Managing mechanical ventilation is a core component of an intensivist’s daily work, and optimizing ventilator management continues to be a major focus of pediatric critical care research (1). Physiologic models of human respiration help illustrate core principles and can guide clinicians in titrating ventilator parameters (2). Such physiologic models, when implemented in software with graphical user interfaces, can give rise to virtual patient simulators, allowing clinicians to test the impact of potential ventilator changes, to suggest ventilator changes, or even to make ventilator changes autonomously (3). However, accurate implementation of such models, particularly for children, has proven challenging (2). Are the models reliable? Do they respond to inputs the way real patients do?

In this issue of Pediatric Critical Care Medicine, Pelletier et al (4) highlight these challenges by testing how well one classic physiologic equation, described by Wexler and Lok (5) and originally evaluated in 50 adult patients, performed in a large pediatric cohort. The equation demonstrates that, assuming constant carbon dioxide production (V̇co2), unchanged ratio of dead space to tidal volume (Vd/Vt), and no spontaneous breathing, Paco2 varies inversely with minute ventilation. Pelletier et al (4) examined 484 patients who were mechanically ventilated while under neuromuscular blockade. Among 15,121 arterial blood gas (ABG) values, there was a median prediction error of 0.00 mm Hg (interquartile range, –3.07 to 3.00 mm Hg). However, practically speaking, the authors rightly note that only 68% of measured ABG values were within ± 5 mm Hg of the predicted values. Pelletier et al (4) made a pragmatic choice to include ABG values up to 6 hours after ventilator changes. In their particularly ill patient population—with frequent comorbidities (87% of patients), prolonged neuromuscular blockade (median 4 d), and high mortality (12%)—other changes in patient physiology, including changes in V̇co2, bicarbonate clearance, or alveolar dead space, likely also impact ABG values, independent of the product of respiratory rate and tidal volume (6).

Pelletier et al (4) also examined clinicians’ ventilatory rate changes in a subset of ABGs in which adjustment of minute ventilation to maintain an acceptable pH was indicated. Of note, the authors constrained the problem space by excluding scenarios where ventilator parameters other than the set rate (e.g., tidal volume) were adjusted. They compared the clinician’s change in respiratory rate vs. the change recommended by the equation and used the subsequent blood gas to determine whether the clinician or the calculator made the “better” choice. In most scenarios (75% vs. 23%), the authors found that the calculator’s recommendation outperformed the clinician’s judgment. Furthermore, there was a marked difference when comparing ABGs that required escalation vs. weaning of minute ventilation. When the ABG indicated respiratory acidosis requiring increased minute ventilation, clinicians’ rate changes outperformed calculator recommendations for 53% of ABGs. In contrast, when the ABG suggested a wean was indicated, the clinician’s choice was better in only 21% of scenarios. This finding is concordant with prior literature that computerized clinical decision support (CDS) systems for mechanical ventilation management are particularly effective for the weaning phase (7). Prior work suggested that these benefits may be achieved because closed-loop systems, such as those reviewed in (7), can effect ventilator changes more frequently and more consistently than humans. The study by Pelletier et al (4) suggests that an additional factor may be that PICU clinicians tend to wean the ventilator in smaller increments than might be physiologically warranted. A key element of well-implemented CDS systems is users’ understanding of and faith in the algorithm’s recommendations (8). The findings by Pelletier et al (4) could be incorporated into a future CDS system that nudges clinicians toward making larger ventilator weans by showing the mathematical model behind the recommendation and reminding clinicians that in most cases in the study by Pelletier et al (4) the larger wean resulted in a “better” subsequent ABG than the clinicians’ intuition.

By limiting their analysis to patients receiving neuromuscular blockade, Pelletier et al (4) removed the effects of varying breath-to-breath minute ventilation in spontaneously breathing patients. This constraint is necessary to apply the Wexler and Lok (5) equation but reduces applicability at the bedside. Additional limitations of the current work, also appropriately noted by the authors, include the lack of clinical detail about the patient population and, in particular, whether some patients may have had clinical indications for the clinicians’ more conservative weaning choices. For instance, in patients with pulmonary hypertension, where respiratory acidosis may be more poorly tolerated, decreasing the ventilatory rate in smaller increments may be logical. It is important to note that serial ABGs in patients are not statistically independent. Indeed, a future computerized CDS system could use modern machine learning methods like reinforcement learning (9) to adjust the system’s recommendation for subsequent ventilator changes based on whether a Paco2 after an index ventilator change resulted in an ABG that was higher or lower than expected, much as clinicians do at the bedside.

