Modelling primary blast lung injury: current capability and future direction

Primary blast lung injury frequently complicates military conflict and terrorist attacks on civilian populations. The fact that it occurs in areas of conflict or unpredictable mass casualty events makes clinical study in human casualties implausible. Research in this field is therefore reliant on the use of some form of biological or non-biological surrogate model. This article briefly reviews the modelling work undertaken in this field until now and describes the rationale behind the generation of an in silico physiological model.


INTRODUCTION
First described by Hooker in 1924 1 as a 'single gross lesion found post mortem after exposure to air concussion due to high-explosive', primary blast lung injury (PBLI) is currently defined as 'radiological and clinical evidence of acute lung injury occurring within 12 hours of exposure and not due to secondary or tertiary injury'. 2 It is a disease characterised by intraparenchymal haemorrhage, laceration and pneumothoraces. 3 In the absence of a specific biomarker or radiological hallmark, it can be difficult to distinguish with confidence PBLI from other forms of lung damage in complex patterns of injury. PBLI occurred in 7% of UK casualties in the most recent conflict in Afghanistan despite the rudimentary nature of the opposition forces 4 and it is likely that PBLI will be increasingly encountered by UK Defence Medical Services (DMS) in future, more industrialised, conflicts due to a combination of factors. A more economically capable opponent could be equipped with the wide variety of thermobaric weaponry that is currently readily available and has been used in the recent Balkan and Chechnian conflicts. 5 British military casualties exposed to such weapons are more likely to survive to reach hospital as improvements in personal protective equipment 6 and prehospital care have reduced immediate fatalities due to penetrating injury. 7 There is thus a need to increase our understanding of the pathophysiology of this disease and to create accurate research models of PBLI.

MODELLING-BASED RESEARCH
Modelling is the use of a surrogate entity to represent a complex system in a readily reproducible manner. Models can be either biological or non-biological. Biological models are further subdivided into in vitro (cell culture), ex vivo (live organ) or in vivo (live animal). 8 Non-biological models are either computational ('in silico') or physical (anthropomorphic) surrogates of the biological system of interest.
As a research technique, the validity of modelling parallels that of clinical trials or laboratory study. 9 Non-biological-based research is cheaper than animal modelling, requires less stringent ethical approval and can accommodate scenarios that are unachievable in live animal or human research, such as multiple casualties with multiple injury events. It can do this in an easily repeatable manner so that adequately powered studies which can achieve statistical significance can be undertaken. Modelling also facilitates the Ministry of Defence's ambition of limiting animal experimentation 10 and the impetus for the scientific community to 'Replace, Reduce and Refine' (the 3R's) when considering the use of live animals in research. 11 12 Both biological and non-biological models of primary blast injury to the chest exist and are in use. The original biological PBLI modelling work of note was undertaken by Bowen and colleagues in 1968 13 and is still used as a benchmark comparator by subsequent researchers despite significant weaknesses. Limitations include its use of a broad range of large and small animal species, the mixing of long and short duration blasts and the mixing of blast overpressure measuring modalities (reflected and incident measurements differ significantly for any given explosion introducing significant differences in recorded overpressure). This work suggested exposure injury and lethality thresholds, but having been undertaken almost 50 years ago does not reflect the significant advances in medical care achieved over this period. It also does not describe the severity of injury in survivors and the likely requirement for, and duration of, intensive care management. Blast injury research continues using both in vivo and ex vivo biological models. [14][15][16] Rodents are commonly used to model lung injury from a variety of mechanisms, including blast. 17 and human cadaveric specimens have been used to examine the effects of under-vehicle explosions on the lower limb. 18 A more recent example of in vivo blast research is the porcine work undertaken by Garner et al 19 at the Defence Science and Technology Laboratories (DSTL) in Porton Down. This work demonstrated a significant increase in mortality when haemorrhagic shock and blast exposure are combined, which subsequently lead to a change in resuscitation protocol within the DMS. The four arms of this study were limited to six to eight subjects because of the need to observe the 3R's and so could only accommodate the study of an immediately life-threatening combination of injuries (ie, coarse data) and not the intermediate-term and more subtle outcomes normally sought in medical intervention research.
One of the earliest examples of anthropomorphic modelling in blast lung research was the blast test device developed by the US military; it consists of a chest-shaped metal cylinder with four pressure gauges, one on each wall of the chest. This simple device allowed the reliable measurement of blast loading even in complex scenarios such as a confined space. 20 The Swedish Defence Research Institute subsequently developed a more complex chest surrogate -the Swedish Dummy Torso (Figure 1) aiming to produce a more biofidelic model. This model was constructed from a combination of strengthened rubber and foam with acoustic transmission facilitated through the use of water compartments; it was able to match the human chest in terms of compressibility and natural frequency. 21 All such models facilitate the measurement of physical blast loading in any given scenario but do not inform as to the physiological consequences of such loading.
Computational modelling has evolved in parallel to advances in computing power. Finite element modelling (FEM) treats the subject of interest as a 3D mesh of finite blocks, each of which has known mechanical properties. These individual components effect change on neighbouring units in a predictable manner and thus physical effects on the subject as a whole can be predicted. A FEM model was commissioned by the UK coroner's office after the suicide bombings on the London transport network in 2005. 22 This quickrunning model looked both at primary injury and secondary (fragmentation) injury resulting from detonation of an explosive device in a crowded area. It was able to generate an abbreviated injury score for casualties based on blast injury threshold limits and likely fragmentation injury and represents a significant step towards arming civil authorities and clinicians with clinically useful information. While much faster than most FEM models, it still requires 5 hours of computer run time to re-create 30 min of simulated time. 23 A FEM model of PBLI in sheep has recently been developed, which can accurately predict the volume of injured lung following a blast but remains unable to inform the medical community regarding the likely level of care such casualties would need and it does not facilitate the study of potential medical interventions. 24 FEM, however, normally takes several days per scenario and requires computing power that is not widely available. Despite advances in computer technology, FEM remains predominantly a tool to study structural, rather than physiological, consequences of injury.

