The Causal Analytic Workflow and A Practical Guide to Structured Data Processing for Antimicrobial Resistance Surveillance

The Causal Analytic Workflow and A Practical Guide to Structured Data Processing for Antimicrobial Resistance Surveillance | Research Square Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Help Center Sign In Submit a Preprint Cite Share Download PDF Short Report The Causal Analytic Workflow and A Practical Guide to Structured Data Processing for Antimicrobial Resistance Surveillance Michelle Louise Harrison, Mengyao Tang, Alkim Ozaygen, Paul Turner, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8962388/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background: Efforts have been made to standardise Antimicrobial Resistance (AMR) data collection globally. There remain challenges in improving the efficiency of data collection and processing to facilitate the comparison of high-quality AMR data between institutions and across countries. We aimed to design a causal analytic workflow to standardise data processing and provide structure for quality data monitoring using a generic raw data structure. Methods: We identified challenges in the processing of AMR and infection surveillance data, including a high time burden to process data, differing raw data structures, differences in identification and management of likely contaminants, and the laborious task of managing antimicrobial prescription and antimicrobial susceptibility testing (AST) data. Using two neonatal infection surveillance databases (ACORN2 and NeoSEAP), we sought to develop a standardised processing method to provide a generic data structure for robust cross-database comparisons. Results: We developed an 8-step causal analytic process that integrated multidisciplinary input from data analysts, decision makers, domain experts, data collectors, and researchers to streamline data processing. We implemented standardised scripts to convert AMR data from differing structures (wide and long) into a generic relational data structure representing four main concepts: ‘persons’, ‘episodes’, ‘organisms’ and ‘specimens’. We implemented standardised algorithms to interpret and group antimicrobial and AST data and to categorise pathogens by degree of pathogenicity. This process led to the production of three data monitoring reports: a raw data report for data quality and validity checks, a monitoring report for identifying important patterns in the data, and an analytic report for answering key research questions. Conclusions: More efficient and low-resource dependent methods are essential in the processing of AMR data. This causal analytic process will allow improved structure for reliable, generalisable and comparative reporting of AMR across clinical sites globally. AMR Antimicrobial Resistance Surveillance data Data processing Causal-analytic workflow relational database data engineering Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Introduction As we enter an era in which large datasets are cultivated, analysed, and processed, there is an ever-increasing need to ensure that methods for data collection and interpretation are standardised and of high quality. The need to optimise data collection and processing is increasingly important as we concurrently enter an era in which antimicrobial resistance (AMR) poses a major threat to public health and has global implications that must be urgently addressed. Since its recognition as a risk to global health, United Nations member states have adopted national action plans to tackle AMR and reducing rates of AMR has been included as a sustainable development goal (SDG) with a focus for improvement by the year 2030. 1 The United Nations has declared an urgent need to respond to the rise of resistant bacteria through ‘innovative approaches’ to ‘report quality surveillance data’. 2 However, despite this strong political recognition, gains have been slow, and challenges remain in identifying, quantifying, and managing resistant infections. There have been many efforts to standardise data collection globally, including the development of WHONET 3,4 and WHO GLASS 5,6 , as well as others that link to laboratory information management systems (LIMS). 7 These web-based systems allow systematic data entry via manual entry or direct extraction and upload from LIMS, with free, open-access online training to support their use. AMR researchers have highlighted inconsistencies in how clinical sites use WHONET: some upload only positive cultures, while others upload all culture results. Inconsistencies make comparison of data between institutions and across countries challenging. While these systems facilitate data capture, they continue to rely on laboratory staff uploading data in specific formats and collect only microbiological data without links to the associated clinical data, creating challenges in the applicability of surveillance data. In particular, the absence of linked clinical data has been a significant limitation of laboratory or microbiological AMR data, 8 preventing the translation of insights from the laboratory into action at the bedside. The acquisition of resistant pathogens is likely to lead to adverse clinical outcomes for patients, but it may not always be the case, and we cannot make this assumption. Linked clinical information, such as that captured by the ACORN2 project, where electronic case report forms are linked to laboratory systems via WHONET, enables policymakers to assess and report the true clinical impact of resistant infections. The use of strong surveillance systems such as this, helps to prevent reporting errors caused by the inclusion of non-sterile or duplicate samples, yet processing challenges remain. 