Prediction and simulation of PEEP setting effects with machine learning models / Predicción y simulación de los efectos de la configuración de la PEEP con modelos de aprendizaje automático

Objective To establish a new machine learning-based method to adjust positive end-expiratory pressure (PEEP) using only already routinely measured data. Design Retrospective observational study. Setting Intensive care unit (ICU). Patients or participants 51811 mechanically ventilated patients in multiple ICUs in the USA (data from MIMIC-III and eICU databases). Interventions No interventions. Main variables of interest Success parameters of ventilation (arterial partial pressures of oxygen and carbon dioxide and respiratory system compliance). Results The multi-tasking neural network model performed significantly best for all target tasks in the primary test set. The model predicts arterial partial pressures of oxygen and carbon dioxide and respiratory system compliance about 45 min into the future with mean absolute percentage errors of about 21.7%, 10.0% and 15.8%, respectively. The proposed use of the model was demonstrated in case scenarios, where we simulated possible effects of PEEP adjustments for individual cases. Conclusions Our study implies that machine learning approach to PEEP titration is a promising new method which comes with no extra cost once the infrastructure is in place. Availability of databases with most recent ICU patient data is crucial for the refinement of prediction performance. (AU)

Objetivo Establecer un nuevo método basado en el aprendizaje automático para ajustar la presión positiva al final de la espiración (PEEP según sus siglas en inglés) utilizando únicamente datos ya obtenidos de forma rutinaria. Diseño Estudio retrospectivo de observación. Ámbito Unidad de cuidados intesivos (UCI) Pacientes o participantes 51811 pacientes ventilados mecánicamente en múltiples UCIs de EE.UU. (tomados de las bases de datos MIMIC-III y eICU). Intervenciones Sin intervenciones. Variables de interés principales Parametros de éxito de la ventilación (presiones parciales arteriales de oxígeno y dióxido de carbono y distensibilidad del sistema respiratorio). Resultados El modelo de red neuronal multitarea obtuvo los mejores resultados en todos los objetivos del conjunto de pruebas primario. El modelo predice las presiones parciales arteriales de oxígeno y dióxido de carbono así como la distensibilidad del sistema respiratorio con aproximadamente 45 minutos de anticipación, mostrando errores porcentuales absolutos medios de aproximadamente 21.7%, 10.0% y 15.8%, respectivamente. El uso propuesto del modelo se demostró en situaciones hipotéticas en las que se simularon los posibles efectos de los ajustes de PEEP para casos individuales. Conclusiones Nuestro estudio implica que el enfoque de aprendizaje automático para el ajuste de la PEEP es un método nuevo y prometedor que no supone ningún coste adicional una vez que se dispone de la infraestructura necesaria. La disponibilidad de bases de datos con información de pacientes de UCI más recientes es crucial para perfeccionar el rendimiento de la predicción. (AU)

Prediction and simulation of PEEP setting effects with machine learning models.

OBJECTIVE: To establish a new machine learning-based method to adjust positive end-expiratory pressure (PEEP) using only already routinely measured data. DESIGN: Retrospective observational study. SETTING: Intensive care unit (ICU). PATIENTS OR PARTICIPANTS: 51811 mechanically ventilated patients in multiple ICUs in the USA (data from MIMIC-III and eICU databases). INTERVENTIONS: No interventions. MAIN VARIABLES OF INTEREST: Success parameters of ventilation (arterial partial pressures of oxygen and carbon dioxide and respiratory system compliance) RESULTS: The multi-tasking neural network model performed significantly best for all target tasks in the primary test set. The model predicts arterial partial pressures of oxygen and carbon dioxide and respiratory system compliance about 45 min into the future with mean absolute percentage errors of about 21.7%, 10.0% and 15.8%, respectively. The proposed use of the model was demonstrated in case scenarios, where we simulated possible effects of PEEP adjustments for individual cases. CONCLUSIONS: Our study implies that machine learning approach to PEEP titration is a promising new method which comes with no extra cost once the infrastructure is in place. Availability of databases with most recent ICU patient data is crucial for the refinement of prediction performance.

Effect of routine suction on lung aeration in critically ill neonates and young infants measured with electrical impedance tomography.

Endotracheal suctioning is a widely used procedure to remove secretions from the airways of ventilated patients. Despite its prevalence, regional effects of this maneuver have seldom been studied. In this study, we explore its effects on regional lung aeration in neonates and young infants using electrical impedance tomography (EIT) as part of the large EU-funded multicenter observational study CRADL. 200 neonates and young infants in intensive care units were monitored with EIT for up to 72 h. EIT parameters were calculated to detect changes in ventilation distribution, ventilation inhomogeneity and ventilation quantity on a breath-by-breath level 5-10 min before and after suctioning. The intratidal change in aeration over time was investigated by means of regional expiratory time constants calculated from all respiratory cycles using an innovative procedure and visualized by 2D maps of the thoracic cross-section. 344 tracheal suctioning events from 51 patients could be analyzed. They showed no or very small changes of EIT parameters, with a dorsal shift of the center of ventilation by 0.5% of the chest diameter and a 7% decrease of tidal impedance variation after suctioning. Regional time constants did not change significantly. Routine suctioning led to EIT-detectable but merely small changes of the ventilation distribution in this study population. While still a measure requiring further study, the time constant maps may help clinicians interpret ventilation mechanics in specific cases.