- Research
- Open Access

# Can computer simulators accurately represent the pathophysiology of individual COPD patients?

- Wenfei Wang
^{1}, - Anup Das
^{1}, - Tayyba Ali
^{2}, - Oanna Cole
^{2}, - Marc Chikhani
^{2}, - Mainul Haque
^{2}, - Jonathan G Hardman
^{2}and - Declan G Bates
^{1}Email author

**2**:23

https://doi.org/10.1186/s40635-014-0023-0

© Wang et al.; licensee Springer. 2014

**Received: **4 April 2014

**Accepted: **21 July 2014

**Published: **20 September 2014

## Abstract

### Background

Computer simulation models could play a key role in developing novel therapeutic strategies for patients with chronic obstructive pulmonary disease (COPD) if they can be shown to accurately represent the pathophysiological characteristics of individual patients.

### Methods

We evaluated the capability of a computational simulator to reproduce the heterogeneous effects of COPD on alveolar mechanics as captured in a number of different patient datasets.

### Results

Our results show that accurately representing the pathophysiology of individual COPD patients necessitates the use of simulation models with large numbers (up to 200) of compartments for gas exchange. The tuning of such complex simulation models ‘by hand’ to match patient data is not feasible, and thus we present an automated approach based on the use of global optimization algorithms and high-performance computing. Using this approach, we are able to achieve extremely close matches between the simulator and a range of patient data including PaO_{2}, PaCO_{2}, pulmonary deadspace fraction, pulmonary shunt fraction, and ventilation/perfusion (V̇/Q) curves. Using the simulator, we computed combinations of ventilator settings that optimally manage the trade-off between ensuring adequate gas exchange and minimizing the risk of ventilator-associated lung injury for an individual COPD patient.

### Conclusions

Our results significantly strengthen the credibility of computer simulation models as research tools for the development of novel management protocols in COPD and other pulmonary disease states.

## Keywords

## Background

Chronic obstructive pulmonary disease (COPD) is a leading cause of mortality and disability internationally and is predicted by 2020 to be the third most likely cause of death [1],[2]. COPD is a progressive disease and is associated with increasing frequency and severity of exacerbations. Mechanical ventilation (MV), either invasive or non-invasive, may be a life-saving measure in managing respiratory failure due to an acute exacerbation of COPD [3]. However, mechanical ventilation is also associated with significant rates of morbidity and mortality. An improved understanding of the underlying pathophysiologic mechanisms of COPD is essential for the development of more effective and more individualized ventilation strategies for COPD patients.

Computer simulators that can accurately represent the particular disease state of an individual COPD patient could be an extremely valuable research tool for investigating the respiratory pathophysiology of COPD and predicting the effects of specific MV settings on the patient. Many researchers have worked on the development of physiological simulators, and various types of mathematical models have been proposed in the literature (e.g., [4]–[11]). However, these models generally employed only a very small number of compartments for gas exchange - in this paper, we show that such model cannot provide an accurate representation of the particular heterogeneous effects of COPD on alveolar mechanics. Also, most previous efforts to match physiological simulators to specific disease states have relied on manually manipulating the parameters in the simulator. Loeppky and co-workers [7] used a two-compartment model to match the COPD patient data generated from the multiple inert gas elimination technique (MIGET). Besides the use of a rather simplistic model, the study suffered from the use of only two parameters to represent V̇/Q mismatching; this is a significant limitation, since gas exchange in COPD patients is often characterized by different V̇/Q patterns with two or three modes [12]. In more sophisticated multi-compartmental models, matching has been achieved with some success by adjusting the resistance for 100 compartments manually [13], a challenging and time-consuming task. Indeed, it is obvious that manual matching is only practical for relatively simple models with a small set of adjustable parameters and a limited set of patient data for matching.

A small number of previous studies have investigated the use of numerical optimisation approaches for the model matching task; a tidally breathing model has been matched to MIGET measurements from emphysema and embolism patients using the interior-reflective Newton algorithm [14], and two and three compartmental models were matched to data from intensive care patients with acute lung injury (ALI) using Brent's method [15]. These studies again used rather simple models with only a few compartments, and both studies employed very simple optimization algorithms, for which the quality of the matching achieved depended completely on the initial values chosen for the optimisation parameters. Since there is very little information available about how to choose these initial estimates, the likelihood of such approaches finding the best possible match to patient data is very small. Moreover, due to the simplified nature of the models used, continuous V̇/Q distributions could not be produced, and thus, the model outputs can only be compared to patient data at two or three isolated points on the V̇/Q curve.