9 , 10 Despite the benefits and the increasing global reach of surveillance projects such as ACORN2, clinically driven, high-quality AMR data remain sparse. A significant lack of high-quality surveillance data persists, with variable collection methods and inconsistent analytics contributing to a poor understanding of how AMR evolves over time. 11,12 The inherent complexity of the AMR problem domain requires multidisciplinary interaction among domain experts, decision-makers, researchers, data analysts, and data collectors. Systematic, clinically-oriented surveillance generates large amounts of data that need to be processed and analysed for global standardised reporting. Often, these data are collected in varying formats, which produce different data extracts, making the accurate interpretation and analysis time-consuming and labour-intensive. The time required to collect, monitor and process large amounts of high-quality surveillance data is a limiting factor. In low-resource settings where the burden of AMR infection is the highest, research and data collection are largely driven by clinician-researchers; they have first-hand knowledge of the impacts of AMR but limited time and expertise in data processing and limited access to those who can carry out complex data monitoring and analytics.  Ensuring data is of high quality requires regular and consistent monitoring with regular audits and data checks, 13 which is heavily resource-dependent. Artificial intelligence (AI) – the use of computer processing to conduct tasks requiring the interpretation, creation or complex reasoning that humans possess 14 – has the potential to enhance the efficiency of surveillance, processing and interpretation of AMR data. 15 AI-based analytic tools will be valuable in providing actionable insights into complex clinical problems by making predictions about complex systems.  It has the capacity to improve our ability to interpret and report large volumes of data and the potential to contribute to data-driven decision-making for clinicians treating infectious diseases. 15 Despite this great promise, a first step in achieving this potential is a clear understanding of the data generation process and how it can be transparently prepared for specific AI needs. Challenges have been reported in AI-model creation due to inadequately structured data, including missing data, heterogeneity of collection methods, variation in reporting styles and poor quality. 16 The requirement for high-quality data 17 makes it essential that we structure data collection and raw processing methods to ensure we can maximise the benefits AI may provide. To address the challenges of generating monitoring reports and processing data collected in variable formats, and to standardise data outputs and reduce the burden of developing analytic methods, we aimed to devise and define a data processing method for use in AMR surveillance.  We aimed to implement a causal analytic workflow (CAW) to guide the transparent generation of reproducible scripts that convert raw AMR data into a generic data format, which can then be monitored, analysed and interpreted at scale. We aimed to produce a generalisable step-by-step workflow that can be followed and replicated by any site collecting, interpreting and monitoring AMR data. We highlight the cross-disciplinary nature of coordination among diverse roles in implementing our proposed workflow, using concrete, illustrative project examples, and discuss its generalisability to other research contexts. Methods This paper describes the method of processing AMR surveillance data collected through 1) the ACORN2 project in Manila, Philippines, and 2) the multicentre NeoSEAP prospective study across South and Southeast Asia between November 2024 and November 2025. We devised the causal analytic workflow using the ACORN2 project dataset and tested its applicability in the NeoSEAP prospective study. Both projects collect prospective data on neonatal infections, antimicrobial use, antimicrobial resistance and clinical outcomes, yet both use differing data collection methods and data structures. Raw data structures and systems: The ACORN2 project captures clinical data in a secure online database (REDCap) 18 and links it to microbiology data collected in the WHONET laboratory information system via an R-Shiny app, providing data visualisation. Clinical staff and hospital administrators can visualise the data on a password protected online dashboard.  At the Philippine General Hospital, WHONET is used to report culture-positive results, while a separate database called RADISH is used to report culture-negative results. An additional REDCap electronic case report form (CRF) was designed to capture birth information, not originally included in the ACORN2 database. As such, the ACORN2 project at PGH requires the combination of four raw data files for processing:  two from REDCap, one from WHONET and one from RADISH. These data are in a ‘long’ format, with multiple rows per participant for participants who have multiple episodes of infection and the potential for multiple cultures per episode. Raw data collected for antimicrobial susceptibility testing (AST) are commonly collected and presented in a long format. Conversely, the NeoSEAP_PS dataset is designed in a ‘wide’ format with one row per participant that encompasses all infection episodes and isolated pathogens. The data processing methods outlined in this causal analytic workflow convert both structures into a common generic structure to facilitate ease of interpretation and comparison of common variables. Data processing methods: The raw data variables from each data source are divided into four main concepts using relational database concepts. 