Current model development
Our model is a modification of an existing in silico cardiorespiratory simulator developed by Nottingham University. 25 26 The Nottingham Physiological Simulator models the cardiorespiratory components of the human body via mathematical equations using the Matlab software package (The Mathworks, Natick, Massachusetts, USA).
The model assumes that a patient is mechanically ventilated and not contributing to respiratory effort. Both the cardiovascular and respiratory systems are divided into a series of individual components, each of which is described by a set of independent variables (Figure 2). At the beginning of a modelling study, these variables are initially set so that they represent the patient population to be studied. Once initiated, the model undertakes a series of predetermined physiological equations for a period of 30 ms, which represents one physiological time slice t. The end product of this series of equations then determines the values of the variables used in the next time slice; this iterative process continues for the study run time T. The respiratory element of the model consists of the mechanical ventilator and breathing circuit, physiological dead space (60 mL), anatomical and alveolar shunts and a variable number of ventilated alveoli, each of which has its own vascular component. Inhaled gases consist of oxygen, nitrogen, carbon dioxide, water vapour and gas α (anaesthetic or toxic gases). The cardiovascular element is composed of 19 compartments, each of which is described by both fixed parameters (unstressed volume and elastance coefficients, resistance and viscosity) and iteratively updated variables ( pressure, flow and volume). The systolic/diastolic cycling is modelled through a repeating pulsatile activation function of variable duration.
The numerical simulations of the integrated model provide results that agree with clinical data available in the published literature and the model has also been validated in a number of earlier studies. 27 Figure 1 Pictorial representation of the Swedish Dummy Torso.

ADAPTING THE MODEL TO REFLECT PBLI WITHIN THE MILITARY CONTEXT
In order to adapt this model to reflect PBLI, the known physiological responses to blast injury 28 were codified mathematically and applied to the model. This new model was then validated against the porcine cardiovascular and pulmonary data collected at DSTL. 19 29 In this model, terminally anaesthetised adult white pigs are exposed to a fixed sublethal blast dose and ventilation continued under anaesthesia for the duration of the trial period prior to terminal anaesthesia. 30 Our model has produced results closely matching these in vivo data for both blast and combined blast and haemorrhagic shock.
For the model to be of relevance to the DMS, it needs to meet several criteria. Primarily, it must be validated against human injuries experienced by UK service personnel suffering PBLI in combat. To this end, a clinical database of UK PBLI victims from the recent conflict in Figure 2 Pictorial representation of our current in silico primary blast lung injury model. Afghanistan is being created, which will be used to inform the model's blast dose-related physiological effect and outcome. The model must be able to be used throughout the chain of care from the point of wounding to rehabilitation, so needs to be able to accommodate the study of buddie-buddie care in a prehospital environment, potential medical interventions in a Role II/III emergency department and a variety of ventilatory approaches while mechanically ventilated in intensive care. In order to achieve this, several adaptions need to be made. It must be able to model spontaneous ventilation in the prehospital environment, the effect of possible modulators of pulmonary inflammation and biotrauma that could be administered both in the prehospital and in the emergency department and finally it should be able to replicate the consequences of intensive care management, including ventilator-induced lung injury, oxygen toxicity and a fluctuating fluid volume status. In addition, it is hoped to make the software sensitive to the age and gender of the casualty.
Concurrently, diagnostic CT criteria for identifying and quantifying PBLI are being developed, as well as early attempts to identify potential mRNA-based biomarkers for the disease. CT images consist of voxels (3D pixels), each of which can be interrogated for their density measured in Hounsfield units (HU). Existing imaging software (Analyze) allows 3D reconstructions of CT lung images from PBLI casualties to be created, which only display voxels from poorly or non-aerated lung tissue (voxel range of −250 to +250 HU; Figure 3). These data can also be used to quantify the proportion of lung tissue that is poorly or non-aerated as a consequence of PBLI (Figure 4). 31 Early evidence suggests that this method may prove useful in the identification of casualties with PBLI 32 and this work will be used to inform the computerised model of the proportion of non-functioning alveoli in human casualties in order to increase its fidelity and clinical range. Figure 4 The histogram data from the 3D CT lung reconstruction denoting the distribution of voxels (y-axis) and their densities in Hounsfield units (HU). Aerated lung exists between -1000 and -500 HU.

FUTURE DIRECTION
Despite this extensive modelling activity, it has not kept pace with advances in medicine such as physician-led prehospital care, highly orchestrated and effective emergency department management of critically injured casualties, intensive care therapy and CT imaging. It also fails to recognise the fact that improved prehospital care will result in increasingly severe cases of PBLI requiring management by the DMS. No model or measurable parameter exists that will inform clinicians of the degree of injury resulting from shockwave exposure alone, can predict the ongoing physiological compromise surviving casualties will suffer or allow clinicians to model different treatment, mitigating or preventative strategies. 33 It is the ambition of our group to create a militarily relevant blast lung injury model validated against human combat injury and augmented by specific serological and CT markers of disease severity that will facilitate future research in this field.
Contributors TS is the main author and guarantor. EH compiled section on CT analysis. MH, JH compiled section on our current model. EK contributed to section on previous modelling work. PM is the paper originator and main editor.