In this paper, we investigate the minimum level of complexity required in a computer simulation model in order to provide a representation of COPD pathophysiology that is accurate enough to allow the simulator to be used for studies on the design of novel therapeutic strategies for individual patients. In order to address this question, we employ a sophisticated simulation model of lung physiology that incorporates tidal ventilation, pulsatile pulmonary blood flow, hypoxic pulmonary vasoconstriction (HPV), a realistic and validated oxygen-hemoglobin model, and up to 200 individually configurable alveolar compartments. Our simulator has been developed over the past decade and has been used and validated in a number of previous studies [16],[17],[13]. In the present study, we couple this simulator to software implementing a global optimization algorithm [18], based on evolutionary principles (a *genetic algorithm*) which allows large numbers of model parameters to be simultaneously optimized in order to match the model outputs to detailed COPD patient data. An attractive feature of our approach from the point of view of clinical researchers and practitioners is that it can be highly automated, so that the user does not need to have detailed knowledge of the underlying algorithms but can use the software to fit models of varying complexity to different patient data sets. To illustrate how the proposed simulator could be used to investigate questions of clinical relevance, we present the results of an investigation into computing ventilator settings that optimally manage the trade-off between ensuring adequate gas exchange and minimizing the risk of ventilator-associated lung injury for an individual COPD patient.

## Methods

### The computational simulator

The simulation used in this study is a multi-compartmental computational model that uses an iterative technique to simulate integrated respiratory and cardiovascular pathophysiological scenarios [17],[19],[20]. A detailed description of the principles and mathematical equations underlying the computational model implemented in our simulator is provided in Additional file 1. In contrast to previous models of COPD pathophysiology that included only two or three alveolar compartments, our model allows the user to define the number of compartments (each with its own individual mechanical characteristics) to be implemented in the simulation. This allowed us to investigate in detail the relationship between the number of compartments in the model and its ability to match individual patient data. Each *i* th alveolar compartment has a unique and configurable bronchiolar resistance *R*_{
B,i
} [cmH_{2}O∙s/l], pulmonary vascular resistance *R*_{
V,i
}[dyn∙s/cm^{5}], stiffness index *S*_{
i
} [cmH_{2}O/ml^{2}], and extrinsic pressure *P*_{ext,i} [cmH_{2}O](giving 4×N adjustable parameters in total, where *N* is the number of compartments). The ability to adjust these parameters individually across up to 200 alveolar compartments allows the model to recreate the heterogeneous effects of COPD on the overall physiology of the lung. The model also includes specific equations to represent the effects of alveolar collapse, threshold opening pressure, alveolar stiffening, and airway obstruction. The net effect of these components of the simulation is that the defining, clinical features of COPD may be observed in the model: alveolar gas trapping (with intrinsic positive end-expiratory pressure (PEEP)), collapse-reopening of alveoli (with gradual reabsorption of trapped gas if reopening does not occur), limitation of expiratory flow, and increased functional residual capacity - see Additional file 1 for further details.

### Patient data

_{2}, PCO

_{2}, deadspace, shunt, mean and standard deviation of V̇ and Q, and ventilation-perfusion distribution across a number of ranges (see Table 2). The second set of patient data is from [12]. The 64-year-old patient is reported to be in a stable condition, and relevant model and ventilator configuration parameters are shown in Table 3. For this case, we attempt to match our model directly to patient V̇/Q curves generated via MIGET measurements.

