Tariq A Dam1, Luca F Roggeveen2, Fuda van Diggelen3, Lucas M Fleuren2, Ameet R Jagesar2, Martijn Otten2, Heder J de Vries2, Diederik Gommers4, Olaf L Cremer5, Rob J Bosman6, Sander Rigter7, Evert-Jan Wils8, Tim Frenzel9, Dave A Dongelmans10, Remko de Jong11, Marco A A Peters12, Marlijn J A Kamps13, Dharmanand Ramnarain14, Ralph Nowitzky15, Fleur G C A Nooteboom16, Wouter de Ruijter17, Louise C Urlings-Strop18, Ellen G M Smit19, D Jannet Mehagnoul-Schipper20, Tom Dormans21, Cornelis P C de Jager22, Stefaan H A Hendriks23, Sefanja Achterberg24, Evelien Oostdijk25, Auke C Reidinga26, Barbara Festen-Spanjer27, Gert B Brunnekreef28, Alexander D Cornet29, Walter van den Tempel30, Age D Boelens31, Peter Koetsier32, Judith Lens33, Harald J Faber34, A Karakus35, Robert Entjes36, Paul de Jong37, Thijs C D Rettig38, Sesmu Arbous39, Sebastiaan J J Vonk40, Tomas Machado40, Willem E Herter40, Harm-Jan de Grooth2, Patrick J Thoral2, Armand R J Girbes2, Mark Hoogendoorn3, Paul W G Elbers2. 1. Department of Intensive Care Medicine, Laboratory for Critical Care Computational Intelligence, Amsterdam Medical Data Science, Amsterdam UMC, Vrije Universiteit, Amsterdam, The Netherlands. t.dam@amsterdamumc.nl. 2. Department of Intensive Care Medicine, Laboratory for Critical Care Computational Intelligence, Amsterdam Medical Data Science, Amsterdam UMC, Vrije Universiteit, Amsterdam, The Netherlands. 3. Quantitative Data Analytics Group, Department of Computer Science, Faculty of Science, VU University, Amsterdam, The Netherlands. 4. Department of Intensive Care, Erasmus Medical Center, Rotterdam, The Netherlands. 5. Intensive Care, UMC Utrecht, Utrecht, The Netherlands. 6. ICU, OLVG, Amsterdam, The Netherlands. 7. Department of Anesthesiology and Intensive Care, St. Antonius Hospital, Nieuwegein, The Netherlands. 8. Department of Intensive Care, Franciscus Gasthuis & Vlietland, Rotterdam, The Netherlands. 9. Department of Intensive Care Medicine, Radboud University Medical Center, Nijmegen, The Netherlands. 10. Department of Intensive Care Medicine, Amsterdam UMC, Amsterdam, The Netherlands. 11. Intensive Care, Bovenij Ziekenhuis, Amsterdam, The Netherlands. 12. Intensive Care, Canisius Wilhelmina Ziekenhuis, Nijmegen, The Netherlands. 13. Intensive Care, Catharina Ziekenhuis Eindhoven, Eindhoven, The Netherlands. 14. Department of Intensive Care, ETZ Tilburg, Tilburg, The Netherlands. 15. Intensive Care, HagaZiekenhuis, Den Haag, The Netherlands. 16. Intensive Care, Laurentius Ziekenhuis, Roermond, The Netherlands. 17. Department of Intensive Care Medicine, Northwest Clinics, Alkmaar, The Netherlands. 18. Intensive Care, Reinier de Graaf Gasthuis, Delft, The Netherlands. 19. Intensive Care, Spaarne Gasthuis, Haarlem en Hoofddorp, The Netherlands. 20. Intensive Care, VieCuri Medisch Centrum, Venlo, The Netherlands. 21. Intensive Care, Zuyderland MC, Heerlen, The Netherlands. 22. Department of Intensive Care, Jeroen Bosch Ziekenhuis, Den Bosch, The Netherlands. 23. Intensive Care, Albert Schweitzerziekenhuis, Dordrecht, The Netherlands. 24. ICU, Haaglanden Medisch Centrum, Den Haag, The Netherlands. 25. ICU, Maasstad Ziekenhuis Rotterdam, Rotterdam, The Netherlands. 26. ICU, SEH, BWC, Martiniziekenhuis, Groningen, The Netherlands. 27. Intensive Care, Ziekenhuis Gelderse Vallei, Ede, The Netherlands. 28. Department of Intensive Care, Ziekenhuisgroep Twente, Almelo, The Netherlands. 29. Department of Intensive Care, Medisch Spectrum Twente, Enschede, The Netherlands. 30. Department of Intensive Care, Ikazia Ziekenhuis Rotterdam, Rotterdam, The Netherlands. 31. Antonius Ziekenhuis Sneek, Sneek, The Netherlands. 32. Intensive Care, Medisch Centrum Leeuwarden, Leeuwarden, The Netherlands. 33. ICU, IJsselland Ziekenhuis, Capelle aan den IJssel, The Netherlands. 34. ICU, WZA, Assen, The Netherlands. 35. Department of Intensive Care, Diakonessenhuis Hospital, Utrecht, The Netherlands. 36. Department of Intensive Care, Adrz, Goes, The Netherlands. 37. Department of Anesthesia and Intensive Care, Slingeland Ziekenhuis, Doetinchem, The Netherlands. 38. Department of Anesthesiology, Intensive Care and Pain Medicine, Amphia Ziekenhuis, Breda, The Netherlands. 39. LUMC, Leiden, The Netherlands. 40. Pacmed, Amsterdam, The Netherlands.
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.
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.
Authors: V Marco Ranieri; Gordon D Rubenfeld; B Taylor Thompson; Niall D Ferguson; Ellen Caldwell; Eddy Fan; Luigi Camporota; Arthur S Slutsky Journal: JAMA Date: 2012-06-20 Impact factor: 56.272
Authors: Pratik Sinha; Carolyn S Calfee; Jeremy R Beitler; Neil Soni; Kelly Ho; Michael A Matthay; Richard H Kallet Journal: Am J Respir Crit Care Med Date: 2019-02-01 Impact factor: 30.528
Authors: Claude Guérin; Richard K Albert; Jeremy Beitler; Luciano Gattinoni; Samir Jaber; John J Marini; Laveena Munshi; Laurent Papazian; Antonio Pesenti; Antoine Vieillard-Baron; Jordi Mancebo Journal: Intensive Care Med Date: 2020-11-10 Impact factor: 41.787