19 The ‘persons’ concept is for variables that pertain to a particular participant and exist only once in the dataset; these include factors such as birthweight, gestational age at birth, age at onset of first infection, mode of delivery and any maternal and antenatal history that may be relevant. The ‘episodes’ concept encompasses variables that pertain to a particular episode of infection, including age-at-onset of infection, clinical signs of infection, details of anti-infective prescription and clinical outcomes of each infection. The distinction between a person and episode concept is important, particularly in the neonatal population as neonates may have multiple infection episodes over long stays in the neonatal intensive care unit. Confusing these concepts can result in double-counting of the numerator or the wrong denominator being used during analysis. The final two concepts are the ‘specimens’ and the ‘organisms’ concepts. These two concepts relate to the specific laboratory data. A single episode may result in the collection of multiple specimens, and each specimen may yield the isolation of one or more organisms, each with its associated antimicrobial susceptibility profile. In processing the data for these two projects, we identified the need to systematically account for contaminants within the organism data. Appropriate reporting of bacteria that are commonly present as contaminants is another challenge for epidemiologists reporting on neonatal sepsis. 20 Determining whether an isolated organism is a true pathogen is important, particularly when determining appropriate antimicrobial policy. The other identified need during the data process was the need to describe antibiotics in a systematic and efficient way. Standard antibiotic groupings would allow more efficient reporting of access, watch and reserve (AWaRe) categories, 21 antibiotic classes and the linkages with antimicrobial susceptibility results. We used GitHub 22 to communicate project progress and tasks, share code updates, and monitor version control of the data process. Following the dataset's conceptual structure, we developed a fully scripted, reproducible extract–transform–load (ETL) workflow to standardise data processing across sites. The workflow consists of three annotated modules: Extract module – imports raw data files, reconstructs unique identifiers, reshapes wide laboratory fields into long format (if applicable), and harmonises date formats. This module produces the initial relational tables (persons, episodes, specimens, organisms, and antibiotics raw). Transform module – performs cleaning and enrichment of all extracted tables. This includes standardisation of organism names, integration of free-text organism entries into the main organism list (“Other – please specify”), assignment of pathogenicity classes using the harmonised ACORN2 ruleset, and grouping of antibiotics using global Aware categories 23 and pharmacological classes. All transformation steps are fully deterministic and reproducible. Load module –creates version-controlled datasets for downstream analysis. Each processing run generates both a date-stamped set of outputs and a corresponding “latest” version to support continuous monitoring and report generation. The ETL workflow was implemented to ensure transparent, reproducible, and scalable data processing across multiple countries and sites. All scripts were version-controlled on GitHub, and fully annotated versions of the Extract, Transform, and Load modules are provided in the supplementary materials (Supplementary Files S1–S3), together with the automated quality-control templates used for data monitoring (Supplementary File S4). While these data scripts are designed using R-studio software, the same modular structure is reproducible across other platforms such as Python. Generation of the ETL workflow, combines the knowledge, skill and expertise of researchers, data engineers and analysts in response to surveillance questions posed by the problem-owner, the domain experts and the researchers. Together, using a multidisciplinary response, and using communication channels such as GitHub projects and a common data schema, they derive variables for raw data monitoring, for surveillance data monitoring, and for analytics Results Comparing the implementation of the data processing method between two AMR data sets with differing structures allowed us to test the functionality, generalisability and reproducibility of the proposed CAW.  We identified an eight-step process to transform raw AMR data, regardless of their original structure, into a generic data structure suitable for comparison and monitoring, while carefully considering denominators (Figure 1). Directed Acyclic Graph (DAG) Once a research question is proposed, an expert knowledge elicitation process with global health practitioners, researchers and clinicians can highlight the causal relationships between the main variables of interest. These relationships are captured in a directed acyclic graph (DAG) 24 of the mechanistic pathway to the target-reported outcome. For these two projects, the target outcome is infection with a resistant pathogen. Each node in the DAG corresponds to variables to extract for the analytic report, and is highlighted in the data schema used for communication. Definitions of roles and data architecture Given the complexity of data collection, processing and interpretation for AMR, a multidisciplinary approach is needed. This includes problem owners, researchers, domain experts, causal modellers, data engineers, project coordinators, and site staff. An individual can exist across multiple roles with one person concurrently possessing the role of problem owner, domain expert and researcher, for example (Figure 2). A problem owner in this scenario would be a nurse unit manager, for