**Model configuration for the first patient dataset**

Parameter | Value |
---|---|

Respiratory frequency [bpm] | 13 |

Tidal volume [ml] | 590 |

FIO | 0.4 |

Inspiratory flow pattern | Constant flow |

Cardiac output [l/min] | 5.0 |

PEEP [cmH | 0 |

IE | 1:3 |

RQ | 0.8 |

**Matched parameter values and the reference data for the first patient dataset**

No. | Parameter | Data | Model outputs (N = number of compartments) | |||
---|---|---|---|---|---|---|

N = 10 | N = 25 | N = 50 | N = 100 | |||

1 | PaO | 125.2 | 165.29 | 143.18 | 133.73 | 127.92 |

2 | PaCO | 46 | 43.08 | 44.73 | 43.62 | 45.76 |

3 | Dead space fraction | 64.9 | 61.24 | 62.76 | 61.28 | 65.28 |

4 | Shunt fraction | 6.8 | 12.9 | 9.02 | 8.07 | 7.2 |

5 | mean_ V̇ | 0.99 | 1 | 1 | 1 | 1 |

6 | mean_Q | 0.27 | 0.2 | 0.2 | 0.25 | 0.2 |

7 | sd_V̇ | 1 | 0.92 | 1.15 | 0.92 | 1.15 |

8 | sd_Q | 1.34 | 1.38 | 1.38 | 1.38 | 1.38 |

9 | 0.1 < V̇/Q < 1, V̇ | 19.9 | 20.43 | 20.06 | 22.68 | 19.94 |

10 | 1 < V̇/Q < 10, V̇ | 14.3 | 19.69 | 16.57 | 16.47 | 14.48 |

11 | 0.01 < V̇/Q < 0.1, P | 15.8 | 22.63 | 18.86 | 14.64 | 16.16 |

12 | 0.1< V̇/Q < 1, P | 62.1 | 55.21 | 58.12 | 65.57 | 64.92 |

13 | 1 < V̇/Q < 10, P | 10.9 | 11.05 | 11.58 | 10.63 | 10.56 |

Total matching error | 1.33 | 0.46 | 0.11 | 0.10 | ||

Simulation time [h] | 11 | 32 | 41 | 67 |

**Model configuration for the second patient dataset**

Parameter | Value |
---|---|

Respiratory frequency [bpm] | 16 |

Tidal volume [ml] | 410 |

FiO | 0.21 |

Inspiratory flow pattern | Constant flow |

Cardiac output [l/min] | 3.4 |

PEEP [cmH | 0 |

IE | 1:3 |

RQ | 0.8 |

### Automated matching to patient data

Exacerbations of COPD are frequently associated with deterioration in gas exchange and associated hypoxemia. Unsurprisingly, increased inequality in V̇/Q relationships appears to be the major determinant of these changes [1]. Therefore, a key requirement for the simulation of COPD pathophysiology is the ability to accurately match the V̇/Q distributions seen in patient data. In our simulator, the V̇/Q distribution can be manipulated by adjusting the bronchiolar resistance and pulmonary vascular resistance, stiffness, and extrinsic pressure for each compartment in the model. For example, by increasing vascular resistance in a region of the lung, an area of relative dead space can be created. While all parameters can be manually adjusted, this becomes impractical as the number of compartments *N* in the model increases, since the total number of parameters is 4×N.

_{2}, PaCO

_{2}, and mean and standard deviation of ventilation and perfusion. For the second case, the whole V̇/Q curve is considered (i.e., every point on the curve is to be matched). The matching error can then be defined based on these data, which is given as:

where *E*_{
T
} is the total residual error representing the matching accuracy,${E}_{i}=\frac{{x}_{i}-{x}_{\mathit{id}}}{{x}_{\mathit{id}}}$is the error for parameter *i*, *x*_{
i
} is the model output value for parameter *i*, and *x*_{
id
} is the value of the data for parameter *i*.

*E*

_{ T }, i.e., minimize the difference between the model outputs and the data. The procedure is illustrated in Figure 1 - in each iteration, a set of parameter combinations are sent to the simulator, and the outputs from the simulator are evaluated by the optimization algorithm which then generates the updated parameter values for the next iteration until the termination criterion is reached (i.e., the condition to get a best matching is found).

