Can National Administrative Data Identify Optimal Nurse Staffing? Evidence of Estimand Instability and Identification Challenges in CMS Home Health Panel Data

<title>Abstract</title> <p>Objective. To determine whether national administrative panel data permit reliable identification of the staffing-quality relationship assumed by production-function models of optimal nurse staffing in home health care. Data Sources and Study Setting. CMS Home Health Compare 2016–2018 merged panel (final analytic sample: 16,828 agency-years after multi-stage data quality screening). Quality monetization analyses used a FY2017–2018 subpanel (n ≈ 10,496). Study Design. Feasibility and identification study applying three estimators—within-agency (CRE-equivalent), pooled, and between-agency—to assess dose–quality and dose–cost relationships; nonparametric validation via Bayesian Neural Network (BNN) and equivalence testing (TOST). Data Extraction Methods. Dose: RN FTE per 100 skilled-nursing patients. Quality: T-standardized composite of four CMS measures (Average 4Ts). Cost: skilled nursing labor costs per patient. All models used CCN-clustered SEs, state/year fixed effects, and Correlated Random Effects (Mundlak) terms. Principal Findings. Cost (C(N)) rose robustly across all estimators (β = +212 to + 447/patient, p &lt; 0.001). Quality (Q(N)) showed estimand instability: within-agency yielded a null result (β = +0.014, p = 0.50) that reversed sign under leverage removal; pooled and between-agency estimators showed significant negative associations (β ≈ −0.036 to − 0.037, p = 0.02–0.03) with unresolved confounding. TOST equivalence testing confirmed the quality effect is negligible relative to clinically meaningful benchmarks (p ≈ 4 × 10⁻¹⁴⁰). Break-even analysis shows the observed quality response upper bound is approximately 365× below the cost-offset threshold. Conclusions. CMS home health administrative panel data, as currently structured, face substantial identification challenges for estimating staffing-quality optima. Estimator choice changes the sign and significance of the quality-staffing relationship, and no available specification resolves the underlying endogeneity. These findings do not imply that nurse staffing is inconsequential; rather, they delineate the data conditions and study designs that future identification efforts require.</p>