Using Machine Learning to Identify Social Determinants of Health that Impact Discharge Disposition for Hospitalized Patients.
He Ren, Chun Wang, David J Weiss and 4 others
PMID 40023505WHAT IT FOUND
In 134,807 hospitalized adults, lack of a partner, older age, more comorbidities, unstable finances, no employment, poor transportation, limited dental care, little exercise, or few social contacts were linked to a higher chance of skilled nursing facility discharge.
Key findings
01Patients without a partner had 15% to 38% higher odds of skilled nursing facility discharge than married or partnered patients.
02More frequent alcohol use, regular dental check-ups, active employment, stable finances, olive oil use, exercise, phone interactions, and adequate transportation were associated with lower odds of skilled nursing facility discharge.
03The model had high accuracy and specificity but low sensitivity, and the authors attribute this to an extremely imbalanced outcome.
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 study was retrospective and used existing records, so it describes associations rather than tested effects. Patients needed sufficient English fluency to complete the questionnaire, so the results may not apply to patients who could not participate. The analysis used only social determinant items available in Mayo Clinic's electronic health record, so other social factors were not considered. The sample was mostly white, non-Hispanic, and married, and minority groups were smaller, so findings may not apply to more diverse populations. Missing data were common, with 15.70% overall missingness, and were handled by statistical methods rather than complete records. The model had low sensitivity, so it did not reliably identify patients who went to a skilled nursing facility. The models were developed and tested using a split of one healthcare system's dataset, and real-time prediction was not available because the missing-data method was computationally intensive. Fairness results were mixed: one check suggested lower accuracy for men, while calibration looked acceptable.
Declared interests
The study was part of Project HoPE (R01AG077706–02). The authors declared no conflicts of interest.
The easy way to misread this
Do not use this model as a bedside decision tool or assume the social factors caused skilled nursing facility discharge. The study was retrospective, the model had low sensitivity, and the findings describe associations rather than tested interventions.