Study design and data Collection.
This retrospective study was carried out in a tertiary hospital in Jiangxi Province, southeast of China. Participants of the study were inpatients with diagnosis of infection. The AST report, antimicrobials prescribed after reporting AST results and related inpatient information of participants were collected. Firstly, The AST reports of the inpatients were collected according to the following conditions: 1. The report time was between 1 January 2019 and 30 June 2021. 2. The bacteria for detection were extracted from a sterile-site specimen, which was more likely to be the causative agent of infection, and mentioned in Clinical and Laboratory Standards Institute (CLSI) M100 Table 2. 3. The AST report firstly gave interpretive categories (ie. sensitive, intermediary or resistance) during this hospitalization. Secondly, antimicrobials prescribed after reporting AST results were collected according to the following conditions: 1. If there was only one AST report during the hospitalization, the antimicrobials prescribed after reporting AST results were collected. 2. If there were equal to or greater than two AST reports and the time interval between the first and second AST reports was equal to or greater than 7 days, the antimicrobials prescribing between the first and second AST reports were collected. 3. If the time interval between the first and second AST reports was less than 7 days, the antimicrobials prescribed within 7 days after the first AST report were collected. All the information was exported from hospital electronic information system. A total of 709 cases were enrolled according to the above conditions.
Identification to the species level and AST of all isolates were performed using the Vitek 2 system (bioMérieux, Marcy I’Etoile, France) at the department of clinical laboratory in the Hospital. The protocols for identification and susceptibility testing of isolates in China were the same as those followed by the Central Laboratory of International Health Management Association[28]. AST was performed by disk diffusion method and broth microdilution method in accordance with the CLSI recommendations.
Variables and Measurement.
Firstly, IQ indicators were established through reference to relevant literature, which contained an overall indicator of IQ namely the total IQ and four sub IQ namely completeness, usefulness, accuracy and consistency[17, 20, 21, 24, 29] (All indicators were presented in Appendix I). The evaluation of the total IQ and four sub IQ was based on the specifications, including CLSI M100 and the related standardization of AST report in China which combined requirements in CLSI M100 and China’s situation[30–33]. Basis for inclusion, definition and calculation formula were as follows:
The total IQ. To evaluate the overall situation of the IQ of AST report, the total IQ was defined as the degree to which the information of AST report meeting the requirements of all criteria in the four sub IQ. Simple ratio was used as the function forms to assess the level of the total IQ and four sub IQ according to Batini C’s research[22]. The calculation formula of the total IQ was as follow: The total IQ = The sum of information fulfilled the criteria in four sub IQ / The sum of information involved in the judgement of four sub IQ.
Completeness. Completeness was chosen because it was one of the basic dimensions according to the research of Batini C. et al.[21], and incompleteness was a prominent problem in AST report such as incomplete annotation information[15, 17, 21]. Completeness was defined as the degree to which the information that guidelines recommended AST reporting was given in AST report, with reference to Cario B’s definition, that was the degree to which a given data collection included data describing the corresponding set of real-world objects[21]. In this study, “information that guidelines recommended AST reporting” corresponded to “data describing the corresponding set of real-world objects”. For example, according to the guideline requirements, AST report should contain interpretation comments such as the annotation of symbols and abbreviation, interpretation of microbiology results and recommendations of antimicrobial prescriptions[30, 33]. The information which guideline recommended AST reporting would be seen as complete information if given in the AST report. The calculation formula of completeness was as follows: Completeness = The amount of information given in AST report which guidelines recommended AST reporting / The amount of information which guidelines recommended AST reporting.
Usefulness. Usefulness was chosen because improving IQ was able to reduce information overload, and in turn, help users better understand and make more informed choices according to previous researches[13, 14, 34], and researches have pointed out that AST report existed useless information such as reporting nitrofurantoin for bloodstream infection[18]. Usefulness was defined as the degree to which the information given in the AST report was in the range of the information recommended by guidelines, because guidelines have regulated the useful information in AST report[30, 35]. For example, the antimicrobials in AST report were useful information if listed in group A of CLSI M100, while daptomycin given in the AST report of respiratory specimen was useless information based on the guidelines[33]. The calculation formula of usefulness was as follows: Usefulness = The amount of information given in AST report which guidelines recommended AST reporting / The amount of information given in AST report.
Accuracy. Accuracy was chosen because it was one of the basic dimensions according to the research of Batini C. et al., and inaccuracy was a prominent problem in AST report such as the errors of interpretive categories and bacterial identification[18, 21]. Accuracy was defined as the degree to which the information given in AST report was correct according to the guidelines, with reference to Wang’s definition, that was the extent to which data were correct, reliable and certified[24]. In this study, the criteria to judge the accuracy of the information given in AST report was the standard information in the guidelines. For example, guidelines gave the standards of antimicrobial’s generic name, susceptibility testing methods and breakpoints. The information in AST report was accurate information if the same as the standard information in guidelines. The calculation formula of accuracy was as follows: Accuracy = The amount of accurate information / The amount of information given in AST report.
Consistency. Consistency was chosen because it was one of the basic dimensions according to the research of Batini C. et al.[21], and it was the guideline recommendation to assess the consistency of information in AST reports[21, 33]. Consistency was defined as the internal consistency between the information of AST report with reference to the definition of Batini C. et al.[21], such as the consistency of the results from individual agents within a specific drug class and the established hierarchy of activity rules. The calculation formula of consistency was as follows: Consistency = The amount of consistent information / The information should be consistent according to the guideline recommendations.
