Nurse Staffing and Patient Outcomes: Analyzing Within- and Between-Variation

Abstract

Objectives: To study and compare the longitudinal and cross-sectional relationship between nurse hours perpatient day and patient outcomes (30‐day mortality and length of stay [LOS]). Data source: Retrospective administrative register data (2015-2017) with all hospital admissions, LOS, andmortality rates from five medical departments combined with monthly data on staffing levels of registerednurses, physicians, and nurse assistants from the hospital’s payroll systems, as well as detailed patient-levelmorbidity and sociodemographic characteristics. Study design: We used a flexible within‐between random effect (REWB) model to exploit longitudinal andcross-sectional variation among homogenous medical departments. We applied a rich patient‐level dataset, leaving little risk of omitted variable bias due to patient‐level heterogeneity. Data Collection: The study population covered all hospital inpatient discharges from five medical departments over the period 2015-17 (N=172,132). Hospital payroll data were merged using hospital department identification codes. Principal findings: For both outcomes, we found evidence of endogeneity in within estimates when failing to control for patient heterogeneity. When controlling for patient characteristics, we found that a greater nurse to-patient ratio was associated with a statistically significant decrease in LOS when using both within- and between‐department variations. However, only between estimates were significant for nurses when it came to mortality, whereas the significance of the within estimate was absorbed by physicians. Conclusions: Most longitudinal studies apply fixed effects and, hence, only assess within variations. We found that between estimates were higher in magnitude and were more robust to omitted variable bias than within estimates. Therefore, as between variations are likely to identify structural recruitment problems, we argue for the importance of studying between estimators as well as in longitudinal studies.