Discussion
This study showed the reliability of Simulresp when predicting blood gas values of healthy child over 8 years old, confirming O. Flechelles et al. data (2,4). The inconsistencies of Simulresp’s prediction regarding under 8 years old and/or ventilator supported subjects might be due to the fact that 1977 CJ Dickinson’s model didn’t considered physiologic and physio-pathologic specificities of these conditions (8). Children respiratory system differs from that of the adult, anatomically and mechanically, mainly during the first 6 to 8 years. Less than 20% of “adult” cells are present at birth. In the first years of life, lung growth occurs by adding or creating new alveoli (18). This alveolarization is accompanied by an increase in the capacity of the lung to perform gas exchanges (18). Alveolarization is considered complete around the age of 6 to 8 years. Thereafter, the alveolar surface will grow due to the growth of the child. This growth of the alveolar surface is related to lung growth but is not associated with an increase in the number of alveoli (18). In addition, from birth to the age of 2 to 3 years, the shape of the rib cage will evolve from a circular shape to a more oval or rectangular shape (19). This shape causes a lower mechanical efficiency of the thoracic cavity compared to adults (20), especially since it is associated with a flattened less efficient diaphragm. Besides, the discrepancies between a high compliance rib cage and a low compliance lung will altered the residual functional capacity. Thus, the child must develop several dynamic compensation mechanisms to maintain its reserves in oxygen and avoid atelectasis. Mechanical ventilation substitutes positive pressure insufflation by an external machine induced to negative pressure insufflation by diaphragmatic contraction. The installation of a ventilatory support requires four elements: a suitable interface between the respirator and the patient, an energy source that operates the machine, an insufflation whose magnitude and rhythm will be regulated or controlled and a system to monitor the performance of the respirator and the condition of the patient. These four elements are associated with mechanic, physic and physiologic constraints that can be difficult to predict, record or standardize in the form of mathematical equations. In addition, if these four elements are mandatory, they are not sufficient to ensure the effectiveness of the respiratory support and other elements such as the adaptation of the parameters, the synchrony between the patient and the machine, the need for sedation or the type of respirator are to be considered and added as constraints and limits to the success of the modeling.
The goal of validation is giving confidence to users on simulator results because the future use of a simulator depends on it. The quality assessment procedures of a simulator are essential to ensure its validity and usability (5). The validity of a simulator will depend directly on the objective for which it was developed (2,13,21,22). Targets, safety ranges, gap between simulated values and expected values can differ depending on simulator goals. For example, when the simulator is used for teaching, an inaccurate output can likely be tolerated if the evolution or prediction approaches a physiologically normal value, and both repeatability and reproducibility are satisfactory. On the other hand, use in care cannot be conceived with a simulator that would not be accurate.
Several cardio-respiratory simulators are described in the literature (2,22,23), but there are few descriptions of the validation process. Most of the publications on the subject focus on the description of the simulator’s performance rather than on the actual technical report of the process applied to guarantee the quality of the simulator and its prediction (24). Some teams took the trouble to evaluate the performance of their simulator by comparing the simulated data with data observed in real patients in mechanical ventilation (25,26). However, the content of these articles remains focused on the description of the purpose of the simulator and how it is developed. The idea of carrying out and presenting a complete validation process, aimed at judging the ability of the simulator to provide a fair and reliable prediction in time and situations, only rarely seems to be part of the research protocol (5,13,27).
The strength of this work lies in the large number of tests performed. In addition, we evaluated each quality component of the simulator: accuracy, repeatability, reproducibility and robustness. We have thus been able to highlight the limitations of the simulator, particularly as regards the prediction of the patient under 8 years of age, the patient who is ill and the patient in mechanical ventilation. This work has major limitations. The use of graphical representations to judge the accuracy of the simulator, although described in the literature (5,13) may seems trivial and in all cases particularly subjective and unreliable. In addition, the question arises as to the choice of statistical tests carried out. While it is certain that the measurement of Pearson or Spearman correlation coefficients is insufficient to judge the accuracy and concordance between simulated and reference values, the relevance of the other statistical methods applied remains equally questionable (15,28). Nonetheless, the performed tests were based on previously described simulator evaluations (25,29). Finally, the question of external applicability arises. The results we obtained apply only to SimulResp and cannot be extrapolated to other simulators, even if they were developed using the Dickinson model.
This work allowed us to better define the next steps in the development of SimulResp. In this actual form, SimulResp is limited by the limits of the model it is built on. For this reason, we are currently working on the content of the algorithm in order to improve SimulResp. we need to study and modified the formula within the model in order to make it accurate when simulating blood gas value of under 8 years old patients. To do so, we have gathered a high-resolution database (30,31), on which we intend to apply several machine learning approaches (32,33) Besides, we have begun to develop the method that will allow us to calibrate the simulator to be reliable in several respiratory physio-pathological situations (normal compliance (> 2 ml/cmH2O/kg), abnormal compliance, increased resistance) and hemodynamic (shock states) in both spontaneous and invasive ventilation (2,30). SimulResp, once completely validated, will be integrated in future clinical decision support systems and will collect data from real patient and then simulate breathing pattern with accelerated time. Resulting simulated blood gases will be presented to physician whom could test different ventilator settings to determine the best one and adjust the patient’s therapy in an individualized but still protocolized care.