COMPARING SINGLE & MULTILEVEL RESULTS FOR PATIENTS NESTED IN HOSPITAL UNITS

Purpose: For meaningful statistical and clinical results, this presentation demonstrates the need to account for multilevel or nested effects of patient risk and hospital unit factors on the mediator (patient complexity) and the outcome (actual length of stay from expected).

Rationale: Reports from the Institute of Medicine cite the emerging body of literature that links patient quality and safety directly to nursing care. However, systematic literature reviews show that methods used and outcomes found in these studies are often incomplete or misleading. Studies using new and sophisticated analytic methods help clarify and evaluate the complex effects of nursing on patients nested in hospital units.

Methods:  Following IRB approvals, this study examined two 10% random samples by unit after dividing 60,156 inpatients from 38 units in 8 hospitals into two independent groups to form base (n=5,987) and cross-validation (n=5,896) samples. These samples included calendar year 2012 inpatients admitted to medical, surgical, critical care, pediatric, and perinatal units. The study utilized mediated multilevel latent path analyses and Bayesian estimation to examine actual length of stay (LOS) from expected as affected by patient risk factors (table below) and unit characteristics of unit staff quality (RN education, certification, experience), unit staff quantity (RN direct HPPD, RN to patient ratio, skill mix), unit culture (NDNQI RN Satisfaction survey), perceived workload (unit complexity and NDNQI RN satisfaction surveys), and unit churn (patient admits, discharges, transfers).

Results: The table below illustrates a small but key subset of results that compares the outcomes from patients aggregated into a single (within) level and also nested into their unit (between) level. Without accounting for nested effects, the researcher might falsely determine that patient complexity as a mediator had a slight but statistically significant effect on length of stay (LOS) and that only surgical admits demonstrated both a statistical and clinical significant effect on patient risk.  By accounting for multilevel effects, the researcher may conclude that the patient risk factors of admits via ED, surgery admits, readmits within 30 days, and age at admit significantly influence patient risk.

Implications: Using sophisticated analytical methods, scientists in education, social studies, and medicine account for nested effects to help avoid mixed and inaccurate results. To meaningfully evaluate outcomes, nursing scientists must further explore multilevel methods to account not only for patients nested in units, clinics, and hospitals, but also for students in classes, individuals in families, and clients by healthcare providers.