Background

Noninvasive respiratory support modalities are common alternatives to mechanical ventilation in acute hypoxemic respiratory failure. However, studies historically compare noninvasive respiratory support to conventional oxygen rather than mechanical ventilation. In this study, we compared outcomes in patients with acute hypoxemic respiratory failure treated initially with noninvasive respiratory support to patients treated initially with invasive mechanical ventilation.

Methods

This is a retrospective observational cohort study between January 1, 2018 and December 31, 2019 at a large healthcare network in the United States. We used a validated phenotyping algorithm to classify adult patients (≥18 years) with eligible International Classification of Diseases codes into two cohorts: those treated initially with noninvasive respiratory support or those treated invasive mechanical ventilation only. The primary outcome was time-to-in-hospital death analyzed using an inverse probability of treatment weighted Cox model adjusted for potential confounders. Secondary outcomes included time-to-hospital discharge alive. A secondary analysis was conducted to examine potential differences between noninvasive positive pressure ventilation and nasal high flow.

Results

During the study period, 3177 patients met inclusion criteria (40% invasive mechanical ventilation, 60% noninvasive respiratory support). Initial noninvasive respiratory support was not associated with a decreased hazard of in-hospital death (HR: 0.65, 95% CI: 0.35-1.2), but was associated with an increased hazard of discharge alive (HR: 2.26, 95% CI: 1.92-2.67). In-hospital death varied between the nasal high flow (HR 3.27, 95% CI: 1.43-7.45) and noninvasive positive pressure ventilation (HR 0.52, 95% CI 0.25-1.07), but both were associated with increased likelihood of discharge alive (nasal high flow HR 2.12, 95 CI: 1.25-3.57; noninvasive positive pressure ventilation HR 2.29, 95% CI: 1.92-2.74).

Conclusions

These data show that noninvasive respiratory support is not associated with reduced hazards of in-hospital death but is associated with hospital discharge alive.

PURPOSE: To compare the evaluation metrics for deep learning methods that were developed using imbalanced imaging data in osteoarthritis studies. MATERIALS AND METHODS: This retrospective study utilized 2996 sagittal intermediate-weighted fat-suppressed knee MRIs with MRI Osteoarthritis Knee Score readings from 2467 participants in the Osteoarthritis Initiative study. We obtained probabilities of the presence of bone marrow lesions (BMLs) from MRIs in the testing dataset at the sub-region (15 sub-regions), compartment, and whole-knee levels based on the trained deep learning models. We compared different evaluation metrics (e.g., receiver operating characteristic (ROC) and precision-recall (PR) curves) in the testing dataset with various class ratios (presence of BMLs vs. absence of BMLs) at these three data levels to assess the models performance. RESULTS: In a subregion with an extremely high imbalance ratio, the model achieved a ROC-AUC of 0.84, a PR-AUC of 0.10, a sensitivity of 0, and a specificity of 1. CONCLUSION: The commonly used ROC curve is not sufficiently informative, especially in the case of imbalanced data. We provide the following practical suggestions based on our data analysis: 1) ROC-AUC is recommended for balanced data, 2) PR-AUC should be used for moderately imbalanced data (i.e., when the proportion of the minor class is above 5% and less than 50%), and 3) for severely imbalanced data (i.e., when the proportion of the minor class is below 5%), it is not practical to apply a deep learning model, even with the application of techniques addressing imbalanced data issues.