A handmade journal that turns validated statistical evidence into an interactive prediction experience. No black boxes — every number knows where it came from.
Every chart from the notebook, reborn on the web — each one with its purpose, what we observed, how we interpreted it, and the business impact.
Strongest validated predictor. Monotonic rise from ~8.44% (0 admissions) to 31.40% (5), the most defensible signal in the dataset.
Readmission climbs from ~5.94% (1 diagnosis) to 17.65% (10). Nearly threefold — disease complexity matters.
Rises from 7.44% (5 meds) to 13.11% (20). A proxy for burden of illness, not a cause in itself.
8.18% (1 day) → 14.35% (10 days). Longer stays associated with more severe illness.
20–30 band spikes to 14.2%; elderly dominate absolute counts. Interpret carefully — never in isolation.
Used for feature exploration only. Never invoked to claim causation.
Rejected — sparse categories produced unstable 100% spikes. Excluded from the engine.
No machine learning — a transparent, evidence-weighted score that always shows how each factor contributed. Move a slider, watch the story change.
Move a slider — the evidence-weighted risk updates instantly, with a full breakdown.
Excluded by evidence review
Comparable to the cohort average — monitor discharge follow-up.
Number of Diagnoses is currently the dominant factor, adding 11.1 points to the evidence-weighted risk. Previous inpatient admissions carry the highest validated confidence (★★★★★) and shape the overall verdict most strongly; age is deliberately down-weighted because its per-band spikes reflect population imbalance more than causal risk.
Includes your evidence journey, slider inputs, and the final explainable prediction.
Repeated hospitalizations were the strongest indicator of future readmissions.
Disease complexity — captured through diagnosis and medication counts — significantly increased readmission likelihood.
Medication burden almost certainly reflects illness severity rather than causing readmissions itself.
Longer hospital stays were associated with greater readmission risk.
Age alone should never be interpreted in isolation — the 20–30 spike is a small-cohort artefact.
Emergency and outpatient visit counts were intentionally excluded because the evidence was unreliable.
The correlation matrix informed feature exploration but never claimed causation.
The best model is the one that explains itself. Every number on this page can be traced back to a chart, a cohort, a validated finding. That is what makes it trustworthy — not the size of a network, but the clarity of a story.
Patients with 5 previous inpatient admissions had nearly 4× the readmission rate of patients with none.