Predicting responders to prone positioning in mechanically ventilated patients with COVID-19 using machine learning

Abstract

Background For mechanically ventilated critically ill COVID-19 patients, prone positioning has quickly become an important treatment strategy, however, prone positioning is labor intensive and comes with potential adverse effects. Therefore, identifying which critically ill intubated COVID-19 patients will benefit may help allocate labor resources. Methods From the multi-center Dutch Data Warehouse of COVID-19 ICU patients from 25 hospitals, we selected all 3619 episodes of prone positioning in 1142 invasively mechanically ventilated patients. We excluded episodes longer than 24 h. Berlin ARDS criteria were not formally documented. We used supervised machine learning algorithms Logistic Regression, Random Forest, Naive Bayes, K-Nearest Neighbors, Support Vector Machine and Extreme Gradient Boosting on readily available and clinically relevant features to predict success of prone positioning after 4 h (window of 1 to 7 h) based on various possible outcomes. These outcomes were defined as improvements of at least 10% in PaO2/FiO2 ratio, ventilatory ratio, respiratory system compliance, or mechanical power. Separate models were created for each of these outcomes. Re-supination within 4 h after pronation was labeled as failure. We also developed models using a 20 mmHg improvement cut-off for PaO2/FiO2 ratio and using a combined outcome parameter. For all models, we evaluated feature importance expressed as contribution to predictive performance based on their relative ranking. Results The median duration of prone episodes was 17 h (11–20, median and IQR, N = 2632). Despite extensive modeling using a plethora of machine learning techniques and a large number of potentially clinically relevant features, discrimination between responders and non-responders remained poor with an area under the receiver operator characteristic curve of 0.62 for PaO2/FiO2 ratio using Logistic Regression, Random Forest and XGBoost. Feature importance was inconsistent between models for different outcomes. Notably, not even being a previous responder to prone positioning, or PEEP-levels before prone positioning, provided any meaningful contribution to predicting a successful next proning episode. Conclusions In mechanically ventilated COVID-19 patients, predicting the success of prone positioning using clinically relevant and readily available parameters from electronic health records is currently not feasible. Given the current evidence base, a liberal approach to proning in all patients with severe COVID-19 ARDS is therefore justified and in particular regardless of previous results of proning.

Changes in Driving Pressure vs Oxygenation as Predictor of Mortality in Moderate to Severe Acute Respiratory Distress Syndrome Patients Receiving Prone Position Ventilation

“…They found that the PaO 2 /FiO 2 improvement was significant between responders and non-responders which was concurrent with our findings. 12 – 14 Guerin et al concluded that though PPV improves oxygenation and ventilation and reduces mortality, there was no association between them. 7 , 33 This infers that the better prognosis in ARDS patients receiving PPV is due to the ability of PPV to reduce VILI.…”

“…With this background, we aimed to determine the changes and the role of DP as compared to improvement in oxygenation (PaO 2 /FiO 2 ) at different time points as predictors of mortality outcomes in moderate-severe ARDS patients with PPV. 14 …”

Validity of the total SOFA score in patients ≥ 80 years old acutely admitted to intensive care units: a post-hoc analysis of the VIP2 prospective, international cohort study

“…Second, the introduction of a complex clinical tool using machine learning, assessment of statistical interactions and other sophisticated mathematical tools would not have been feasible in 1996. Conversely, the current availability of smartphones, much more powerful than personal computers means there is great potential for the creation of new prognostic tools and this should be considered when the decision is made to update the SOFA score [ 14 ].…”

Predicting the Length of Mechanical Ventilation in Acute Respiratory Disease Syndrome Using Machine Learning: The PIONEER Study

“…ML flexibility detects relationships between potential clinical features, physiologic parameters, and an outcome [ 13 ]. Despite extensive modeling and a large number of clinically relevant features, the discrimination using ML approaches for predicting responders to the prone position in mechanically ventilated patients with COVID-19 [ 14 ] was very poor. Scarce studies have evaluated the role of ML in predicting the duration of MV in ARDS patients [ 11 , 15 , 16 , 17 ].…”