Hospital Readmissions Analytics — Clinical Quality & Adoption Metrics Target completion: 30 May 2026
Which hospitals show consistently low adoption of discharge protocol improvements, and what patterns in clinical quality metrics predict readmission rates?
This mirrors the analytical challenge in hospital medical device contexts: understanding where clinical workflow changes are not being adopted, and what measurable signals predict poor outcomes. Directly relevant to Philips’s focus on hospital medical device analytics and adoption measurement.
CMS Hospital Readmissions Reduction Program data Source: kaggle.com/datasets/dansbecker/hospital-readmissions
Real CMS (Centers for Medicare & Medicaid Services) data. Hospital-level readmission rates, excess readmission ratios, discharge counts, and condition categories across US hospitals.
Real-world equivalent: MIMIC-III (PhysioNet) — de-identified ICU clinical data from Beth Israel Deaconess Medical Center. Requires credentialed access (physionet.org). The architecture of this project is designed to extend directly to MIMIC-III.
hospital-readmissions-dbthospital_analyticsraw, staging, intermediate, martsraw.hospital_readmissions via BigQuery UI or bq loadproject and dataset in dbt profiles.ymlstaging/)Clean and type-cast raw data. No business logic. One model per source table.
stg_hospital_readmissions.sql
stg_conditions.sql
Tests: not_null and unique on hospital identifier. accepted_values on condition categories.
intermediate/)Apply business logic and calculate derived metrics.
int_hospital_readmission_rates.sql
int_hospital_adoption_signals.sql
Tests: not_null on all calculated fields. relationships between intermediate and staging layers.
marts/)Analytics-ready tables for reporting and visualisation.
mart_adoption_summary.sql
mart_condition_performance.sql
mart_national_benchmarks.sql
Tests: unique on primary keys. not_null on all dimension and measure fields.
Every model has a description in schema.yml. Every column has a description. dbt docs generated and linked from the GitHub README.
The documentation layer is the data catalogue — this directly demonstrates data governance thinking relevant to the Philips analytics engineering role.
The project answers a question that hospital medical device teams ask continuously: adoption of protocol improvements does not happen uniformly across hospitals. Some hospitals consistently perform above national benchmarks across multiple conditions — others do not. Understanding which hospitals are in which tier, and what distinguishes them, is the analytical foundation for targeted intervention.
In a Philips device context: which clinical sites are adopting new device features or workflow changes, and which are not? The dbt architecture here surfaces that signal from outcome data — the same pattern applies to telemetry and usage data from medical devices.
| Task | Time |
|---|---|
| BigQuery setup + data load | 30 minutes |
| Staging models + tests | 2 hours |
| Intermediate models + tests | 2 hours |
| Mart models + tests | 1 hour |
| Documentation (schema.yml) | 1 hour |
| README + dbt docs | 30 minutes |
| Total | ~7 hours |
Recommended split: two sessions. Session 1: BigQuery setup + staging layer. Session 2: intermediate + marts + documentation.
Use Claude Code for all model writing and debugging. Open with the ForAI.md prompt in this folder.
# Hospital Readmissions Analytics — dbt + BigQuery
## Business Question
## Dataset
## Architecture (with DAG screenshot from dbt docs)
## How to Run
## Key Findings
## Real-World Extension (MIMIC-III)
The “Real-World Extension” section notes that this architecture is compatible with MIMIC-III (PhysioNet) — the gold standard clinical dataset — and explains what additional models that would enable. Signals clinical domain awareness without requiring credentialed access.
This project demonstrates:
Plan version: May 2026 — updated stack to BigQuery (native GCP environment, no setup friction) Target role: Philips UX Telemetry Expert — Adoption & Insights, Eindhoven