Statistical analysis

The results from the two control hospitals were combined. Pearson’s chi square tests were used to identify any differences in proportions within the two groups at the two points of measurement. For each questionnaire item, differences in proportions between baseline sample and follow-up sample were calculated using binary logistic regression. We conducted analyses for the intervention hospital and the control hospitals, respectively. Demographic variables with trends for difference (p<0.1) within each group (intervention group; current position, control group; gender and current position, Table 2) were added to the time of measurement variable in the regression model. To compare the odds ratio (OR) in the intervention group to the OR in the control group we used a test of proportion [43] to calculate ratio odds ratios (ROR). A ratio odds ratio tells us whether there are any differences between the changes in two different groups. A test of proportions is used to compare two estimated quantities, such as means or proportions (in this case odds ratios) of the same quantity in two independent samples. All statistical analysis was done with SPSS 21.0 for Windows (IBM Corp., Armunk, NY) with a significance level of 5% (p<0.05).

Numbers are percentages of total N for each sample unless otherwise stated.

* The N in the four samples varied for each question due to missing answers on the variables (0.8%–2.5%).

^† P-value calculated using Pearson’s chi square.

^‡ P-value calculated using independent samples t-test.