Midnight census revisited: Reliability of patient day measurements in US hospital units

BackgroundPatient days are widely used in nurse staffing research and for nursing quality measurement. Nursing hours per patient day (NHPPD) and fall rates incorporate patient days in the denominator and are endorsed by the US National Quality Forum (NQF) as nursing sensitive consensus measures. Measurement error introduced by patient days would affect the accuracy of these nursing quality indicators.ObjectivesThe aim of this study was to assess the reliability of five patient day reporting methods accepted by the National Database of Nursing Quality Indicators (NDNQI). The specific aims were (1) to investigate the agreement of five patient day measurements with a defined quasi-gold standard, (2) to explore method bias by investigating the association of potential confounding variables with the differences between the routine measurements and the quasi-gold standard, and (3) to extrapolate the potential effect of bias of the patient day methods on nursing quality indicators.DesignA multiple census study with a national convenience sample of hospital units in the US was conducted.Setting260 out of 282 units (92%) from 54 hospitals sent bi-hourly patient census data for seven randomly selected days in September 2008.MethodsThe multiple census data comprised the quasi-gold standard and was compared with data routinely submitted to the database. Intraclass correlations were calculated for an agreement analysis. A Bayesian regression analysis was conducted to explore the impact of different data collection methods and the degree of short stay patients.ResultsOverall agreement between routine data and the quasi-gold standard was excellent (ICC [95% CI]: 0.967 [0.958–0.974]). A Bayesian regression analysis identified that two methods underestimated patient days and an interaction between the degrees of short stay patients and one of the data collection methods also affected patient day measurement by up to 7.6%.