Prediction of 30-Day All-Cause Hospital Readmissions Using Limited Structured Electronic Health Record Data: Retrospective Comparative Study
Hospitals can identify readmission-risk patients within days using basic billing codes, allowing earlier prevention efforts.
Using a large dataset of 50,000 inpatient encounters, this study shows that a minimal structured code feature set (first 5 ICD-10 codes + 5 CPT codes + Charlson score) retains meaningful predictive signal for 30-day readmission — enabling earlier identification of at-risk patients without waiting for complete discharge documentation. Four ML models were compared with similar findings, with DistilBERT mapping codes to text for transformer-based prediction.
What the study was
- Study design
- Retrospective comparative ML modeling study
- Population
- Inpatient encounters from NY State Emergency Department Database 2019
- Sample size
- 50000
- Category
- Diagnostics
- Maturity
- Exploratory
- Journal
- JMIR Formative Research
Why it surfaced
Large-scale ML readmission study with clinically-friendly minimal feature approach. n=50K, external generalizability to be confirmed, but timeliness-accuracy trade-off findings directly relevant to hospital implementation. JMIR Form Res.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.