Artificial intelligence to improve antimicrobial prescribing: A protocol for a systematic review

Research question

Can the use of AI improve antimicrobial prescribing in human patients?

Study design

The selected studies will be observational in nature, including cohort studies; and any study that examines the potential or actual use of AI, or machine learning, or data analytics to improve the antimicrobial prescriptions or use in human patients will be included in this review.

Eligibility criteria

Types of participants: Participants will be human patients, and there will be no restrictions on age, gender, weight, or morbidities.

Settings: There will be no restriction on the setting of the studies.

Time: There will be no restriction on the time of publication of the studies included.

Language: There will be no restriction on the language of the articles included.

Data sources and search strategy

Data sources: A search strategy will be developed and applied to the following databases: ScienceDirect, Scopus, OVID, EMBASE, and Web of Science. Since part of the studies will have been carried out through a computer science lens, ACM Digital library, and IEEE Xplore databases will also be searched to collect as many relevant publications as possible.

Search strategy: The Search for keywords will be used across three different concepts: 1. Artificial Intelligence, 2. Prescriptions, 3. Population, and it will be customized to every database searched depending on its special filters. The detailed search strategy can be found in Extended data¹⁶.

Study records

Data extraction and management: Citations resulting from searching the different databases will be imported into EndNote, merged, and checked for duplicates. After removing duplicates, studies will be imported into an excel sheet, and the screening for titles and abstracts, for inclusion or exclusion, will be conducted.

Study selection: Two reviewers will independently screen the titles and abstracts to define eligible studies using the inclusion and exclusion criteria. When it is unclear whether a study is to be included or not, it will be included for the full text screening stage. If any reviews are identified among the studies, a manual search of their included studies will be conducted to ensure that all relevant studies are included. Conflicts between reviewers will be resolved through discussion till a consensus is reached, else a third reviewer will resolve the conflict. If there is missing information in a study, the author will be contacted for clarification. A PRISMA flow diagram will be used to keep track of the studies selection process.

Study outcome measures

Primary outcome: The relative reduction in antimicrobial prescriptions.

Secondary outcome: The relative reduction in re-consultation of patients, whether for infection recurrence or worsening condition (i.e.: symptoms).

Assessment of risk of bias

For the internal validity of this review, Newcastle Ottawa Scale¹⁷ tool will be used to assess the risk of bias for the included studies. The domains of this tool are selection (“representativeness of the exposed cohort”, “selection of the non-exposed cohort”, “ascertainment of exposure”, and “demonstration that outcome of interest was not present at start of study”), comparability (“comparability of cohorts on the basis of the design or analysis”), and outcome (“assessment of outcome”, “was follow-up long enough for outcomes to occur”, “adequacy of follow up of cohorts”). The results of the risk of bias assessment will be used to inform the synthesis, the interpretation of the results as well as the discussion of findings for this review.

Data synthesis

As most studies in this review will be observational, which may lead to heterogeneity of the data, a Random Effects Model will be used to obtain odds ratios with a 95% confidence interval. The I² statistic will be calculated to studying the heterogeneity of the studies included. If any missing data is suspected in one or more studies during the review, the authors of the study will be contacted to clarify, and thus minimize the risk of bias during the meta-analysis. If sufficient data is available, this review will conduct subgroup analysis for population age groups.

Study status

This study is currently in the “abstract-title” screening phase.

Strengths and limitations of this study

This study will explore the advantages of AI in reducing antimicrobial prescribing, and the “no-restriction” on language, time and setting of the studies will lead to inclusion of more studies, and thus higher applicability of findings. On the other hand, the expected heterogeneity due to the observational nature of the studies may hinder meta-analysis, and the studies, with poor quality, will represent some difficulty in the interpretation of findings.

Underlying data

No data are associated with this article.

Extended data

Figshare: Search Strategy - Artificial Intelligence to Improve Antimicrobial Prescribing- A Protocol for a Systematic Review. https://doi.org/10.6084/m9.figshare.20195723.v1¹⁶

This project contains the following extended data:

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).

Reporting guidelines

Figshare: PRISMA-P Checklist for “Artificial intelligence to improve antimicrobial prescribing: A protocol for a systematic review”. https://doi.org/10.6084/m9.figshare.20195573¹⁸

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).