United StatesHealthcare and life sciences2026 – ongoing

Duly Health and Care

30-day hospital readmission prediction model

Duly Health and Care cut hospital readmissions from 14% to 10% with a prediction model built on Databricks

WorkingVendor case studyVendor ↗

Problem

Duly Health and Care, a physician-directed medical group in Chicago, runs fee-for-service billing alongside Medicare Advantage value-based care. Readmissions after hospital stays affected both patient health and cost.

Approach

The team built a 30-day readmission model with MLflow and AutoML, comparing Random Forest, LightGBM and XGBoost. It was trained on 100,000 historical patients and reached a ROC AUC of 89.2%. The group also built Genie agents for plain-language questions on live data and a conversational tool to help providers refer patients to in-network physicians.

Results

29%reduction in hospital readmission rates, from 14% to 10%Vendor case study[1]

$4.5Mannual savings from lower readmissionsVendor case study[1]

89.2%ROC AUC of the readmission modelVendor case study[1]

Takeaway

The readmission rate fell by 29% in relative terms. The company did not say how much of the drop came from the model itself rather than from other care changes.

Sources

Keeping every patient healthier as fast as care demands
Databricks10 Jul 2026Vendor publication

Source summaries

[1]Databricks10 Jul 2026

Databricks describes how Duly Health and Care moved its data to the Databricks platform and built a readmission prediction model on it. The page reports a 29% fall in readmissions, from 14% to 10%, worth $4.5M a year, and a model ROC AUC of 89.2%. It also reports 30% lower data platform costs than SQL Server and about 106 projects completed in 2025 against 51 in 2024.

Full details are in the original sources.

Added 4 Nov 2026 · Updated 4 Nov 2026