuries have been shown to depend on proximity to blast origin, as well as exposure to reinforced pressure waves reverberating within enclosed spaces such as vehicles or bunkers.^8–10

Primary blast lung injury (PBLI) is associated with dyspnea, coughing, and hypoxia resulting from pulmonary hemorrhage and parenchymal injury.¹¹ Blast-wave propagation can produce rapid deformation of the thin lung parenchymal tissues, as well as rapid expansion and compression of gases contained in the airways and alveoli.^12,13 The transmission of such forces may result in the development of petechiae, gas emboli, pneumatoceles, and pneumothoraces.¹ Internal reflection of blast waves at the boundaries between the lung and surrounding tissue may result in constructive interference of pressure waves within the lungs, known as “stress concentration.”¹⁴ Such injurious mechanisms might explain the radiographic presentation of PBLI, often characterized as a “butterfly” pattern of bilateral perihilar infiltrates, depending on the orientation of the victim with respect to the pressure wave origin.^15–17 Disruption of ventilation and/or perfusion in regions of concentrated lung injury may result in widespread ventilation-to-perfusion mismatching and deficiencies in gas exchange.

Patients with PBLI may experience severe respiratory failure, necessitating mechanical ventilatory support to maintain gas exchange.^18,19 However, ventilator-induced lung injury (VILI) may also develop as a consequence of injurious parenchymal stresses, strains, and/or strain rates associated with cyclic overdistension and recruitment/derecruitment.^20–22 The resulting injury, inflammation, and cell death are suggested to accelerate the progression of respiratory distress and multisystem organ failure.²³ Preventative measures to reduce the risk of VILI include lung-protective ventilation, which relies on low tidal volumes or driving pressures to minimize cyclic overdistension, and positive end-expiratory pressure to minimize cyclic recruitment/derecruitment.^24–26 High-frequency oscillatory ventilation (HFOV) is an alternative therapy that relies on high mean airway pressures as well as tidal volumes less than the anatomic deadspace; these are theoretically ideal characteristics for lung-protective ventilation.²⁷

Oscillatory gas flows are not distributed uniformly throughout the lung, but rather according to regional mechanical properties.^28,29 This means that despite the reduction in tidal volume (i.e., to the entire lung) that is provided by lung-protective ventilation and HFOV, VILI may still occur in regions of locally amplified or diminished gas flows, especially in the presence of exacerbated mechanical heterogeneity.³⁰ Furthermore, it is difficult to ascertain the degree of heterogeneity in distributed parenchymal deformation at the bedside, especially during rapid flow oscillations. Nevertheless, such information may be useful for assessing the efficacy of any lung-protective ventilation strategy to mitigate extreme parenchymal strains, which may also be frequency-dependent and regionally heterogeneous.^31,32

In this study, we developed a computational model to simulate the distribution of parenchymal strains and gas transport during conventional mechanical ventilation (CMV) and HFOV under healthy and PBLI conditions. We hypothesized that the heterogeneity and frequency-dependence of regional ventilation in our model would be strongly dependent on the severity of PBLI, resulting in substantial differences in parenchymal strain, strain rate, and regional ventilation-to-perfusion matching.

METHODS

A glossary of acronym and symbol definitions is provided in the Supplementary Material. Our computational model consisted of a three-dimensional asymmetric branching airway network, based on the central airway tree segmented from a computed tomographic scan of an adult male human subject.³³ The segmentation included all airways from the trachea to the first generation of sub-segmental branches in each bronchopulmonary segment. Peripheral airway trees with smallest branches of 0.4 mm in inner diameter were then generated using a space-filling algorithm, with the central airway tree segmentation as the initial condition and lobar segmentations as boundary constraints.³³ The model consisted of 60,494 cylindrical airway segments, with 30,243 terminal bronchi subtended by viscoelastic acini.³³