**Model parameters – nominal values and allowable ranges (for** N **= 100)**

Parameters | Nominal value | Variation ranges |
---|---|---|

Bronchiolar resistance | 600 | [300, 3 × 10 |

Vascular resistance | 1.6 × 10 | [8 × 10 |

Stiffness coefficient | 0.05 | [0.025, 0.15] |

Extrinsic pressure | 28.8 | [−20, 28.8] |

the resistance (*R*) in a single tube is directly proportional to the length (*L*) of the tube and the viscosity (*η*) and inversely proportional to the fourth power of the radius (*r*^{
4
}). To achieve the required precision in matching the extreme ranges of V̇/Q distributions for the second dataset, we have investigated the effect of increasing the upper bound on compartmental resistances from 500 to 2,000 times their nominal values, i.e., the radius of the airways can be reduced by as much as 85%. To ensure that this does not result in model configurations that produce unrealistic values for total lung resistance, we incorporated a constraint function into the optimization algorithm that limits the increase in the total lung inflow resistance to ten times its basic (healthy) value, as expected in patients with COPD [22].

Previous attempts to use optimization for matching models of pulmonary disease states to patient data have used rather simple algorithms that require good initial ‘guesses’ for the parameters in order to be effective. In the case of COPD, however, current understanding of the associated pathophysiology provides little guidance into how to choose initial values for the model parameters across large numbers of alveolar compartments. In this study, we therefore employed an advanced global optimisation algorithm known as a genetic algorithm. This general purpose stochastic search and optimization procedure, based on genetic and evolutionary principles [23], has been shown to have a much higher chance of finding optimal solutions for difficult problems with large numbers of variables. Full details of the particular optimization algorithm used in this study and how it was implemented with the model are provided in Additional file 1.

### Using the simulator for clinical investigations

To illustrate how the simulator can be used for clinical investigations, we consider the problem of identifying ventilator settings that minimize the risk of ventilator-associated lung injury (VALI). We consider the following five key ventilator settings as variable parameters that may be adjusted to optimize the trade-off between effective gas exchange and minimizing the risk of VALI: (1) tidal volume (Vtidal, [ml]) - the volume of air traveling in or out of the patient's lungs during every breath; (2) ventilation rate (VentRate, [breaths/min]) - the number of breaths per minute; (3) duty cycle (I:E) - the ratio of inspiratory time to total ventilatory cycle duration; (4) PEEP, [cmH_{2}O] - the positive pressure in the lungs at the end of exhalation; and (5) fraction of inspired oxygen (FIO_{2}) - the fraction of oxygen constituting the inhaled volume of gas as provided by the mechanical ventilator.

_{2}O, and FIO

_{2}is bounded within 0.21 to 1. A summary is provided in Table 5.

**MV setting parameter variation bounds and desired model outputs**

Variation ranges | ||
---|---|---|

Acceptable values | Desired | |

MV setting parameters | ||

Vt [ml] | [390, 650] | |

VentRate [bpm] | [9,16] | |

I:E | [0.25, 0.5] | |

PEEP [cmH | [0, 5] | |

FiO | [0.21, 1] | |

Model outputs | ||

PO | >8 | 12 |

PCO | >4, <8 | 5.3 |

Palv [kPa] | <4 | - |

Three key physiological indicators are also defined. To monitor effective arterial oxygenation, partial pressure of oxygen, PaO_{2}, needs to be considered. In order to maintain effective arterial oxygenation, PaO_{2} is constrained to be higher than 8 kPa, with a desired value of 12 kPa. Arterial partial pressure of carbon dioxide, PaCO_{2} is another key indicator of alveolar ventilation that also indirectly reflects acid-base balance. PaCO_{2} is constrained to be between 4 and 8 kPa with a desired value of 5.3 kPa. The risk of barotrauma is proportional to the peak alveolar pressure, (Palv, kPa above atmospheric pressure), and Palv is limited to 4 kPa, where Palv is calculated as the average of the peak pressure in the most highly pressurized 25% of all alveoli.

Large values of *J*_{1} will be produced by ventilator settings that provide poor gas exchange, while large values of *J*_{2} will be produced by combinations of ventilator settings that cause high peak alveolar pressures (and hence increase the risk of VALI). By requiring both *J*_{1} and *J*_{2} to be minimized simultaneously, we can search for combinations of ventilator settings that optimally manage the trade-off between effective gas exchange and minimizing the risk of VALI. w_{1},w_{2}, and w_{3} are weighting functions that are used in the optimization process to ensure that equal priority is given to each of the different objectives. Since we are trying to minimize two objectives *J*_{1} and *J*_{2} at the same time, a multi-objective optimization algorithm called non-dominated sorting genetic algorithm II (NSGA-II) was used here [26].

