When Does the Model Pay? Postpartum Readmission Economics
When Does the Model Pay?
An interactive explorer of the decision economics of predicting readmission in the first 14 days after a first birth.
The question
A risk model saves money only if it beats the two strategies that need no model: contact every new mother after discharge, or contact none. Whether it does depends on what one outreach contact costs, how many readmissions a contact prevents, what a readmission costs, and how well the model ranks risk. The explorer lets you set each of these and see where, if anywhere, a model earns its place.
What you can explore
- Cost of a contact. Move it from a few dollars to hundreds, and see when contacting everyone, no one, or only the women the model flags is the better choice.
- The models. Logistic regression, gradient boosting, a stacking ensemble and two automated machine learning frameworks, trained on nuMoM2b and scored on an untouched test set. You can also compare them with a hypothetical better model.
- States. The test set reweighted to each US state’s first births, with each state’s Medicaid payment for a home visit or a phone call. This view is post hoc and exploratory.
- How it was built. The explorer explains, step by step, how the state estimates were made, including how the nuMoM2b insurance variable links the study to state birth and Medicaid data.
Data and privacy
The explorer contains no participant-level data. Predictions are grouped so that every group holds at least 11 readmissions and 11 non-readmissions, and states are described only by state-level summaries, as required by the NICHD data use agreement for nuMoM2b.