Mechanical impedance was computed for each airway and acinus as a function of frequency, according to Colletti et al.²⁹ Briefly, each cylindrical airway segment was assigned a longitudinal impedance to account for resistive and inertial pressure losses due to gas flow, as well as a shunt impedance to account for gas compression and airway wall distension.^34,35 Airway wall thickness and cartilage fraction were extrapolated from a canine airway tree based on airway size.^36,37 The impedance of each nth acinus was modeled as a parallel arrangement of a gas compression compliance and a viscoelastic tissue impedance, with parameters for tissue elastance $(H_n)$ and hysteresivity $(\eta )$. Both $H_n$ and $\eta$ were allowed to vary as quadratic functions of the local mean transpulmonary pressure.³⁸ Pleural pressure was spatially distributed throughout the model, with an average value of −5 cmH₂O and a linear gravitational gradient of 0.25 cmH₂O cm⁻¹.³⁹ Mean airway lumen pressure was 10 cmH₂O for all simulations, resulting in 156 mL anatomic deadspace $(V_{\mathrm{D}})$ and 3,516 mL total acinar volume for an assumed healthy 70 kg adult male. One baseline model was constructed to simulate healthy conditions, and subsequently altered to simulate varying degrees of PBLI.

PBLI Conditions

PBLI was simulated by bilateral derecruitment and increased tissue elastance, focusing on the perihilar regions of the model. The value of tissue elastance in each nth acinus for the PBLI condition $(H_n^{\mathrm{PBLI}})$ was obtained using a spatially varying distribution function ${\mathrm{\Psi }}_n$:

$$H_n^{\mathrm{PBLI}}={\mathrm{\Psi }}_n{H}_n,$$ (1)

where ${\mathrm{\Psi }}_n$ was defined as a function of the distance between the nth acinus and each of two centroids located at the left and right hila:

$${\mathrm{\Psi }}_n=1+\sum _{i=1}^2\left({\mathrm{\Psi }}_i^{\max }-1\right)e^{-{\left({\xi }_n\frac{x_{n,i}}{{\tilde{x}}_i}\right)}^2},$$ (2)

where $i$ indexes the left and right centroids, ${\mathrm{\Psi }}_i^{\max }$ is the maximum multiplier at the nth centroid, $x_{n,i}$ is the distance between the ith centroid and nth acinus, ${\tilde{x}}_i$ is a radial length constant for the ith centroid, and ${\xi }_n$ is a pseudorandom gradient noise distribution evaluated at the nth acinus’ location.⁴⁰ Acinar derecruitment was determined by a threshold value for ${\mathrm{\Psi }}_n$, such that the nth acinus was considered derecruited if ${\mathrm{\Psi }}_n\ge 8$. Derecruited acini were modeled with infinite impedance, and thus received no flow. In addition to the healthy condition with zero acinar derecruitment, three injured conditions were simulated with the model representing mild, moderate, and severe PBLI, corresponding to 8%, 26%, and 40% derecruitment, respectively. Total lung compliance was reduced from approximately 77 mL cmH₂O⁻¹ for the healthy condition to 50, 37, and 28 mL cmH₂O⁻¹ for the mild, moderate, and severe PBLI conditions. These alterations in lung compliance and derecruitment (or shunt fraction) are consistent with reported values for patients with varying severity of the acute respiratory distress sydrome (ARDS), which is characterized by diffuse alveolar damage, hypoxemia, and ventilation-to-perfusion mismatching.^41,42 Figure 1 shows the spatial distributions of acinar tissue elastance for each condition in transverse sections, along with the corresponding three-dimensional airway tree renderings indicating the derecruited acini.

FIGURE 1.

Computational modeling of PBLI using distributions of modulated acinar tissue elastance (A) and derecruitment (B). Pseuorandom noisy distributions were used to mimic spatial distributions of parenchymal injury commonly associated with PBLI, concentrated most severely near in the perihilar regions of the lung.