## Results

### Matching results for the first dataset

For the first dataset, the ventilation-perfusion distribution is characterized by several segments based on the V̇/Q ratio, which includes the amount of ventilation in the ranges 0.1< V̇/Q < 1 and 1< V̇/Q < 10, and the amount of perfusion in the ranges 0.01< V̇/Q < 0.1, 0.1< V̇/Q < 1, and 1< V̇/Q < 10 together with the shunt and dead space. Other data considered are the PO_{2}, PCO_{2}, mean V̇/Q ratio of the pulmonary blood flow, and ventilation distribution as well as standard deviations (dispersion) of pulmonary blood flow and ventilation. The simulation was performed separately for four cases with the total number of compartments in the model being varied between 10, 25, 50, and 100 to represent different levels of model complexity.

### Matching results for the second dataset

### Computation of optimal ventilation strategies using the simulator

_{2}(from 6.1 to 6.3 kPa), while the value of PaO

_{2}(12.2 kPa) is now very close to its desired value of 12 kPa. This ‘rebalancing’ of the ventilator settings in favor of a more lung protective strategy has been achieved principally by reducing tidal volume from 590 to 506 ml, while simultaneously increasing the ventilation rate from 13 to 15 bpm. Changes in the other three ventilator settings are more modest, indicating that they have a relatively smaller influence on the chosen physiological indicators for this patient. Moving along the curve shown in Figure 4, different combinations of ventilator settings are computed which place more emphasis on one objective or the other. Point No. 2, for example, corresponds to a maximally protective strategy, where Palv has been further reduced to 2.8 kPa, at the cost of increasing PaCO

_{2}to 7.4 kPa and reducing PaO

_{2}to 11.9 (note that both PaCO

_{2}and PaO

_{2}are still within their specified limits).

**Optimal MV settings and model outputs compared with first patient dataset**

Data | Optimal | |
---|---|---|

MV setting parameters | ||

Vt [ml] | 590 | 506 |

VentRate [bpm] | 13 | 15 |

I:E | 0.33 | 0.32 |

PEEP [cmH | 0 | 1 |

FiO | 0.4 | 0.3 |

Model outputs | ||

PaO | 17 | 12.2 |

PaCO | 6.1 | 6.3 |

Palv [kPa] | 5.1 | 3.8 |

## Discussion and conclusions

The high mortality and morbidity associated with COPD will make the condition a critical health burden for the future, particularly if current trends are allowed to continue [27],[28]. The main treatment option for critical exacerbations of COPD remains mechanical ventilation (non-invasive and invasive) and for long-term management, oxygen therapy [25]. Prognosis of COPD is worsened with underlying respiratory failures, adding greatly to the cost of treating the disease [29],[30]. Research into the condition using animal models of lung injury and *in vitro* experiments have failed to show a clear way forward, and large-scale clinical trials to monitor long-term effects in this fragile population are not possible [31],[6].

Computer simulations [6],[32] offer a potentially powerful alternative platform for research into the development of new treatment strategies, but no previous studies have been able to demonstrate adequate matching to the particular disease characteristics of COPD. In this study, we have described a computer simulation model that can accurately simulate an individual COPD patient receiving MV. Our study has focused on single patient subjects described in the literature, and the two studies that have been selected represent the spectrum of COPD cases that are found in clinical practice. Our results clearly show the importance of employing mathematical models of sufficient complexity to allow an accurate representation of disease states in individual patients. In particular, the heterogeneous effects of COPD on lung alveoli in different regions of the lung make it essential that large numbers of compartments are included in the model - for all datasets considered, matching between the simulation outputs and patient data was significantly improved as the number of alveolar compartments in the model was increased.

