Skip to main content
Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

< Back to Article

Fig 1.

Study cohort.

Flow diagram of 2016 model derivation cohort and 2017 testing cohort.

More »

Fig 1 Expand

Fig 2.

Timeline for prediction.

At the patient’s time of eligibility (i.e., when they develop moderate hypoxia), the patient’s risk of future ARDS was predicted using the most recent six hours of data.

More »

Fig 2 Expand

Fig 3.

Model training, validation and testing pipeline.

The data was split temporally into a training/validation dataset (2016) and testing dataset (2017). Step 1: 5-fold cross validation was performed using the 2016 dataset to identify the optimal model hyperparameter. Step 2: The model was re-trained using the optimal hyperparameter on the entire 2016 dataset to learn model parameters. Step 3: The model was evaluated on held-out test data from 2017.

More »

Fig 3 Expand

Table 1.

Study population characteristics.

More »

Table 1 Expand

Fig 4.

Model Performance.

Performance of the ARDS risk prediction model (L2-regulized model) in the 2017 test cohort. A. ROC curve and 95% interval estimates. B. Confusion matrix with 95% interval estimates.

More »

Fig 4 Expand

Fig 5.

Model Sensitivity stratified by ARDS time of onset.

Model performance in subgroups of ARDS patients based on time from ARDS risk stratification to ARDS onset.

More »

Fig 5 Expand

Table 2.

Top predictive factors.

Top 10 risk factors and top 10 protective factors identified in the model to risk stratify patients for ARDS.

More »

Table 2 Expand