Secondly, antimicrobial adherence was used to measure the rational antimicrobial use based on the interpretive categories of AST report. Medication adherence was generally defined as the extent to which the patient’s actual history of drug administration corresponds to the prescribed regimen[36]. In this study, it was the evaluation of antimicrobial prescription behavior, thus antimicrobial adherence was defined as the antimicrobial prescribed was effective against isolated bacteria with reference to the interpretive categories of AST report, which included direct and indirect situations. Direct adherence was defined as using antimicrobials to which the bacterium was susceptible, or using antimicrobials to which the bacterium was intermediate without reporting susceptible results. Indirect adherence was defined as using antimicrobials that did not be reported but the susceptible result was able to be inferred according to equivalent agents in CLSI M100, or using antimicrobials that did not be reported but the intermediate result was able to be inferred according to equivalent agents in CLSI M100 without reporting susceptible and referred susceptible results (The summary of equivalent agents in CLSI M100 was presented in Appendix II.)[33]. The adherence rate was used to measure the degree of antimicrobial adherence. The calculation formula was as follows: The adherence rate = The number of direct and indirect adherence antimicrobial prescriptions after reporting AST results / The number of antimicrobial prescriptions after reporting AST results.
Data analysis.
In the analysis of the impact of IQ of AST report on the rational antimicrobial use, the total IQ and four sub IQ were independent variables, and the adherence rate was dependent variable. Considering the dependent variable defined and observed only on the standard unit interval, 0 ≤ y ≤ 1, fractional logit regression model (FLRM) was chosen because it was more common and easily applied alternative that would yield more precise and trustworthy research results[37]. The functional form for the conditional mean of the fractional outcome was: \(\text{E}\left(\text{y}\right|\text{X}) = \text{G}(\text{X}{\beta })\), where the nonlinear function \(\text{G}\left( \right)\). ensured that predictions lied inside the unit interval. RESET test, which was proposed by Papke and Wooldridge (1996), was used to detect whether the model had specification problems, and P-values > 0.10 indicated that the model was appropriate specified at 10%[38]. In nonlinear models, the magnitude of the change in the dependent variable caused by a unit change in the predictor varied with the starting level of the latter. The procedure for the estimation of average partial effects (APE) within the FLRM was particularly appealing due to the fact that its identification required no assumptions in terms of serial dependence in the dependent variable of the economic magnitude of the relations of interest[39]. The interpretation of APE’s was similar to that of linear regression coefficients[40].
We added some control variables to excluded other possible impact factors of antimicrobials use. Age and sex were used to control the impact of patient characteristics. The Age-adjusted Charlson Comorbidity Indicator (ACCI) was used to control the impact of disease severity[41]. Average Administration Time (AAT) and Total Numbers of Antimicrobials (TNA) were used to control the impact of length and number of prescriptions[42].
Two FLRM equations were built to evaluate the impact of the total IQ and four sub IQ separately, which were as follows:
$$\text{y}\left(\text{t}\text{h}\text{e} \text{a}\text{d}\text{h}\text{e}\text{r}\text{e}\text{n}\text{c}\text{e} \text{r}\text{a}\text{t}\text{e}\right)= \text{G}({\beta }1 + {\beta }2(\text{t}\text{h}\text{e} \text{t}\text{o}\text{t}\text{a}\text{l} \text{I}\text{Q}) + {\beta }3(\text{A}\text{g}\text{e}) + {\beta }4(\text{S}\text{e}\text{x}) + {\beta }5(\text{A}\text{C}\text{C}\text{I})+ {\beta }6(\text{A}\text{A}\text{T}) + {\beta }7(\text{T}\text{N}\text{A}\left)\right) \left(1\right)$$
$$\text{y}\left(\text{t}\text{h}\text{e} \text{a}\text{d}\text{h}\text{e}\text{r}\text{e}\text{n}\text{c}\text{e} \text{r}\text{a}\text{t}\text{e}\right) = \text{G}({\beta }1 + {\beta }21(\text{c}\text{o}\text{m}\text{p}\text{l}\text{e}\text{t}\text{e}\text{n}\text{e}\text{s}\text{s})+ {\beta }22(\text{u}\text{s}\text{e}\text{f}\text{u}\text{l}\text{n}\text{e}\text{s}\text{s}) + {\beta }23(\text{a}\text{c}\text{c}\text{u}\text{r}\text{a}\text{c}\text{y})+ {\beta }24(\text{c}\text{o}\text{n}\text{s}\text{i}\text{s}\text{t}\text{e}\text{n}\text{c}\text{y})+ {\beta }3(\text{A}\text{g}\text{e}) + {\beta }4(\text{S}\text{e}\text{x}) + {\beta }5(\text{A}\text{C}\text{C}\text{I})+ {\beta }6(\text{A}\text{A}\text{T})+ {\beta }7(\text{T}\text{N}\text{A}\left)\right) \left(2\right)$$
The computation of the IQ of AST report and antimicrobial adherence were implemented in Python programming language using PyCharm 2022 community edition. Eq. (1) and Eq. (2) was modeled using the “fractional logit regression” package in R (The Python code was presented in Appendix III).
This study included four datasets, which was AST reports, the IQ of AST, antimicrobial adherence and FLRM. The datasets were attached in the Appendix IV to Appendix VII.
Ethics Statement
The Ethics Committee of Tongji Medical College, Huazhong University of Science and Technology approved the study. Any information that could identify participants was guaranteed confidentiality.