The need for a large number of compartments with heterogeneous dynamics creates an additional challenge - how to choose appropriate values for the large number of corresponding model variables. Indeed, a second major conclusion of our study is that ‘manual’ tuning of such parameters is simply not feasible for models of the complexity required. Instead, we have shown how global optimization algorithms, implemented using sophisticated parallel processing protocols, can allow optimal values of very large numbers of parameters to be identified. A significant advantage of this approach is that it can be largely automated, with the clinical researcher needing only to input the patient data and the simulation platform taking care of all other aspects of the model-matching process.

Due to the nature of COPD, multiple comorbidities are associated with the symptoms and development of COPD including inflammation and heart and kidney dysfunctions. This presents a unique challenge to the clinician in terms of maintaining the correct balance between raising pressures and delivering oxygen, while keeping hypercapnea and hypoxia to a minimum [33],[34]. The clinician needs to monitor multiple aspects of physiology simultaneously, and many adverse effects are not always immediately clear. Currently, MV is applied in a rather generic manner; however, the clinicians have recognized the need for accounting for inter-patient variation [35],[36]. VALI due to overstretching of lung parenchyma can be minimized by applying lower tidal volumes [37], but this needs to be achieved without allowing PaCO_{2} and pH to rise to dangerous levels (permitted hypercapnea). Sophisticated computer simulation platforms that can accurately represent individual COPD patient characteristics could provide a powerful experimental tool for the development of improved treatment strategies for COPD that address many of the above issues.

## Additional file

## Declarations

### Acknowledgements

This work was supported from the UK Engineering and Physical Sciences Research Council (EP/F057016/2, EP/F057059/1, EP/1036680/2) and Medical Research Council (G1002017, MR/K019783).

