Outcome evaluation
The primary objective of this study is to examine the change of the antibiotic prescription rate in targeted patients within three intervention arms and the comparison between the three intervention arms. The confirmatory analysis of the primary endpoint will be conducted on the basis of the Intention-To-Treat (ITT) population where all patients will be included in the analysis and assigned to the group they were randomised to.
To account for multiple testing and to assure a global significance level of 5%, a multiple test procedure for hierarchically ordered hypotheses will be applied. In the first stage, differences regarding the primary endpoint (prescription of antibiotics) before and after the intervention in intervention groups B and C are tested. To assure a significance level of 5% within the first level, the hypotheses for the pre- and post comparison will be tested at a local significance level of 5/2%. The second step includes the pre-and post comparison in intervention group A. The significance level of the rejected hypotheses from step one is taken (2.5% if only one hypothesis can be rejected, 5% if both hypotheses can be rejected). If significance is reached in intervention groups A and B, differences between groups A and B will be tested at a local significance level of 5/2%. Similarly, difference between groups A and C will be only tested if significance is reached for groups A and C in the steps 1 and 2. If the hypotheses in step 1 or 2 cannot be rejected, the group comparison will not be tested primarily.
In all stages described above, a logistic mixed effects model will be applied to assess the respective hypotheses regarding the primary endpoint. In the first and second stage, the time (before/after the intervention) is included as fixed effect. A random intercept will be included for patients and for practices. For the comparisons in the third stage, the intervention group will be included as fixed effect additionally. Furthermore, age and gender will be included as covariates. Finally, sensitivity analyses will be conducted using different populations (per protocol population where patients with major protocol violations are excluded and appropriate subgroups). Since all these analyses are based on claims data, missing values cannot be tracked.
Descriptive methods will be used for the analysis of the secondary outcomes, including the calculation of appropriate summary measures of the empirical distribution (mean, standard deviation, median, IQR, minimum and maximum for continuous variables and frequency in percentages for categorical variables). In addition, similar mixed models as described for the primary endpoint will be used. All calculated p values regarding secondary outcomes will have descriptive character only.
Furthermore, differences between the intervention groups and an untreated comparison group will be tested. The comparison group consists of claims data of patients insured by AOK health insurance in non-participating practices in Bavaria and North Rhine-Westphalia. Therefore, propensity score matching will be used to generate 14 virtual clusters of non-participating practices.
A detailed statistical analysis plan is written prior to the final analysis. SAS (SAS Institute Inc., Cary, NC, USA), version 9.4 or higher, is used to carry out the analyses.