Demystifying Quality Metrics and Unveiling the True Measure of Quality of Care in Nursing Homes: Mixed Effects Analysis

Background: The 5-Star Quality Rating System for nursing homes plays a central role in evaluating quality of care, although it has both strengths and limitations. This system relies heavily on the Minimum Data Set and derives several quality measures (QMs) from it. In this study, we validated the effectiveness of the 5-Star Quality Rating System for nursing homes and its underlying QMs in estimating quality of care. We constructed a panel dataset of US nursing homes (n=15,416) active from May 2020 to June 2023, retrieving data from three major sources: (1) COVID-19 nursing home data, (2) Payroll-Based Journal data, and (3) nursing home QM snapshots. The outcome variables included (1) resident infection, (2) staff infection, or (3) resident and staff deaths. The predictor variables were the 5-Star Quality Rating System for nursing homes and its underlying QMs classified as structure, process, or outcome (SPO) QMs. Objective: This study aims to evaluate the effectiveness of nursing home QMs by regressing nursing home COVID-19 outcomes on nursing home QMs, classified using the Donabedian SPO framework. We hypothesized that nursing homes with better structural quality (eg, greater staff availability, better skill mix, and so on), better process quality (eg, lower restraint use and higher vaccination rates), and better outcome quality (eg, lower number of residents with pressure ulcers and a lower number of resident falls) experienced better COVID-19 performance in terms of resident and staff infections and deaths. Methods: To examine the association between the COVID-19 outcomes and SPO QMs, we imputed missing values in the dataset using random forest. Subsequently, we modeled the imputed dataset using hurdled zero-inflated negative binomial mixed effects models. The zero inflation model included factors influencing initial susceptibility to COVID-19 or factors influencing the possibility of death after COVID-19 had been contracted. The model estimates were conditioned on zero inflation and random effects. Results: Staffing measures (P