example, responsible for monitoring and acting upon rising AMR, rising rates of hospital-acquired infections and observing antimicrobial stewardship. Responsibilities and actions undertaken within each role are further outlined in Figure 2. Additionally, the architecture of the data processing method, which takes raw data and converts it into a generic data structure, is shown in Figure 3. Stages of the analytic process A data dictionary representing the data collection tools is used to develop a raw data processing schema by categorising the collected data variables into the relational concepts of Persons, Episodes, Organisms, and Specimens. This schema is then used to inform the variable collected into four data comma-separated value (.csv) files using standardised R scripts (Supplementary material S1-S3). Once raw data files are generated, we use generic functions in R 25 to transform the data, first searching for organisms commonly considered bacterial contaminants using a standardised pathogenicity table (Supplementary material S5). This table, derived by domain experts and informed by published literature, 26 classifies pathogenicity as ‘high’ or ‘low’ to determine the likelihood that a particular organism is a causal pathogen rather than a contaminant. The intention for this table is to make it reproducible across projects and to serve as a live, editable resource that can be continuously updated based on the pathogens likely to be isolated within the surveillance population. Additionally, a second generic algorithm for antibiotic data has been developed for use (Supplementary material S6), which creates a summary of commonly prescribed antibiotic agents, their World Health Organization (WHO) ‘Access’, ‘Watch’ and ‘Reserve’ category (AWaRe) 21,23 , their antibiotic class and the associated antimicrobial susceptibility data. This list can also be amended for use across different populations, and new antimicrobials can be added as required. Example designs for data monitoring and analytic investigations The data monitoring reports were designed by the researcher and generated by the analyst in R-markdown to allow monitoring of the frequency of data entry by the decision-maker. Concept Monitoring can be directed and driven by the decision-making needs of data collectors and site staff, together with the researcher and operations co-ordinator, focused on the most important variables for a particular project. For the two projects, the most pertinent variables presented included the number of new enrolments each day, the number of culture-positive infections identified per week or per month, mortality rates, outbreaks and missing data points. The first report run is a Raw data monitoring report (Supplementary material S6). This is run to ensure validity and logic checks for smooth data linkage and processing. This includes checking that participant identification numbers are recorded and align across different data collection tools, that the same number of participants are included in lab data and clinical data and that there aren’t any errors that can be easily identified. The aim of this step is to minimise future data processing and to optimise efficiency. Figure 4 provides examples of what can be identified in this phase, with panel A showing a table of missingness and panel B highlighting an obvious data-collection error in which the incorrect date has been entered for the day-28 follow-up data. The second report generated is a data insights report, which can give monitoring insights into the data collection. These insights include enrolment frequency, which can give the researcher, operations co-ordinator, and site data collectors an indication of the enrolment pace and the distribution of collected data. Figure 5 provides examples of the monitoring report with panel A and panel B showing the pace of acquisition of new infection episodes, showing (A) the frequency of infection episodes per month, and (B) the distribution of community- and hospital-acquired infections within the monitored period. Panels (C) and (D) can provide clinicians and policymakers with insights into the impact of antimicrobial stewardship practices, using data on the timing of antimicrobial administration (C) and blood culture collection (D). Finally, the analytic report is designed and generated to efficiently answer the growing number of research questions that arise with the collection of surveillance data. These reports, after being generated by data analysts, are analysed and communicated by the decision makers, researchers and domain experts for clinical uptake and implementation. This process was used to answer research questions during analysis of ACORN2 surveillance data collected at the Philippine General Hospital. 24 The development of analytic reports requires input from domain experts, decision-makers, and researchers to ensure that pertinent questions are answered. This step requires further processing of the same raw data points, but to facilitate different interpretations and insights. The examples included here include analysis of mortality for community and hospital-acquired infections caused by different identified pathogens. The second example shows the degree of processing where pathogenicity was analysed to determine the presence of causal pathogens and likely contaminants. Discussion Processing and reporting AMR data remains a challenge. Despite the promise of AI models that can interrogate big datasets, the quality and accuracy of source data currently limit their applicability. Rising AMR with forecasts predicting 1.9million AMR attributable deaths by the year 2050 27 emphasises the need to better understand, monitor and report the spread of resistant pathogens. Many groups globally are working to improve surveillance and reporting of neonatal AMR data. Yet the quality remains variable: some reports include only microbiological data without a linked clinical picture, others report commonly regarded contaminants as pathogens, and some contain data errors, such as using the wrong denominator or no denominator reported at all. In a systematic review and meta-analysis of the causes of early- and late-onset sepsis conducted in 2023, GRADE 28 scoring of the data quality revealed predominantly low-quality reporting, with only 12% (6/48) receiving GRADE level B for moderate-quality evidence. 