The demographic details of the patients in each of the three time periods are presented in Table . During the computerised decision support period (CDSS), patients were generally older than those in the other two time periods (a greater proportion were aged >85 years), and less likely to have received antibiotic therapy prior to presentation. The observed death rate during the CDSS period appeared to be higher than for the other two periods, but this was largely explained by differences in the proportion of patients aged over 85 years, and differences in the number of patients who died in the ED for whom supportive therapy was not thought appropriate
Table details the comparisons in prescribing behaviour over the three time periods. The odds ratio for having received the recommended empiric antibiotic therapy to cover both typical and atypical pathogens ('concordant therapy') in the ED for the academic detailing period compared to the baseline period was 2.58 [1.78, 3.73], p < 0.01, and after adjustment for age, severity (PSI class) and suspicion of aspiration, OR = 2.79 [1.88, 4.14], p < 0.01. The odds ratio for concordant therapy in the computerised decision support period compared to the academic detailing period was 2.03 [1.13, 3.66], p = 0.01, and after adjustment for age, severity and aspiration, OR = 1.99 [1.07, 3.69], p = 0.02. The estimated effect over time within each cohort did not appear to be substantially altered by the inclusion of these covariates.
The effect of change over time was observed in more detail. Figure illustrates the percentage of empiric antibiotic prescriptions that were concordant with recommendations per month over the entire period. Prescribing patterns improved slowly over time. One year after release of the guideline, in the absence of any promotional efforts, (that is, at the end of the baseline period), the concordance rate was around 60%. The change in the proportion of concordant prescribing between the last month of the baseline period and the first month of the academic detailing period was +10.8% over 12 months. The change in the proportion of concordant prescribing between the last month of the academic detailing period and the first month of the computerised decision support period was +21.5% over 5 months. At the end of the study period, the rate of concordant prescribing was high. The first month post the CDSS intervention had a very high concordance rate (100%) and thereafter the rate remained around 90%, although the study was not long enough to demonstrate whether this level was maintained beyond 6 months.
Percentage of empiric antibiotics prescribed that were concordant with recommendations per month.
Further analysis was performed to compare the observed results with that which would be expected based upon an underlying trend in improvement over time [11
]. The observed behaviour in the preceding time periods (over 3 years) were used to predict the expected prescribing behaviour in the latter 6 month period of the study. Figure shows the three regression lines that best fit the observed rate of concordance over the three separate time periods, and the concordance predicted from a logistic regression model based upon the first and second time periods extrapolated forward through the third time period (the 'expected' concordance). While it is important to note that such a regression line may be sensitive to outliers, there were in fact few actual outliers in these actual data and the likelihood of effect would be low.
Figure 2 Proportion of concordant therapy prescribed over time. The solid lines indicate regression lines that best fit the observed data in each of the three time periods, demonstrating the percentage of empiric antibiotic therapy that was concordant with recommendations (more ...)
During the first six months of the CDSS period, the proportion of patients who were prescribed concordant therapy was greater than would be expected based on the observed trend. A confidence interval around the trend line was determined, and this described the likelihood of the observed results in the first month of the CDSS period as having a p value of 0.06 based on the existing trend alone.
Secondary outcomes were analysed as a measure of the impact of the changes in prescribing on key areas of interest. Regarding those patients who required ICU support, the likelihood that recommended broad spectrum empiric antibiotics were received in the ED increased over time. The odds ratio for the academic detailing period compared to the baseline period was 1.48 [0.35,6.25], p = 0.59, and the odds ratio for the CDSS period compared to the AD period was 10.80 [0.99, 116.99], p = 0.051. Improved early recognition of patients with severe illness was suggested, with a greater proportion of patients requiring ICU care going directly to ICU from the ED over time (65% in baseline, 75% in the academic detailing period and 80% in the period of computerised decision support). The number of patients was too small to comment upon whether this change was statistically significant.
There appeared to be a lower likelihood of inappropriately prescribing an antibiotic to a patient who had a documented allergy to that drug during the computerised decision support period. Specifically, comparing the AD period with the baseline, the odds of an allergy prescribing error were 0.99 [0.31, 3.16], p = 0.99; whereas when the comparing the CDSS with the AD period, the odds ratio for such a prescribing error was 0.47 [0.10, 2.19], p = 0.33.
Table describes the most frequent antibiotic combinations prescribed to patients in each of the three time periods. The percentage of patients empirically prescribed a cephalosporin was 38.2% for the baseline period, 38.1% for the academic detailing period, and 42.8% for the computerised decision support period. The average cost of antibiotic therapy per patient was calculated for the three patient groups. This calculation was adjusted for changes in pricing over time, though in fact, very little change occurred in the price of the antibiotics most frequently prescribed for CAP during the study. While the average cost per patient increased between the first and second time periods, it fell in the third time period. Finally, the time between a patient being admitted to the emergency department, and an antibiotic being first administered to the patient did not increase, and was actually found to progressively fall over the three time periods, from 171 to 158 and then 142 minutes, p < 0.01.
The most frequent initial antibiotic combinations prescribed (described as percentage of patients)