healthcare risk · payer analytics
MEPS Healthcare Risk Stratification
Prospective models that rank people by next-year medical cost and prescription burden, built for limited outreach capacity and audited by subgroup.
The question
Can prior-year survey, utilization, expenditure, insurance, access, chronic-condition and prescription information identify people at elevated risk of high medical spending or high prescription out-of-pocket burden in the following year?
The data
Medical Expenditure Panel Survey (MEPS) Panel 24 longitudinal records and Prescribed Medicines event files. Features from 2021 predict outcomes in 2022 for the same people, which avoids the temporal leakage in a same-year baseline. The weighted targets are next-year medical expenditure at or above the 90th percentile (~$15,256) and next-year prescription self/family payment at or above the 90th percentile (~$517).
How it was done
- Preserve MEPS survey weights in descriptive estimates and model evaluation.
- Compare logistic regression, random forest, and gradient boosting.
- Evaluate discrimination, calibration, precision, recall, and lift at fixed outreach capacity.
- Audit performance by age, income, insurance, race or ethnicity, and chronic-condition burden.
- Treat thresholds as resource-allocation choices rather than defaulting to 0.50.
What the evidence showed
| Target | Best model | ROC-AUC | PR-AUC | Precision | Recall |
|---|---|---|---|---|---|
| Next-year high medical expenditure | Random forest | 0.835 | 0.611 | 0.464 | 0.667 |
| Next-year high prescription burden | Random forest | 0.883 | 0.616 | 0.490 | 0.740 |
| Target | Outreach share | Capture rate | Precision |
|---|---|---|---|
| High medical expenditure | 10% | 39.3% | 74.6% |
| High medical expenditure | 20% | 57.6% | 55.2% |
| High prescription burden | 10% | 47.5% | 66.4% |
| High prescription burden | 20% | 71.9% | 52.2% |
What I recommend
- Use the prospective scores to prioritize care-management, affordability, or pharmacy-navigation outreach when capacity is limited — at 10% capacity the ranking reaches 39.3% of high-cost members at 74.6% precision.
- Do not use these scores for coverage denial, pricing, restricting care, or clinical decision-making.
- Treat the threshold as a capacity decision made with the operating team, not a modeling default: moving from 10% to 20% outreach raises capture to 57.6% but drops precision to 55.2%.
- Performance is not uniform across groups. Calibration and recall vary by age, income, insurance, race or ethnicity, and chronic-condition burden, and several subgroups are too small for stable estimates. Group-specific thresholds are offered as sensitivity analysis, not as an approved fairness policy.
What this does not prove
- MEPS is nationally representative survey data, not a production claims feed.
- The longitudinal cohort is smaller than the cross-sectional full-year sample.
- Prescription features summarize events and therapeutic classes rather than individual drug identities.
- Fixed-threshold performance may shift across populations and years.
- Small subgroup estimates require cautious interpretation.
- Predictive performance does not establish that outreach improves health or financial outcomes.