Vector 1

Demo · Causal Care

Did the care program work? And who should the nurses call?

A care management program cuts 30-day readmissions, but enrolled patients come back to the hospital more often than everyone else, because the care team enrolls its sickest patients first. This demo runs Vector 1's open-source causal-health-starter library in your browser on synthetic patients with a known true effect, and shows how to get both questions right.

Loading…

The synthetic health system

Each patient has an age, comorbidity score, prior admissions, social risk and living situation. The program truly lowers every patient's readmission risk, more so for patients with social risk. Enrollment favors the highest-risk patients. That's the confounding.

Question 1: Did the program reduce readmissions?

Question 2: Who should the care team call?

Most health systems pick patients with a readmission risk score. But the patients most likely to be readmitted aren't always the ones the program helps most. Rank them by estimated benefit instead, and the same number of calls prevents more readmissions.

True benefit by risk-score decile

Source: github.com/vector1-labs/causal-health-starter (version 0.1.0, MIT), companion code to the Causal Care series. Benefit ranking uses a two-model uplift estimate built from the library's outcome models. Synthetic patients only: no real patient data, and not clinical or medical advice.