Interactive Power BI model simulation • Click any slicer below to filter cohort
Click any bar to quickly slice by that cohort
The 3x Risk Multiplier: Prior admission count is by far the single most actionable predictor. Age showed minimal variation (9.8% to 12.4%), proving that clinical utilization history—not age—drives readmission.
A healthcare analytics project identifying which diabetic patients are at high risk of hospital readmission within 30 days, and what a hospital can do about it.
Portfolio project | Data cleaning, EDA, SQL, statistical testing, interactive BI dashboard
Gulam Mustafa Hospital lacks a systematic way to identify which diabetic patients are at high risk of returning to the hospital within 30 days of discharge, resulting in preventable costs and missed opportunities for early intervention.
Hospital readmissions are a critical quality, cost, and reputational metric across healthcare systems globally — including GCC hospitals, where readmission rates are tracked as part of JCI accreditation and align with national health transformation goals such as Saudi Vision 2030.
(Note: "Gulam Mustafa Hospital" is a fictional name used to frame this portfolio project in a realistic business context. The underlying data is real, publicly available clinical data — see Dataset section below.)
| Stakeholder | Role | What they need from this analysis |
|---|---|---|
| Care Coordinator | Decision-maker | A prioritized list of high-risk patients to schedule follow-up calls or home visits |
| CFO | Influencer | Estimated cost of preventable readmissions to justify budget for intervention programs |
| Insurance Payer | Indirectly affected | Visibility into preventable readmission drivers, since they bear repeat hospitalization costs |
| Tool | Purpose |
|---|---|
| Python (pandas) | Data cleaning, exploratory data analysis, statistical testing |
| SQL (SQLite) | Relational querying and validation of key findings |
| Power BI / React | Interactive dashboard for stakeholder-facing reporting |
| Jupyter Notebook | Documented, reproducible analysis workflow |
| Git / GitHub | Version control and portfolio presentation |
payer_code, medical_specialty).| Factor | Finding | Strength |
|---|---|---|
| Prior inpatient admissions | Readmission rate rises from 8.6% (0 prior admissions) to 26.2% (3+ prior admissions) | Strong — primary risk driver |
| Age | Rate stays roughly flat (9.8%–12.4%) across most adult age groups | Weak — not a meaningful driver, despite intuitive expectations |
| Length of stay | Rate rises from 8.4% (1 day) to ~14.5% (8+ days), likely reflecting case severity | Moderate-strong, correlational not causal |
| A1C testing | Untested patients show a higher readmission rate (11.6%) than tested patients (9.9%), despite only 16.9% of patients being tested | Modest but highly actionable |
Statistical testing confirmed both number_inpatient ($\chi^2 = 6,376$) and age ($\chi^2 = 357.7$) as technically significant ($p < 0.05$) — but the much larger effect size for number_inpatient demonstrates it as the far more practically important factor for intervention targeting.
-- Query: Readmission Rate by Prior Inpatient Admission History
SELECT
CASE
WHEN number_inpatient = 0 THEN '0 inpatient'
WHEN number_inpatient = 1 THEN '1 inpatient'
WHEN number_inpatient = 2 THEN '2 inpatient'
WHEN number_inpatient >= 3 THEN '3+ inpatient'
END AS inpatient_bucket,
COUNT(*) AS total_patients,
SUM(CASE WHEN readmitted = '<30' THEN 1 ELSE 0 END) AS readmitted_30_count,
ROUND(SUM(CASE WHEN readmitted = '<30' THEN 1 ELSE 0 END) * 100.0 / COUNT(*), 1) AS readmit_30_pct
FROM patient_encounters
GROUP BY inpatient_bucket;