Provider exclusions are rising — and clustering around distressed operators

The Office of Inspector General's List of Excluded Individuals and Entities (LEIE) is the federal registry of providers barred from participating in Medicare, Medicaid, and other federal health programs. It is a punitive, trailing record — an exclusion lands long after the conduct that prompted it. That lag is precisely what makes the list useful as a cohort signal rather than a case signal.

Through April 2026, new additions to the LEIE grew 11% against the same period in 2025. The raw count is not the story; exclusion volume fluctuates with enforcement capacity and statutory deadlines. The structure is the story.

Joining exclusions to enrollment

We linked each new exclusion to its PECOS enrollment record where one existed, then to the cost-report markers we use as a rough proxy for financial distress — negative operating margin across two consecutive cost-report periods, and a current ratio below one. The match rate was 63%; the unmatched remainder skews toward individual practitioners who never carried an institutional enrollment.

Among the matched, institutional exclusions were 2.4 times more likely to be attached to an operator already flagged as distressed than to a financially stable one. The direction is intuitive — distress and misconduct share common upstream causes — but the magnitude is larger than we expected, and it has widened since 2024.

Why the clustering matters

If exclusions were randomly distributed across the provider base, the list would be a poor early-warning instrument — by the time a provider is excluded, the damage is done and the signal is spent. But because new exclusions cluster around operators that the cost reports had already flagged, the two datasets together do something neither does alone: the cost report flags the cohort early, and the exclusion list later confirms which members of that cohort actually failed.

This is the same pattern we found in the SNF quality study: the public, frozen-snapshot data carries a leading signal that the official, trailing designation only later ratifies.

The moat is not any single dataset. It is the join — the cost report flags the cohort early, and the exclusion list later confirms which members of that cohort actually failed.

The corollary is that the predictive value decays the moment either source is read alone. The distress proxy on its own over-flags: plenty of thin-margin operators never cross into misconduct. The exclusion list on its own is too late to act on. Read together, across frozen monthly snapshots, the false positives from the first source are pruned by the confirmations in the second, and what remains is a short, defensible watch-list rather than a long, speculative one.

Reproducibility

The exclusion snapshot used here is oig-leie/2026-05. The PECOS join key is the NPI; the distress proxy is computed from HCRIS and its construction is published alongside the composite methodology at /quality/methodology. Every count in this study resolves to a specific row in a specific frozen snapshot. No abstracted claims, no smoothed-away facilities.

We will refresh this analysis when the next LEIE snapshot lands and append the revised figures rather than overwrite these.