Simulations

Gas exchange simulations were performed in one airway network model under four distinct conditions representing healthy baseline lungs, as well as mild, moderate, and severe PBLI with altered tissue elastance and derecruitment. Airway flow delivered at the trachea was distributed throughout the model according to regional mechanical impedance. At each airway bifurcation, flow was distributed to subtending lung regions in inverse proportion to the total impedance of each respective region.²⁹ Gas exchange was simulated using a model of convective and diffusive gas transport during oscillatory flow in a bifurcating airway network.⁴³ Total carbon dioxide elimination $({\dot{V}}_{\mathrm{tot}}{\mathrm{CO}}_2)$ was computed as a function of the delivered tidal volume $(V_{\mathrm{T}})$ and ventilation frequency $(f)$. Cardiac ouput $({\dot{Q}}_{\mathrm{tot}})$ was assumed to be 5.5 L min⁻¹, and perfusion to each acinus $({\dot{Q}}_n)$ varied linearly in the gravitational direction with a gradient of 0.11 L min⁻¹ cm⁻¹.^44,45 Simulations were performed for $f$ ranging from 0.1 Hz to 40 Hz, covering the range of typical rates during CMV (0.1 to 1 Hz) and HFOV (2 to 20 Hz). Eucapnic ventilation at each $f$ was determined according to the tidal volume $(V_{\mathrm{T}}^{\mathrm{euc}})$ for which total CO₂ elimination $({\dot{V}}_{\mathrm{tot}}^{\mathrm{euc}}{\mathrm{CO}}_2)$ was in equilibrium with metabolic CO₂ production (assumed to be 3.3 mL kg⁻¹ min⁻¹).⁴⁶ Figure 2 shows the processing pipeline for an example simulation of 7 Hz oscillation in a lung under moderate PBLI conditions.

FIGURE 2.

Computational processing pipeline for simulated oscillatory flow and gas transport, illustrated for 7 Hz oscillation in a moderate blast injury lung model. Processing pipeline of computational modeling and simulation. Thoracic CT scans are segmented to obtain central airways, which are then used as initial conditions for a space-filling algorithm to generate a three-dimensional network of smaller peripheral airways.³³ Models of mechanical impedance applied to each airway and terminal acinus yield frequency-dependent flow distributions.²⁹ Gas transport models are solved to find flow required for eucapnic carbon dioxide elimination.⁴³

Data Analysis

The relationship between $V_{\mathrm{T}}$ and ${\dot{V}}_{\mathrm{tot}}{\mathrm{CO}}_2$ was quantitatively assessed using model regression over a range of $f$ and $V_{\mathrm{T}}$ within ±10 % of $V_{\mathrm{T}}^{\mathrm{euc}}$ at each frequency. For low frequencies ranging from 0.1 Hz to 1 Hz, the following regression model was applied:

$${\dot{V}}_{\mathrm{tot}}{\mathrm{CO}}_2\left(f,V_{\mathrm{T}}\right)={\beta }_0{f}^{{\beta }_1}{\left(V_{\mathrm{T}}-{\beta }_3\right)}^{{\beta }_2},$$ (3)

where ${\beta }_3$ represents apparent deadspace, such that $(V_{\mathrm{T}}-{\beta }_3)$ is fresh gas penetrating the respiratory zone. The ${\beta }_1$ and ${\beta }_2$ parameters reflect the power law dependence of CO₂ elimination with respect to frequency and alveolar ventilation, respectively, where ${\beta }_0$ is a constant of proportionality. For high frequencies ranging from 2 Hz to 20 Hz, the following regression model was applied:

$${\dot{V}}_{\mathrm{tot}}{\mathrm{CO}}_2\left(f,V_{\mathrm{T}}\right)={\alpha }_0{f}^{{\alpha }_1}{V_{\mathrm{T}}}^{{\alpha }_2},$$ (4)

where ${\alpha }_1$ and ${\alpha }_2$ reflect the power law dependence of CO₂ elimination with respect to rate and amplitude of volume oscillations, respectively, and ${\alpha }_0$ is a constant of proportionality. High-frequency model regressions were applied over two adjacent frequency ranges: 2–10 Hz and 10–20 Hz. Model regressions were performed by a non-linear gradient search algorithm using a least-squares approach (MATLAB, The Mathworks, Inc., Natick, MA, USA). For each regression, a parameter covariance matrix was approximated using the inverse of the Hessian matrix of second derivatives for the sum of squared errors with respect to model parameter values, evaluated at the least-squares solution and normalized by an estimate of the noise variance.⁴⁷ Parameter standard errors were obtained from the square root of the diagonal terms of the covariance matrix.

