The Journal of Clinical Investigation (JCI)Machine Learning Links Unresolving Secondary Pneumonia to Mortality in Patients with Severe Pneumonia, Including COVID-19

In a recent single-center prospective cohort study, the effect of unsuccessful treatment of ventilator-associated pneumonia (VAP) on mortality rates in patients with severe pneumonia was evaluated. The study involved 585 mechanically ventilated patients with severe pneumonia and respiratory failure, including 190 with severe COVID-19, all of whom underwent at least one bronchoalveolar lavage. A novel machine learning approach, CarpeDiem, was employed to group similar ICU patient-days into clinical states based on electronic health record data. Findings indicated that the lengthy ICU stays experienced by COVID-19 patients were largely due to prolonged periods of respiratory failure.

The study discovered that although VAP wasn’t a contributing factor to mortality rates overall, patients with a single episode of unsuccessfully treated VAP demonstrated a significantly higher mortality rate (76.4%) compared to those with successfully treated VAP (17.6%). Furthermore, the CarpeDiem model highlighted that unresolved VAP was linked with transitions to clinical states associated with increased mortality. The study underscores that unsuccessful treatment of VAP is correlated with higher mortality and that patients with COVID-19 are at an elevated risk of VAP due to prolonged respiratory failure.