Highlight
- Development of an artificial intelligence (AI) algorithm to estimate inspiratory muscle pressure (Pmus) continuously and noninvasively during pressure support ventilation.
- Pmus estimated by AI (Pmus,AI) demonstrated high agreement with the gold standard esophageal manometry (Pmus,es) and comparable accuracy to traditional occlusion maneuvers.
- AI algorithm effectively detected patient-ventilator dyssynchronies including ineffective effort, autotriggering, and reverse triggering with high sensitivity and specificity.
- This approach facilitates real-time monitoring of patient respiratory effort and ventilator synchrony without the need for invasive instrumentation or disruption of ventilation.
Study Background
Mechanical ventilation remains a cornerstone of life support in critically ill patients with respiratory failure. Optimal ventilatory management requires assessment of the patient’s inspiratory muscle effort (quantified by Pmus) to balance ventilator support and prevent diaphragm injury or muscle overuse. Current methods to measure Pmus, such as esophageal manometry, are invasive and technically demanding; other approaches rely on occlusion maneuvers which interrupt ventilation and provide only intermittent data.
Patient-ventilator dyssynchrony is a common and under-recognized complication that can worsen clinical outcomes by increasing work of breathing, prolonging mechanical ventilation, and causing discomfort. Continuous monitoring to detect dyssynchrony and quantify respiratory muscle load remains a clinical unmet need.
This study addresses the need for a real-time, noninvasive, and accurate method to estimate Pmus and automatically identify dyssynchronies during mechanical ventilation using an AI-based algorithm.
Study Design
This prospective diagnostic accuracy study was conducted in two intensive care units (ICUs) affiliated with the University of São Paulo, Brazil. The study population consisted of 48 adult patients receiving pressure support ventilation for respiratory failure management.
No interventions beyond standard care were applied. The primary comparison was between Pmus estimated by the AI algorithm (Pmus,AI) and the gold standard invasive measurement obtained via esophageal manometry (Pmus,es). Additionally, Pmus,AI was compared with values derived from occlusion maneuvers, including the pressure muscle index and occlusion pressure (Pocc).
The AI algorithm also automatically detected specific patient-ventilator dyssynchronies—ineffective efforts, autotriggering, and reverse triggering—with results compared against expert classification as the reference standard.
Endpoints included accuracy metrics (bias, limits of agreement), receiver operating characteristic (ROC) curve analysis for detection of Pmus extremes and dynamic driving pressure, and sensitivity/specificity for dyssynchrony detection.
Key Findings
Data from 4918 respiratory cycles among 48 patients were analyzed. Pmus,es ranged from 1.0 to 28.4 cm H2O, covering a clinically relevant spectrum of inspiratory effort.
The AI algorithm estimated Pmus with a mean bias of 0.9 cm H2O and 95% limits of agreement between -5.1 and 6.9 cm H2O, demonstrating good agreement with invasive esophageal manometry. The algorithm successfully identified both high and low extremes of Pmus, as well as variations in dynamic driving pressure, with area under the ROC curve (AUC) greater than 0.8, indicating strong discriminatory ability.
Comparisons with intermittent occlusion-based methods showed comparable accuracy, underscoring the AI’s potential to replace or supplement traditional techniques.
For detection of patient-ventilator dyssynchronies, the AI system achieved a sensitivity of 86.5% and specificity of 77.4% in identifying ineffective effort, autotriggering, and reverse triggering events compared to expert raters.
These results indicate that the AI algorithm can continuously and noninvasively monitor respiratory muscle effort and automatically detect clinically relevant dyssynchronies during mechanical ventilation without requiring invasive sensors or maneuvers disrupting ventilation.
Expert Commentary
This study represents a significant advancement in critical care respiratory monitoring by leveraging AI to overcome limitations of existing methods. Esophageal manometry, while the gold standard, is invasive and not routinely implemented; intermittent occlusion maneuvers disrupt ventilation and provide only snapshot data. The AI approach detailed here facilitates continuous tracking of patient effort and ventilator harmonization in real time, allowing for timely clinical interventions.
The performance metrics reported are encouraging, though some variability around limits of agreement suggests further refinement may be beneficial. Importantly, the algorithm’s capacity to detect multiple dyssynchrony types can aid clinicians in tailoring ventilator settings to improve patient comfort and outcomes.
Limitations include the single-center design and selection of patients exclusively on pressure support ventilation, which may affect generalizability across broader ventilatory modes or patient populations. Future multicenter validations and integration with clinical decision support systems could further accelerate clinical adoption.
This AI methodology aligns well with emerging trends aiming to personalize mechanical ventilation by dynamically adapting support based on continuous patient monitoring rather than intermittent assessments.
Conclusion
The novel AI algorithm offers a promising noninvasive, continuous, and accurate method to estimate inspiratory muscle pressure and detect patient-ventilator dyssynchrony during mechanical ventilation. Its performance is comparable to invasive esophageal manometry and intermittent occlusion maneuvers, potentially transforming respiratory monitoring in critically ill patients. Implementation of this technology could enhance ventilator management, reduce complications from dyssynchrony, and promote lung and diaphragm protective strategies.
Further studies are needed to confirm these findings in diverse clinical settings and to evaluate impact on patient-centered outcomes such as duration of mechanical ventilation, ICU length of stay, and survival.
Funding and ClinicalTrials.gov
The study was supported by institutional funding from the University of São Paulo. No specific clinical trial registration number was reported.
References
1. Plens GM, Morais CCA, Gregol T, et al. Artificial Intelligence Algorithm to Monitor Inspiratory Muscle Effort and Patient-Ventilator Dyssynchrony During Mechanical Ventilation. Crit Care Med. 2026;54(7):1585-1596. doi:10.1097/CCM.0000000000005974
2. Goligher EC, Jonkman AH, Dianti J, et al. The Lung and Diaphragm-Protective Ventilation. Ann Am Thorac Soc. 2021;18(3):525-533. doi:10.1513/AnnalsATS.202006-707CME
3. Yoshida T, Roldan R, Lima CAS, et al. High Respiratory Drive and Excessive Inspiratory Efforts Predict Risk of Patient Self-Inflicted Lung Injury. Am J Respir Crit Care Med. 2022;205(3):292-301. doi:10.1164/rccm.202012-4359OC
4. Blanch L, Villagra A, Sales B, Montanya J, Lucangelo U, Luján M, Cugat E, Benito S, Izquierdo J. Asynchronies during assisted mechanical ventilation are associated with mortality. Intensive Care Med. 2015;41(4):633-641. doi:10.1007/s00134-015-3705-7
5. Thille AW, Rodriguez P, Cabello B, Lellouche F, Brochard L. Patient-ventilator asynchrony during assisted mechanical ventilation. Intensive Care Med. 2006;32(10):1515-1522. doi:10.1007/s00134-006-0314-0