The ventilation-to-perfusion ratio for each acinus was computed based on the corresponding acinar perfusion $({\dot{Q}}_{\mathrm{A},n})$ and ventilation $({\dot{V}}_{\mathrm{A},n})$. Although global alveolar ventilation is typically defined in terms of fresh gas ventilation for the entire lung (i.e., $f\left[V_{\mathrm{T}}^{\mathrm{euc}}-V_{\mathrm{D}}\right]$), this definition is not applicable to HFOV wherein $V_{\mathrm{T}}$ may be less than $V_{\mathrm{D}}$.⁴⁸ Instead, we defined ${\dot{V}}_{\mathrm{A},n}$ using the rate of acinar CO₂ elimination during eucapnic ventilation ${\dot{V}}_n^{\mathrm{euc}}{\mathrm{CO}}_2$:

$${\dot{V}}_{\mathrm{A},n}\left(f\right)=863\frac{{\dot{V}}_n^{\mathrm{euc}}{\mathrm{CO}}_2\left(f\right)}{P_{\overline{\mathrm{a}}}{\mathrm{CO}}_2},$$ (5)

where $P_{\overline{\mathrm{a}}}{\mathrm{CO}}_2$ is the partial pressure of CO₂ in mixed arterial blood, and 863 is a conversion factor between partial pressure in mmHg and gas volume fraction, accounting for the difference between standard temperature and pressure of dry air vs. body temperature and pressure of air saturated with water vapor.

Distributions of peak acinar volumetric strain $({\epsilon }_{\mathrm{V},n})$ and strain rate $({\dot{\epsilon }}_{\mathrm{V},n})$ during eucapnic ventilation were computed based on distributed acinar flow amplitudes. Peak strain $({\epsilon }_{\mathrm{V},n})$ was determined by the peak-to-peak range of acinar volume $(V_{\mathrm{T},n}^{\mathrm{euc}})$ divided by the minimum value:

$${\epsilon }_{\mathrm{V},n}\left(f\right)=\frac{V_{\mathrm{T},n}^{\mathrm{euc}}\left(f\right)}{{\overline{V}}_n-0.5V_{\mathrm{T},n}^{\mathrm{euc}}\left(f\right)},$$ (6)

where ${\overline{V}}_n$ is the acinar mean volume, determined by the mean airway pressure. Peak strain rate $({\dot{\epsilon }}_{\mathrm{V},n})$ was then estimated as:

$${\dot{\epsilon }}_{\mathrm{V},n}\left(f\right)=\pi f\cdot {\epsilon }_{\mathrm{V},n}\left(f\right),$$ (7)

Note that our definitions of strain and strain rate are based on intratidal changes in acinar gas volume, and therefore provide a surrogate for actual stretch experienced by acinar tissue.

RESULTS

Figure 3 shows the total tidal volume required to achieve eucapnea for the healthy and PBLI conditions. The transition frequency between CMV and HFOV $(f_{\mathrm{t}})$ was estimated to be 3.3 Hz, based on the nondimensional characterization of high-frequency gas exchange reported by Venegas et al.⁴⁹ The actual anatomic deadspace of the model, based on the sum of all airway segment volumes in the tree, is also indicated in Figure 3 (top panel). Note that $f_{\mathrm{t}}$ coincides with the transition from $V_{\mathrm{T}}^{\mathrm{euc}}>V_{\mathrm{D}}$ at low frequencies to $V_{\mathrm{T}}^{\mathrm{euc}}<V_{\mathrm{D}}$ at high frequencies. Also note that $V_{\mathrm{T}}^{\mathrm{euc}}$ was 15% higher for the severe PBLI condition compared to the healthy condition for frequencies less than 1 Hz, yet 10% lower for frequencies between 1 Hz and 10 Hz.