Perhaps the most intriguing application of the work by Pelletier et al (4) may be for informing physiologically based computerized teaching platforms for critical care. Such medical simulators are prevalent (10), and some studies show that they can be effective teaching aids (11). Still, less emphasis has been placed on validating the physiologic accuracy of the models underlying these teaching tools (12). The authors provide a link to a prototype teaching platform that they designed and are testing. A formal evaluation of the educational effectiveness of the platform will be a valuable contribution. Insightfully, on their website, the authors explain the effects of changes to V̇co2 and Vd/Vt on their estimates of Paco2. Thus, the authors turn limitations into teachable moments.

Some may wonder why CDS systems and “serious games” continue to rely on simplified physiologic models when our electronic health records provide a wealth of data on ventilator changes and resulting blood gas values from actual patients. These records, however, provide information only on what clinicians actually did. If clinicians never reduced the respiratory rate by 10, the model will struggle to extrapolate the impact of such a reduction in a future scenario. Physiologic equations like that of Wexler and Lok (5) certainly have their limitations, but the work of Pelletier et al (4) suggests that validating and incorporating such models into our training and CDS systems could start nudging us toward bigger ventilator weans. These bigger weans will generate more data on the effects of such weans for future models and could ultimately shorten the time to liberation from mechanical ventilation.

1. Fernandez A, Modesto V, Rimensberger PC, et al.; Second Pediatric Acute Lung Injury Consensus Conference (PALICC-2) of the Pediatric Acute Lung Injury and Sepsis Investigators (PALISI) Network: Invasive ventilatory support in patients with pediatric acute respiratory distress syndrome: From the Second Pediatric Acute Lung Injury Consensus Conference. Pediatr Crit Care Med. 2023; 24(12 Suppl 2):S61–S75 2. Brossier D, Flechelles O, Sauthier M, et al.: Evaluation of the SIMULRESP: A simulation software of child and teenager cardiorespiratory physiology. Pediatr Pulmonol. 2023; 58:2832–2840 3. Jouvet PA, Payen V, Gauvin F, et al.: Weaning children from mechanical ventilation with a computer-driven protocol: A pilot trial. Intensive Care Med. 2013; 39:919–925 4. Pelletier JH, Rakkar J, Au AK, et al.: Retrospective Validation of a Computerized Physiologic Equation to Predict Minute Ventilation Needs in Critically Ill Children. Pediatr Crit Care Med. 2024; 25:390–395 5. Wexler HR, Lok P: A simple formula for adjusting arterial carbon dioxide tension. Can Anaesth Soc J. 1981; 28:370–372 6. Yehya N, Bhalla AK, Thomas NJ, et al.: Alveolar dead space fraction discriminates mortality in pediatric acute respiratory distress syndrome. Pediatr Crit Care Med. 2016; 17:101–109 7. Rose L, Schultz MJ, Cardwell CR, et al.: Automated versus non-automated weaning for reducing the duration of mechanical ventilation for critically ill adults and children. Cochrane Database Syst Rev. 2014; 2014:CD009235 8. Dziorny AC, Heneghan JA, Bhat MA, et al.; Pediatric Data Science and Analytics (PEDAL) Subgroup of the Pediatric Acute Lung Injury and Sepsis Investigators (PALISI) Network: Clinical decision support in the PICU: Implications for design and evaluation. Pediatr Crit Care Med. 2022; 23:e392–e396 9. Otten M, Jagesar AR, Dam TA, et al.: Does reinforcement learning improve outcomes for critically ill patients? A systematic review and level-of-readiness assessment. Crit Care Med. 2024; 52:e79–e88 10. Olszewski AE, Wolbrink TA: Serious gaming in medical education: A proposed structured framework for game development. Simul Healthc. 2017; 12:240–253 11. Wolbrink TA, van Schaik SM, Turner DA, et al.; Game-based Education in Residency (GamER) Study Group: Online learning and residents’ acquisition of mechanical ventilation knowledge: Sequencing matters. Crit Care Med. 2020; 48:e1–e8 12. Gentry SV, Gauthier A, L’Estrade Ehrstrom B, et al.: Serious gaming and gamification education in health professions: Systematic review. J Med Internet Res. 2019; 21:e12994