Mathematica 12

Any single medication administration, including omission, was observed as the standard event, used also as a common denominator. Each medication administration could contain any number of medication errors. However, an omission (i.e., drug omission) excluded other medication errors since it was impossible to observe. For the purposes of MAE assessment, information from SmPC, factual databases (UpToDate, Micromedex, and Drugbank) and process standards of individual hospitals were used. All steps (data collection, data entry, and evaluation) were performed according to a standardised protocol. All medications were categorised following the ATC classification (WHO Collaborating Centre for Drug Statics Methodology 2022 ).
Data was analysed using Wolfram Mathematica 12 (Wolfram Research Inc., Illinois, USA). The output in the form of quantity and percentage for binomial variables (e.g., the occurrence of a given major MAE/specific MAE/procedural MAE) and in the form of mean ± standard deviation or in the form of median and quartiles for numerical variables (e.g., aggregated quantities of some type of major MAE/specific MAE/procedural MAE) was performed. The sum of relevant medication administration was used as the denominator.
The dependency of either major MAE or specific MAE frequency on the nurse or inpatient characteristics, respectively, was tested using the general linear model. In this model, clustering was considered only at the level of the nurse administering the drugs, while data at higher levels (wards and hospitals) were treated as independent and identically distributed. Results were considered statistically significant if the p‐value was below the Sidak‐adjusted threshold (Lee 2010 ): $${1-\left(1-0.05\right)}^{1/\left(\text{number of major}\mathrm{MAE}+\text{specific}\mathrm{MAE}\right)}$$
For each of the significant dependencies, the impact was evaluated by estimating its effect size using η² and classified as small (η² > 0.01), medium (η² > 0.06), and large (η² > 0.14) according to Cohen's convention (Cohen 1992 (link)).
For the dependence of either major MAE or specific MAE frequency on the type of medication (according to the first level of the ATC class), their real occurrence with the probability distribution corresponding to the frequency of a related administration was compared by the Goodness‐of‐fit test. In this model the Sidak correction was also used.
The dependency of the occurrence of major MAE on procedural MAE was analysed using two methods. First, the generalised linear model with a Bernoulli distribution and logit link function (Fahrmeir and Tutz 1994 (link)) was used for estimation and testing the impact of any procedural MAE per se, assuming the other procedural MAE remains constant. Second, the decision tree model with the CHAID algorithm (Hastie, Friedman, and Tibshirani 2009 ) tested the combinations of procedural MAE with the critical impact on major MAE occurrences. The output in the form of a risk ratio with a 95% confidence interval and the threshold for significance of p‐value < 0.05 were used for both models.