FIGURE 3.

Eucapnic tidal volume predicted by simulations for the healthy and PBLI conditions with increasing severity (mild, moderate, severe). Results are shown both in absolute units of mL (top) as well as a percentage relative to the healthy model (bottom). The dotted line indicates the anatomic deadspace volume (V_D). The solid black line indicates the transition between low-frequency CMV and high-frequency oscillatory ventilation.

Tables 1 and 2 provide the coefficients computed for the low- and high-frequency regression models defined by Equations (3) and (4), respectively. A reduction in the ${\beta }_3$ parameter indicates a reduction in the apparent deadspace of the model with increasing severity of PBLI. The relationship among $f$, $V_{\mathrm{T}}$, and $V_{\mathrm{tot}}{\mathrm{CO}}_2$ over the range of frequencies typical of HFOV (i.e., 2–10 Hz) was not substantially different across the severity of PBLI. However, the same regression model produced markedly different parameter values over the range of 10–20 Hz.

TABLE I.

Parameter Estimates (Standard Error) for Low-Frequency Regression Model $V_{\mathrm{tot}}{\mathrm{CO}}_2={\beta }_0{f}^{{\beta }_1}{\left(V_{\mathrm{T}}-{\beta }_3\right)}^{{\beta }_2}$, and Corresponding Coefficient of Determination $R^2$, Given $f$ Between 0.1 and 1 Hz, $V_{\mathrm{T}}$ in L, and $V_{\mathrm{tot}}{\mathrm{CO}}_2$ in L s⁻¹

	${\beta }_0$	${\beta }_1$	${\beta }_2$	${\beta }_3$	$R^2$
Healthy	0.0295 (0.4%)	0.8331 (0.2%)	0.8463 (0.3%)	0.1387 (0.2%)	0.9996
Mild	0.0257 (0.6%)	0.7905 (0.3%)	0.8169 (0.4%)	0.1322 (0.3%)	0.9990
Moderate	0.0224 (0.5%)	0.7580 (0.3%)	0.7827 (0.3%)	0.1208 (0.3%)	0.9993
Severe	0.0196 (0.5%)	0.7242 (0.3%)	0.7419 (0.4%)	0.1155 (0.4%)	0.9991

TABLE II.

Parameter Estimates (Standard Error) for High-Frequency Regression Model $V_{\mathrm{tot}}{\mathrm{CO}}_2={\alpha }_0{f}^{{\alpha }_1}{V_{\mathrm{T}}}^{{\alpha }_2}$, and Corresponding Coefficient of Determination $R^2$, Given $f$ Between 2 and 10 Hz or 10–20 Hz, $V_{\mathrm{T}}$ in L, and $V_{\mathrm{tot}}{\mathrm{CO}}_2$ in L s⁻¹

	${\alpha }_0$	${\alpha }_1$	${\alpha }_2$	$R^2$
$10\mathrm{Hz}\ge f\ge 2\mathrm{Hz}$
Healthy	0.1068 (11.4%)	0.9310 (3.6%)	2.3745 (3.4%)	0.8573
Mild	0.1181 (7.9%)	0.9014 (2.4%)	2.3798 (2.3%)	0.9283
Moderate	0.1441 (6.8%)	0.8891 (2.0%)	2.4419 (1.9%)	0.9510
Severe	0.1437 (4.6%)	0.8721 (1.3%)	2.3933 (1.3%)	0.9777
$20\mathrm{Hz}\ge f\ge 10\mathrm{Hz}$
Healthy	0.0125 (2.9%)	0.7420 (2.1%)	1.2256 (1.8%)	0.9600
Mild	0.0104 (3.0%)	0.7702 (2.3%)	1.1754 (2.0%)	0.9470
Moderate	0.0136 (3.8%)	0.7861 (2.5%)	1.3088 (2.2%)	0.9416
Severe	0.0178 (6.0%)	0.7992 (3.5%)	1.4354 (2.9%)	0.9006