## Authors’ Affiliations

## References

- Kent BD, Mitchell PD, McNicholas WT: Hypoxemia in patients with COPD: cause, effects, and disease progression.
*Int J Chron Obstruct Pulmon Dis*2011, 6: 199.PubMedGoogle Scholar - Murray CJ, Lopez AD: Alternative projections of mortality and disability by cause 1990–2020: Global Burden of Disease Study.
*Lancet*1997, 349(9064):1498–1504. 10.1016/S0140-6736(96)07492-2PubMedView ArticleGoogle Scholar - Reddy RM, Guntupalli KK: Review of ventilatory techniques to optimize mechanical ventilation in acute exacerbation of chronic obstructive pulmonary disease.
*Int J Chron Obstruct Pulmon Dis*2007, 2(4):441.PubMedGoogle Scholar - Gutierrez G: A mathematical model of tissue-blood carbon dioxide exchange during hypoxia.
*Am J Respir Crit Care Med*2004, 169(4):525–533. 10.1164/rccm.200305-702OCPubMedView ArticleGoogle Scholar - Hickling KG: Best compliance during a decremental, but not incremental, positive end-expiratory pressure trial is related to open-lung positive end-expiratory pressure: a mathematical model of acute respiratory distress syndrome lungs.
*Am J Respir Crit Care Med*2001, 163(1):69–78. 10.1164/ajrccm.163.1.9905084PubMedView ArticleGoogle Scholar - Huo B, Fu R: Recent advances in theoretical models of respiratory mechanics.
*Acta Mechanica Sinica*2012, 28(1):1–7. 10.1007/s10409-012-0024-8View ArticleGoogle Scholar - Loeppky JA, Caprihan A, Altobelli SA, Icenogle MV, Scotto P, Vidal Melo MF: Validation of a two-compartment model of ventilation/perfusion distribution.
*Respir Physiol Neurobiol*2006, 151(1):74–92. 10.1016/j.resp.2005.06.002PubMedView ArticleGoogle Scholar - Martin S, Similowski T, Straus C, Maury B (2008) Impact of respiratory mechanics model parameters on gas exchange efficiency. In: ESAIM: Proceedings. Volume 23. EDP Sciences, pp 30–47Google Scholar
- Steimle KL, Mogensen ML, Karbing DS, Bernardino Dela Serna J, Andreassen S: A model of ventilation of the healthy human lung.
*Comput Methods Programs Biomed*2011, 101(2):144–155. 10.1016/j.cmpb.2010.06.017PubMedView ArticleGoogle Scholar - Wilson A, Murphy C, Brook B, Breen D, Miles A, Tilley D: A computer model of the artificially ventilated human respiratory system in adult intensive care.
*Med Eng Phys*2009, 31(9):1118–1133. 10.1016/j.medengphy.2009.07.009PubMedView ArticleGoogle Scholar - Yem JS, Tang Y, Turner MJ, Baker AB: Sources of error in noninvasive pulmonary blood flow measurements by partial rebreathing: a computer model study.
*Anesthesiology*2003, 98(4):881–887. 10.1097/00000542-200304000-00014PubMedView ArticleGoogle Scholar - Wagner P, Dantzker D, Dueck R, Clausen J, West J: Ventilation-perfusion inequality in chronic obstructive pulmonary disease.
*J Clin Invest*1977, 59(2):203. 10.1172/JCI108630PubMedView ArticleGoogle Scholar - Kathirgamanathan A, McCahon R, Hardman J: Indices of pulmonary oxygenation in pathological lung states: an investigation using high-fidelity, computational modelling.
*Br J Anaesth*2009, 103(2):291–297. 10.1093/bja/aep140PubMedView ArticleGoogle Scholar - Yem J, Turner M, Baker A, Young I, Crawford A: A tidally breathing model of ventilation, perfusion and volume in normal and diseased lungs.
*Br J Anaesth*2006, 97(5):718–731. 10.1093/bja/ael216PubMedView ArticleGoogle Scholar - Karbing DS, Kjærgaard S, Andreassen S, Espersen K, Rees SE: Minimal model quantification of pulmonary gas exchange in intensive care patients.
*Med Eng Phys*2011, 33(2):240–248. 10.1016/j.medengphy.2010.10.007PubMedView ArticleGoogle Scholar - Das A, Menon PP, Hardman JG, Bates DG: Optimization of mechanical ventilator settings for pulmonary disease states.
*IEEE Trans Biomed Eng*2013, 60(6):1599–1607. 10.1109/TBME.2013.2239645PubMedView ArticleGoogle Scholar - Hardman J, Bedforth N, Ahmed A, Mahajan R, Aitkenhead A: A physiology simulator: validation of its respiratory components and its ability to predict the patient's response to changes in mechanical ventilation.
*Br J Anaesth*1998, 81(3):327–332. 10.1093/bja/81.3.327PubMedView ArticleGoogle Scholar - Wang W, Menon P, Bates D, Bennani S: Robustness analysis of attitude and orbit control systems for flexible satellites.
*Control Theory Appl IET*2010, 4(12):2958–2970. 10.1049/iet-cta.2009.0531View ArticleGoogle Scholar - Hardman J, Wills J: The development of hypoxaemia during apnoea in children: a computational modelling investigation.
