The panel unanimously agreed on whether the antimicrobial recommendation was inappropriate or not in 163 (82%) of 200 ASP calls after three rounds of reviews. These 163 calls are the basis for the study analyses (see ). 106 different clinicians made the 163 calls. 48% of the 163 study calls were made by 1st-year housestaff, 24.5% by 2nd-year housestaff, 21% by callers of unknown length of training, 6% by ≥ 3rd-year housestaff, and one (0.5%) by a nurse.
Disposition of ASP Study Telephone Calls
Overall, 42% [95% confidence interval (CI) 34%–50%] of the calls had inappropriate antimicrobial recommendations. Common selected reasons included: antimicrobial drugs not indicated (42%), overly broad spectrum of activity (26%), wrong spectrum of activity (12%), and overly narrow spectrum of activity (8%). The panelists wrote in a reason in 7% of cases, most commonly the need for a formal infectious disease consultation.
Forty-two percent (95% CI 35%–50%) of communications included inaccuracies: 37% in one type of data and 5% in two types of data. The levels of inaccuracies for specific types of data are listed in . Calls made to physician ASP practitioners inside the hospital (i.e., with access to patient data) were inaccurate in 50% of cases vs. 38% if outside, p=0.34. The results were similar for pharmacist ASP practitioners (who had data access while in the hospital and from home; inaccuracy 56% if inside the hospital vs. 35% if outside, p=0.10).
Unadjusted Analysis of Specific Types of Communication Inaccuracies Associated with Inappropriate Antimicrobial Drug Recommendations
Fifty-five percent of calls with inaccurate communication resulted in inappropriate antimicrobial drug recommendations compared to 32% of calls without inaccurate communication (in the unconditional bivariable analyses, p=0.05). Patient age and prior patient location were also significantly associated with the outcome in these analyses (). In bivariable analyses conditioned on the individual ASP practitioner, the OR for the effect of inaccurate communications on inappropriate antimicrobial recommendations was 1.8 (95% CI 0.93–3.6, p=0.08). Day of call, prior patient location, and Charlson comorbidity index were also significantly associated with the outcome in these latter analyses ().
Unadjusted Analysis of Factors Associated with Inappropriate Antimicrobial Drug Recommendations
Conditional Analysis of Factors Associated with Inappropriate Antimicrobial Recommendations
Using multivariable conditional logistic regression to control for confounding, communication inaccuracy was significantly associated with inappropriate antimicrobial recommendations with an OR of 2.2 (95% CI 1.0–4.4, p=0.05). Charlson comorbidity index and prior patient location of intermediate or long-term care center were also significant risk factors ().
The measure of association for the effect of inaccuracy on inappropriateness was consistently greater than or equal to one across all ASP practitioners (p-value of 0.89 for rejecting the null hypothesis of homogeneity). When the primary analysis (bivariable, conditional) was limited to calls in which the panelists were unanimous after the first round of ratings, the studied association increased to an OR of 4.50 (95% confidence interval 1.4–14.3, p=0.01). We also examined the how the pattern of ratings changed for each panelist during the three rounds of ratings (). The kappa for the agreement between the first and final rating was higher for Panelist A vs. Panelists B or C. If the final rating was inappropriate, Panelists B and C were more likely to have changed their rating than Panelist A. However, if the final rating was not inappropriate, then the difference in rates between the panelists for changing ratings was less.
Agreement Between Initial and Final Rating by Panelist*
In bivariable sub-analyses of the effect of inaccuracies of individual types of data, only inaccuracies of microbiological test results and patient temperature were significantly associated with inappropriate antimicrobial recommendations (). Multivariable modeling using patient temperature as a covariate was not possible due to sparse data. A conditional logistic regression model was constructed that included inaccurate communication of microbiological data. After controlling for potential confounding with multivariable modeling, we found that inaccuracies of microbiological data were associated with increased odds of inappropriate antimicrobial drug recommendations (OR of 7.5, 95% CI 2.1–27.0, p=0.002). As with the prior analysis, Charlson comorbidity index and prior patient location of intermediate or long-term care center were also associated with the study outcome in this model ().
Conditional Analysis of Inaccurate Communication of Microbiological Test Results and other Factors Associated with Inappropriate Antimicrobial Recommendations