Figure 4 shows the distributions of acinar ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$ for selected frequencies. During low-frequency ventilation, a unimodal log-normal distribution of ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$, centered around unity, was observed for the healthy condition. Under PBLI conditions, the ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$ distribution was bimodal: one mode corresponding to the ventilated and perfused recruited acini, and the other mode (for values of ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}\ll 1$, not shown in Fig. 4) corresponding to perfused but non-ventilated derecruited acini (i.e., shunt). Increasing the severity of PBLI also resulted in a rightward shift of the ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$ distributions for the recruited acini, reflecting increased $V_{\mathrm{T}}^{\mathrm{euc}}$ but diminished lung recruitment. Note that frequencies near $f_{\mathrm{t}}$ (3.3 Hz) resulted in multimodal distributions of ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$, reflecting the various fractions of acini that are either derecruited and non-ventilated, directly ventilated with fresh gas during each tidal oscillation, or indirectly ventilated via intrapulmonary mixing, dispersion, and diffusion.⁴³ The ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$ distribution became bimodal for frequencies greater than $f_{\mathrm{t}}$, reflecting the lack of direct acinar ventilation with fresh gas during HFOV with $V_{\mathrm{T}}^{\mathrm{euc}}<V_{\mathrm{D}}$.

FIGURE 4.

Histograms of distributed acinar ventilation-to-perfusion ratios during eucapnic ventilation of healthy and blast-injured lung models at four selected oscillation frequencies. Acinar derecruitment in blast-injury models result in substantial shunting of pulmonary blood flow (i.e., fraction of acini exhibiting approximately zero ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$).

Figure 5 shows the 90th percentile of acinar strain and strain rate during eucapnic ventilation. Note the substantial increases in strain observed at low frequencies for the PBLI conditions. Although $V_{\mathrm{T}}^{\mathrm{euc}}$ was only increased by 15% for the most severe PBLI condition, the 90th percentile of acinar strain increased more than three-fold. Increasing frequency resulted in minimal acinar strains, yet maximal strain rates.

FIGURE 5.

The 90th percentile of acinar strain (solid lines) and strain rate (dotted lines) induced during eucapnic ventilation as a function of oscillation frequency in healthy and blast-injured lung models. The solid black line indicates the transition from low-frequency CMV to high-frequency oscillatory ventilation.

DISCUSSION

Our computational modeling study of distributed flow and gas transport simulated bilateral perihilar derecruitment and heterogeneous tissue elastance that is consistent with PBLI. The model predicted increased tidal volumes required to maintain eucapnic ventilation in blast-injured lungs, reflecting deficiencies in gas exchange induced by severe ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$ mismatching. The increased amount of perfused but non-ventilated acini resulted in an elevated shunt fraction, as well as increased mechanical burden associated with gas exchange in the remaining recruited acini. Consequently, moderate increases in eucapnic tidal volume resulted in substantially amplified parenchymal strains across the diminished fraction of recruited lung. The concept of ventilating the “baby lung” is colloquially used to encourage mindfulness of the patient’s reduced lung recruitment and compliance.^42,50 Management of mechanical ventilation must support the metabolic demands of the patient, without injuriously straining the “baby lung.”

The discrepant strain heterogeneity between the healthy and PBLI conditions is further intensified by the heterogeneous respiratory mechanics associated with lung injury. If, for example, the recruited 60% of acini in the severe blast-injury model were inflated uniformly using the 15% increased $V_{\mathrm{T}}^{\mathrm{euc}}$, then one would expect individual acini to experience intratidal strains approximately two-fold greater than in the healthy model. However, the simulations performed using heterogeneous injured tissue elastance resulted in more than three-fold increases at the 90th percentile of acinar strain. This finding indicates that severe regional mechanical heterogeneity has potential to amplify injurious tidal stretch in the already overburdened “baby lung.” Such mechanical heterogeneity may also arise from surfactant dysfunction, microatelectasis, and/or parenchymal tissue injury.^51,52