*Br J Anaesth*2006, 97(4):564–570. 10.1093/bja/ael178PubMedView ArticleGoogle Scholar - McCahon R, Columb M, Mahajan R, Hardman J: Validation and application of a high-fidelity, computational model of acute respiratory distress syndrome to the examination of the indices of oxygenation at constant lung-state.
*Br J Anaesth*2008, 101(3):358–365. 10.1093/bja/aen181PubMedView ArticleGoogle Scholar - Jolliet P, Watremez C, Roeseler J, Ngengiyumva J, De Kock M, Clerbaux T, Tassaux D, Reynaert M, Detry B, Liistro G: Comparative effects of helium-oxygen and external positive end-expiratory pressure on respiratory mechanics, gas exchange, and ventilation-perfusion relationships in mechanically ventilated patients with chronic obstructive pulmonary disease.
*Intensive Care Med*2003, 29(9):1442–1450. 10.1007/s00134-003-1864-2PubMedView ArticleGoogle Scholar - Loring SH, Garcia-Jacques M, Malhotra A: Pulmonary characteristics in COPD and mechanisms of increased work of breathing.
*J Appl Physiol*2009, 107(1):309–314. 10.1152/japplphysiol.00008.2009PubMedView ArticleGoogle Scholar - Glodberg DE:
*Genetic algorithms in search, optimization, and machine learning*. Addion wesley, Boston; 1989.Google Scholar - Esteban A, Anzueto A, Alia I, Gordo F, Apezteguia C, Palizas F, Cide D, Goldwaser R, Soto L, Bugedo G: How is mechanical ventilation employed in the intensive care unit? An international utilization review.
*Am J Respir Crit Care Med*2000, 161(5):1450–1458. 10.1164/ajrccm.161.5.9902018PubMedView ArticleGoogle Scholar - Esteban A, Anzueto A, Frutos F, Alia I, Brochard L, Stewart TE, Benito S, Epstein SK, Apezteguia C, Nightingale P, Arroliga AC, Tobin MJ: Characteristics and outcomes in adult patients receiving mechanical ventilation: a 28-day international study.
*JAMA*2002, 287(3):345–355. 10.1001/jama.287.3.345PubMedView ArticleGoogle Scholar - Deb K, Pratap A, Agarwal S, Meyarivan T: A fast and elitist multiobjective genetic algorithm: NSGA-II.
*Evol Comput IEEE Trans*2002, 6(2):182–197. 10.1109/4235.996017View ArticleGoogle Scholar - Mannino DM, Buist AS: Global burden of COPD: risk factors, prevalence, and future trends.
*Lancet*2007, 370(9589):765–773. 10.1016/S0140-6736(07)61380-4PubMedView ArticleGoogle Scholar - Rabe KF, Hurd S, Anzueto A, Barnes PJ, Buist SA, Calverley P, Fukuchi Y, Jenkins C, Rodriguez-Roisin R, Van Weel C, Zielinski J: Global strategy for the diagnosis, management, and prevention of chronic obstructive pulmonary disease: GOLD executive summary.
*Am J Respir Crit Care Med*2007, 176(6):532–555. 10.1164/rccm.200703-456SOPubMedView ArticleGoogle Scholar - Breen D, Churches T, Hawker F, Torzillo PJ: Acute respiratory failure secondary to chronic obstructive pulmonary disease treated in the intensive care unit: a long term follow up study.
*Thorax*2002, 57(1):29–33. 10.1136/thorax.57.1.29PubMedView ArticleGoogle Scholar - Matic I, Danic D, Majeric-Kogler V, Jurjevic M, Mirkovic I, Mrzljak Vucinic N: Chronic obstructive pulmonary disease and weaning of difficult-to-wean patients from mechanical ventilation: randomized prospective study.
*Croat Med J*2007, 48(1):51–58.PubMedGoogle Scholar - Budweiser S, Jörres RA, Pfeifer M: Treatment of respiratory failure in COPD.
*Int J Chron Obstruct Pulmon Dis*2008, 3(4):605.PubMedGoogle Scholar - Tawhai MH, Clark AR, Burrowes KS: Computational models of the pulmonary circulation: insights and the move towards clinically directed studies.
*Pulm Circ*2011, 1(2):224–238. 10.4103/2045-8932.83452PubMedView ArticleGoogle Scholar - Ferrer M, Esquinas A, Leon M, Gonzalez G, Alarcon A, Torres A: Noninvasive ventilation in severe hypoxemic respiratory failure: a randomized clinical trial.
*Am J Respir Crit Care Med*2003, 168(12):1438–1444. 10.1164/rccm.200301-072OCPubMedView ArticleGoogle Scholar - Rodriguez-Roisin R, Drakulovic M, Rodriguez DA, Roca J, Barbera JA, Wagner PD: Ventilation-perfusion imbalance and chronic obstructive pulmonary disease staging severity.
*J Appl Physiol*2009, 106(6):1902–1908. 10.1152/japplphysiol.00085.2009PubMedView ArticleGoogle Scholar - Marini JJ: Lower tidal volumes for everyone: principle or prescription?
*Intensive Care Med*2012, 39(1):3–5. 10.1007/s00134-012-2754-2PubMedView ArticleGoogle Scholar - Fernando F-V, Ferguson ND, Andrés E: Mechanical ventilation: quo vadis?
*Intensive Care Med*2009, 35(5):775–778. 10.1007/s00134-009-1450-3View ArticleGoogle Scholar - Tremblay LN, Slutsky AS: Ventilator-induced lung injury: from the bench to the bedside.
*Intensive Care Med*2006, 32(1):24–33. 10.1007/s00134-005-2817-8PubMedView ArticleGoogle Scholar

## Copyright

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.