1058 — Reconciliation of Different Estimates of the Impact of Nurse Staffing on Patient Outcomes: The Effect of Data Aggregation and Estimation Methods
Phibbs CP, HERC & Stanford University; Bartel AP, Columbia University; de Cordova PB, Rutgers University; Needleman J, UCLA; Schmitt SK, HERC; Stone PW, Columbia University;
Evidence shows that nurse staffing affects patient outcomes. There are large differences in the estimated effects; including some findings with opposite signs. We used a large, longitudinal, patient-level dataset to examine if different levels of data aggregation and/or statistical methods could explain these differences.
: Monthly staffing, by nurse type (RN, LPN, aides, contract) were obtained from all VA acute care units (including ICUs) for 2003-2013. Approximately 4,000,000 discharges from > 600 units. Data were integrated from payroll (PAID), discharge (PTF) and financial (DSS NDEs) sources. Length of stay (LOS) was used as the dependent variable as it captures the combined effect of multiple nursing-sensitive patient outcomes. Patient-level regressions controlled for patient characteristics, nurse staffing, experience, and unit tenure, and were estimated using both ordinary least squares (OLS) and fixed-effects (FE), with different levels of aggregation for the nursing variables (unit-month, unit-year, hospital-month, hospital-year), for all units together, and separately for acute care and ICUs.
: Results were sensitive to estimation method and aggregation. The point estimates of the effects of nurse staffing on LOS of switching from monthly to annual staffing data ranged from 14-117% for the FE models and 13-276% for the OLS models, with some sign reversals. These ranges were even larger across all levels of aggregation. For the same level of aggregation, the difference between the OLS and FE estimates ranged from 0-304% with two cases of sign reversal.
The magnitude and even the direction of the effects of different elements of nurse staffing on patient outcomes are sensitive aggregation level and estimation method. The differences we observed probably explain much of the heterogeneity of findings across studies.
Interpretation of the results of studies of nurse staffing on patient outcomes needs to account for the level of data aggregation and the statistical methods used. Higher aggregation levels, both across time and across units, probably masks effects. Thus, studies that measure nurse staffing at the unit-level data with shorter time intervals yield more reliable estimates. Studies that fail to account for unobserved heterogeneity are probably biased. These factors need to be carefully considered when using research to make policy.