The “baby lung” concept of a smaller functional lung was also evident in the low-frequency relationship among $f$, $V_{\mathrm{T}}$, and $V_{\mathrm{tot}}{\mathrm{CO}}_2$, which demonstrated a reduction in the apparent deadspace parameter (i.e., ${\beta }_3$, see Table 1) with increasing PBLI severity. The reduction in apparent deadspace can be explained by the reduction in recruited lung: flow delivered at the trachea is redistributed among remaining recruited acini, thus the airways terminating in derecruited acini no longer contribute to the effective anatomic deadspace. ${\beta }_3$ therefore describes the anatomic deadspace of the “baby lung.” However, the relative decrease in ${\beta }_3$ was not simply proportional to the reduction in lung recruitment, but rather was dependent on the spatial distribution of derecruitment. In the severe blast injury model with 40% derecruitment, ${\beta }_3$ was only reduced by 16%, indicating that the ratio of apparent deadspace to recruited lung volume was actually increased in the blast-injury models. This imbalance may also contribute to the relative increases in $V_{\mathrm{T}}^{\mathrm{euc}}$ for frequencies below 1 Hz.

By contrast, the imbalance between the apparent deadspace and recruited lung volume may have limited impact on gas exchange during HFOV, since direct acinar ventilation with fresh gas is not necessary to maintain sufficient CO₂ elimination with this ventilatory modality. For frequencies between 1 and 10 Hz, $V_{\mathrm{T}}^{\mathrm{euc}}$ was lower in the PBLI conditions compared to the healthy condition. The reduction in apparent deadspace resulted in more efficient intrapulmonary gas transport via the combined mechanisms of mixing, dispersion, and diffusion. However, this range of frequencies also demonstrated greater heterogeneity in the distributions of acinar ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$, likely due to the transition from CMV to HFOV. The multimodal distributions of ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$ observed in this frequency range reflect the various fractions of acini that are either non-ventilated, directly ventilated with fresh gas, or indirectly ventilated via high-frequency gas transport mechanisms. The relative exponents of $f$ and $V_{\mathrm{T}}$ during HFOV, ${\alpha }_1$ and ${\alpha }_2$, demonstrated consistency with experimental measurements, yielding ${\alpha }_1\approx 1$ and ${\alpha }_2\approx 2$ over the range 2–10 Hz.⁵³ Although the relative ratio of ${\alpha }_2$ to ${\alpha }_1$ (i.e., approximately 2:1) is preserved at the higher frequency range from 10 to 20 Hz, the values of these exponents were sensitive to the bandwidth over which they were regressed. Supporting this notion, the coefficients of determination $(R^2)$ were lower, and parameter uncertainties higher, for regression models covering the transition frequency range of 2–10 Hz compared to 10–20 Hz. Previous summaries of experimental studies have reported widely varying estimates of ${\alpha }_1$ and ${\alpha }_2$, although all consistently demonstrate that ${\alpha }_2>{\alpha }_1$, suggesting that gas exchange is more sensitive to relative changes in $V_{\mathrm{T}}$ compared to $f$ during HFOV.^43,53

