METHODS USED IN THE COMPARISON OF STAFFING MEASURES
To understand how alternative measures of nurse staffing vary, we merged the four datasets to directly compare hospital-level data. We focused our analysis on short-term acute general hospitals, excluding children's and specialty hospitals. We used data from 2000 for most of the analysis, because the CWI survey was a one-time survey conducted in that year. The CalNOC comparison was conducted using data from 2002, because that was the most recent CalNOC data available for matching at the time of the analysis, with the largest number of hospitals providing data.
We began by linking the 2000 OSHPD and AHA datasets, matching 349 hospitals. The AHA-OSHPD comparison focused on total hospital employment, because the AHA survey does not permit disaggregation by type of service.
The OSHPD survey was linked to the CWI survey. The CWI data included two children's hospitals, one specialty hospital, and a federal hospital, which were excluded from this comparison. We thus matched 111 hospitals from the CWI survey with OSHPD data. We limited the OSHPD data to medical–surgical units, to match the type of unit reported in the CWI survey.
Finally, we linked the Year 28 OSHPD data (hospital fiscal years spanning 2001 and 2002) with the 2002 CalNOC data, matching 106 hospitals. Hospitals report unit-by-unit data to CalNOC, which we merged to create totals for types of units. The two categories examined here are medical–surgical units and intensive care units. CalNOC also receives data about step-down units, but OSHPD does not report data for this category. The OSHPD category of “Definitive Observation” was considered as a match to CalNOC's “step-down” definition, but exploratory analyses found large discrepancies in reported patient days and nurse staffing between the OSHPD and CalNOC data. Thus, we excluded step-down units from this analysis.
To compare the datasets, we first calculated equivalent measures of nurse staffing, which are described below for each comparison. We then compared the means and frequency distributions for these measures, using t-tests to determine whether differences were statistically significant. Pearson correlations and Spearman (rank) correlations were computed for each comparison. Key results are presented below. Additional statistics and charts are available in the online supplementary appendix.