Readmission Analytics

The question
behind the work.

Thirty-day readmissions combine clinical complexity, utilization, and equity concerns. The raw encounter data is too messy for reliable modeling or BI without a disciplined data layer.

My role

I built the normalized MySQL schema, staging and mapping workflow, analysis view, R modeling pipeline, subgroup analysis, dashboard exports, and three-page Power BI report.

What I built.

The project starts with normalized tables and analysis views, then moves through regression and equity analysis into a three-page decision report for clinical operations.

• Loaded and normalized roughly 100,000 encounters from the Diabetes 130-US Hospitals dataset.
• Mapped diagnoses, medications, labs, demographics, and readmission outcomes into an analysis-ready view.
• Built a binary 30-day readmission outcome and logistic model in R.
• Published overview, equity, and clinical-driver pages with consistent measures and subgroup comparisons.

The decisions
that shaped it.

1. Separated staging, normalized entities, and the analysis view so cleaning logic is reusable outside Power BI.
2. Reported subgroup gaps with context instead of using equity visuals as causal claims.
3. Combined model results with operationally legible thresholds such as diagnoses, medication count, and length of stay.

Evaluation & results

The cohort showed an overall readmission rate near 11%, a roughly 2.7-point gap between the lowest and highest racial subgroups, and higher observed risk among encounters with more diagnoses, more medications, and longer stays.