Increasing the frequency of ventilation greatly reduced acinar strains to achieve eucapnea, yet concomitantly increased acinar strain rates. Both high strain and high strain rate are injurious, although safe thresholds for either quantity are difficult to define.⁵⁴ Given that both extremes of low and high frequencies result in maximal strains and strain rates, respectively, it may be possible to identify an optimal range of lung-protective frequencies to achieve a balance between the conflicting objectives of minimizing parenchymal strain vs. strain rate. It should be noted, however, that frequencies above 5 Hz produce diminishing returns of acinar strain reduction, despite consistent reductions in total delivered tidal volume. This can be explained by increasing flow heterogeneity at high frequencies, for which regional interactions between resistive, inertial, and elastic mechanical properties culminate in frequency-dependent ventilation heterogeneity.⁴³ Thus the 90th percentile of acinar strain did not continue to fall at the same rate as the total delivered tidal volume with increasing frequency. Furthermore, increasing heterogeneity at high frequencies also intensifies the increasing 90th percentile of acinar strain rate. Thus a frequency range between 1 and 5 Hz may be ideal for balancing the opposing injurious processes of strain vs. strain rate, as well as achieving a relative reduction in $V_{\mathrm{T}}^{\mathrm{euc}}$, despite increasing ${\dot{V}}_{\mathrm{A},n}/{\dot{Q}}_{\mathrm{A},n}$ heterogeneity. Further research in computational modeling of lung injury may reveal an objective relationship between oscillatory strain, strain rate, and VILI at the acinar scale, which may be used to refine optimized mechanical ventilation strategies.

Limitations

The limitations of the oscillatory flow and gas transport models are discussed by Colletti et al²⁹ and Herrmann et al.⁴³ In the present study, the development of a computational model to investigate PBLI involves several critical assumptions. First, flow distribution in the model was computed entirely in the frequency-domain for computational simplicity, assuming steady-state oscillation of a linear time-invariant system. Thus the model does not account for intratidal variations in mechanical properties such as dynamic strain stiffening during overdistension or acinar recruitment/derecruitment. Intratidal varations in lung mechanics may also be important factors in the redistribution of flow and gas transport in the acutely injured lung.^55,56 In addition, our PBLI conditions were simulated by modulating only acinar tissue elastance and a steady level of lung derecruitment. Nonetheless the syndrome of PBLI may feature many other pathophysiologic derrangements, such as airway obstruction, pneumatoceles, or pneumothoraces.¹ Hemodynamic variables such as perfusion distribution and cardiac output were not altered from the values used for the healthy model. Increased physiologic deadspace due to ventilation-to-perfusion mismatch, which can arise from obstruction of pulmonary capillaries, was not considered in our simulations. However such a feature is a significant predictor of mortality in the acute respiratory distress syndrome.^57,58 Aside from lung mechanical alterations representative of PBLI, our model does not account for other physiological or pathological alterations associated with underlying patient condition, environmental hazards, smoking, or aging. Despite constituting a potentially limited description of PBLI in a single model of an anatomical airway network, the computational modeling and simulation study presented here offer several insights into the distribution of regional ventilation and injurious strains within the acutely injured lung.

CONCLUSIONS

Mechanical ventilation of blast-injured lungs may be a life-sustaining intervention, however the frequency and volume amplitude of both conventional and oscillatory ventilation modalities should be chosen with careful consideration for the heterogeneous lung mechanics underlying the patient’s critical condition. Heterogeneity of distributed parenchymal stretch during PBLI exacerbates the risk of VILI within the already reduced and overburdened lung parenchyma that is available for gas exchange. Conventional respiratory rates below 1 Hz may induce injurious parenchymal strains in patients with substantial lung derecruitment and severe ventilation-to-perfusion mismatching. Increasing ventilation frequency may allow reductions in the tidal volume necessary to maintain eucapnia. However, oscillatory frequencies above 5 Hz may be at best ineffective at further reducing parenchymal strain, and at worst detrimental by enhancing parenchymal strain heterogeneity and strain rate.

Supplementary Material

Supplementary material is available at Military Medicine online.

Presentations

Presented as a poster at the 2017 Military Healthy System Research Symposium, Abstract # MHSRS-17-1492.

Funding

This work was supported in part by the Office of the Assistant Secretary of Defense for Health Affairs, through the Peer Reviewed Medical Research Program under Award No. W81XWH-16-1-0434 (D.W.K.), by the Medical Technologies Centre of Research Excellence at the University of Auckland (M.H.T.), by National Institutes of Health Grants R01 HL112986, and R01 HL126383 (D.W.K.), and by the Department of Anesthesia at the University of Iowa (D.W.K., J.H.). This supplement was sponsored by the Office of the Secretary of Defense for Health Affairs.