Methodology & limitations

Percentile normalization

Each measure is winsorized at the 2.5th / 97.5th percentiles within a snapshot, then ranked to a 0–100 percentile scale. Lower-is-better measures (e.g. rehospitalization, abuse share, turnover) are inverted so that higher domain scores always mean better relative performance.

Six domains

• Acute. Short-stay rehospitalization and ED use; long-stay hospitalization / ED rates; 30-day potentially preventable readmissions.
• Rehab. Short-stay residents at or above expected ability to care for themselves at discharge.
• Safety. Abuse-icon share, falls with major injury, pressure ulcers, physical restraints.
• Staff. Nurse and RN hours per resident day; nurse and RN turnover.
• Regulatory. Average fines ($) and average payment denials.
• Stars. Overall, health inspection, staffing, and quality 5-star averages.

Buyer domain weights

Within-domain emphasis also differs by buyer (e.g. MA doubles rehospitalization and abuse share). Domain weights below sum to 1.0 per lens.

Tier thresholds

• PreferredComposite ≥ 70
• Acceptable50 – 69
• Watch30 – 49
• ExcludeComposite < 30
• Exclude (safety)Safety gate triggered (overrides composite)
• Insufficient dataNo scorable composite

Safety gate

Panel trajectories

Chains present in at least five of the six monthly snapshots enter a balanced panel. An OLS fit of MA score on snapshot order yields slope, net change, and R². Classes: Deteriorating (slope ≤ −0.6 and net ≤ −2), Improving (slope ≥ 0.6 and net ≥ 2), Volatile (high period-to-period volatility with small net), otherwise Stable.

Limitations

• PECOS ownership graphs are imperfect — chain assignment can lag ownership changes and miss complex holding structures.
• Chain averages hide bad buildings. A Preferred chain can still contain facilities that would fail a building-level screen.
• This product is informational, not regulatory. It does not replace survey, compliance, or clinical judgment.
• Several staffing measures refresh on a slower cadence than monthly chain files, which limits leading-indicator tests on turnover.
• Archetypes are descriptive k-means clusters on domain scores, not causal types or CMS designations.