Machine learning-based forecast of Helmet-CPAP therapy failure in Acute Respiratory Distress Syndrome patients
Riccardo Campia, Antonio De Santisa, Paolo Colombob, Paolo Scarpazzab, Marco Masserolia
aPolitecnico di Milano, Dipartimento di Elettronica, Informazione e Bioingegneria, Piazza Leonardo Da Vinci 32, Milano, MI, 20133, Italy
bAzienda Socio-Sanitaria Territoriale (ASST) della Brianza, Via Santi Cosma e Damiano 10, Vimercate, MB, 20871, Italy
Email addresses: riccardo.campi@mail.polimi.it (Riccardo Campi), antonio.desantis@polimi.it (Antonio De Santis), paolo.colombo@asst-brianza.it (Paolo Colombo), paolo.scarpazza@asst-brianza.it (Paolo Scarpazza), marco.masseroli@polimi.it (Marco Masseroli)
Helmet-Continuous Positive Airway Pressure (H-CPAP) is a non-invasive respiratory support that is used for the treatment of Acute Respiratory Distress Syndrome (ARDS), a severe medical condition diagnosed when symptoms like profound hypoxemia, pulmonary opacities on radiography, or unexplained respiratory failure are present. It can be classified as mild, moderate or severe. H-CPAP therapy is recommended as the initial treatment approach for mild ARDS. Even though the efficacy of H-CPAP in managing patients with moderate-to-severe hypoxemia remains unclear, its use has increased for these cases in response to the emergence of the COVID-19 Pandemic.
Using the electronic medical records (EMR) from the Pulmonology Department of Vimercate Hospital, in this study we develop and evaluate a Machine Learning (ML) system able to predict the failure of H-CPAP therapy on ARDS patients.
The Vimercate Hospital EMR provides demographic information, blood tests, and vital parameters of all hospitalizations of patients who are treated with H-CPAP and diagnosed with ARDS. This data is used to create a dataset of 622 records and 38 features, with 70-30% split between training and test set. Different ML models such as SVM, XGBoost, Neural Network, Random Forest, and Logistic Regression are iteratively trained in a cross-validation fashion. We also apply a feature selection algorithm to improve predictions quality and reduce the number of features.
The SVM and Neural Network models proved to be the most effective, achieving final accuracies of 95.19% and 94.65%, respectively. In terms of F1-score, the models scored 88.61% and 87.18%, respectively. Additionally, the SVM and XGBoost models performed well with a reduced number of features (23 and 13, respectively). The PaO2/FiO2 Ratio, C-Reactive Protein, and O2 Saturation resulted as the most important features, followed by Heartbeats, White Blood Cells, and D-Dimer, in accordance with the clinical scientific literature.
| File/folder | Description |
|---|---|
| /main.ipynb | code used to clean the datasets and train the models |
| /functions.py | accessory functions and classes |
| /models_vimercate/ | folder containing the trained models |
| /datasets/ | folder containing the dataset "dataset_vimercate.csv" used to train the models |
| /images/ | folder containing the obtained images |
Human participants were involved in this research; the study was conducted in accordance with the Declaration of Helsinki. Our study was approved by the local institution, Vimercate Hospital, ASST-Brianza, according to the legal requirements concerning observational studies (Resolutions 0000573 27/07/2021 and 0000133 22/02/2023).
Due to the nature of the present observational study and data anonymization, the patients' consent to participate was not required, as declared by the ASST Brianza Ethics Committee.