Use of Machine Learning Models to Predict Microaspiration Measured by Tracheal Pepsin A.
Annette Bourgault, Ilana Logvinov, Chang Liu and 3 others
PMID 39740967WHAT IT FOUND
In mechanically ventilated adults, models using heart rate, blood pressure, body mass index, tube duration and age identified the presence of tracheal pepsin A, a marker of microaspiration.
The best model outperformed a simple comparison model, but needs external testing before use.
Key findings
01Pepsin A concentration of 6.25 ng/mL or higher, indicating microaspiration or regurgitation, was found in 39% of patients.
02The random forest model had the highest tracheal pepsin A predictive value, area under the curve 0.844 (95% CI, 0.792-0.897), followed by XGBoost 0.817 (95% CI, 0.758-0.876), SVM-RFE 0.752 (95% CI, 0.685-0.819), and logistic regression 0.627 (95% CI, 0.553-0.701).
03The top five variables identified by the random forest model as important for tracheal pepsin A were heart rate, mean arterial pressure, body mass index, endotracheal tube duration, and age.
STILL TO COME
How it was doneWhat they foundWhat it means for RNs
Read the rest of this summary
You get three full summaries a month, free, and we do not ask for a card. Search, the TL;DRs and your library stay unlimited either way.
What it does not show
The models were built from one existing dataset and the paper does not report testing them in a new patient group. The original study was powered for pulmonary versus gastrointestinal outcomes, not for machine learning model development. Pepsin A was used as a proxy for microaspiration or regurgitation, not as a direct clinical outcome such as ventilator-associated pneumonia. The models were trained with enteral feeding and other available variables, but specific microaspiration variables such as gastrointestinal symptoms were not in the database. The identified predictors did not match predictors for ventilator-associated pneumonia, so the models may not identify microaspiration from all causes. Missing data were imputed using means, and variables considered irrelevant were excluded.
Declared interests
No author conflicts or funding declaration are provided in the supplied text. The publication types note NIH extramural research support.
The easy way to misread this
Do not treat this as a tested tool for identifying microaspiration at the bedside. The models were built and cross-validated on one existing dataset, and the paper does not show that acting on them helps patients.