Personalized Risk Prediction for 30-Day Readmissions With Venous Thromboembolism Using Machine Learning.
Jung In Park, Doyub Kim, Jung-Ah Lee and 2 others
PMID 33617689WHAT IT FOUND
A model built from hospital records flagged adults likely to return within 30 days with blood clots.
The best internal version scored 0.84 on a 0-to-1 performance scale and caught true cases at 0.74, but no patient care was tested.
Key findings
01The balanced random forest model performed best, with a 0.84 overall performance score and true-case detection at 0.74.
02The logistic regression model had the highest specificity at 0.84 but the lowest sensitivity at 0.69.
03The models used 158,804 adult hospital admissions from 92,481 distinct patients, including 2,080 admissions from 1,695 patients with 30-day VTE readmission.
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 data came from one tertiary academic hospital, so it is unknown whether the model works in other hospitals or patient groups. The VTE-positive group was small, 2,080 admissions or 1.31% of admissions, so the model's true-positive performance may be unstable. The study used coded EHR data and dropped admissions with missing values, which may have affected the sample and outcome classification. Validation was internal only, using 70% training, 30% test and 10-fold cross-validation; there was no external validation or real-world deployment. The paper reports model performance, not whether using the score changed discharge decisions, anticoagulation, or patient outcomes.
The easy way to misread this
Do not read the 0.84 score as proof that this model is ready for use in your discharge workflow. It was developed and validated only on internal data from one academic hospital, and the study did not test whether using it changed care or prevented readmissions.