29 Here, we propose a method to standardise data processing, improve and monitor quality, and produce more homogeneous data, which can then be reported at scale. In processing the ACORN2 data collected at the Philippine General Hospital, it became clear that a streamlined, systematic data-monitoring and analytic process was essential, yet the time required for setup was too great an undertaking for local clinical and surveillance staff. Developing the data processing structure and providing open-source scripts that can be customised to each project's needs should significantly reduce the time required to generate reports on raw data, quality concepts, and analytics. Testing the process on two differently with different structures allowed us to develop the method using the ACORN2 data and then test its implementation on the NeoSEAP dataset. We found that generating a relational database that converts data into a generic format based on the concepts of ‘episodes’, ‘specimens’, and ‘organisms’ allowed for easy comparison and reproducible scripts, ensuring denominators were correct and data exports were structured for streamlined review. This responds to calls by those reviewing the use of AI in AMR data cleaning and transformation infrastructure and for generic, standardised and transparent data structures. 17 A multi-disciplinary team approach is required when designing analytics for a complex domain such as AMR, and here we outline a method to do just that. The interaction between roles and team members helps ensure that the problem domain is accurately represented, the main factors driving policy decisions are considered and incorporated, and the research questions are generated with an agreed-upon understanding of the causal process at the centre. This process underscores the power of simple analytics when they are designed to be relevant to decision-makers and communicated effectively across the multidisciplinary team. Antibiotic and pathogenicity algorithms are generalisable and modifiable, meaning they can be reproduced across projects that report pathogens or antimicrobial use, improving the implementation efficacy of all such projects. This could include projects analysing data on other types of infections, not only in infants but also in older children and adults. New antibiotics can be added to the list, and new classifications can be easily added to the antibiotic database if the WHO 19 categories are updated or advanced. Similarly, the pathogenicity table can include new common contaminants or new pathogens; rules can be implemented to treat an organism with low pathogenicity as a pathogen, depending on the clinical context. This adaptability is a real strength of the proposed process, as it enables standardisation and ease of implementation. While we were able to modify and refine the process across the two projects, this analytic process has some limitations. Firstly, it still requires a degree of data knowledge and understanding to tailor the scripts to the project's individual needs. Where the ACORN2 dashboard builds in the analytic process to pull simply and directly from WHONET and REDCap, this proposed method requires some modification of coded-scripts and therefore will need someone with data knowledge to facilitate its use. It is important to recognise that one person can encompass many roles across the proposed roles we defined (e.g., the problem owner may also be the domain expert and the data analyst), and therefore, some basic data processing skills can be acquired to help orchestrate data processing. Finding methods to promote data monitoring is essential, particularly when presenting the large amounts of data collected during systematic surveillance of AMR. Promoting regular data monitoring, which highlights missing data and data entry errors, aims to improve the generalisability and quality representation of  AMR globally, facilitating a greater understanding of the burden of resistance across all settings. Once established, these monitoring reports can be run simply to monitor not only errors but also trends in infection and antimicrobial stewardship, offering an effective solution to a complex problem. Analytics derived within a multi-disciplinary team, driven by the causal pathway and executed to promote high-quality data representation, will help advance our understanding and management of resistant pathogens globally. Declarations Funding was provided in part by the University of Sydney Research Training Program (RTP scheme) and the National Health and Medical Research Council of Australia (NHMRC). This research was additionally funded in part by the Wellcome Trust [222156/Z/20/Z to PT]. For the purpose of Open Access, the author has applied a CC BY public copyright licence to any Author Accepted Manuscript version arising from this submission. Ethical approva l: ethical approval has been granted for the data collection and surveillance in both ACORN2 (2024HE000/366) and NeoSEAP studies (2023/HE000528). Authors contributions: YW conceptualised the project, YW, TS, PCMW, PT and MLH contributed to the academic design and output of the project. MT and AO engineered the data infrastructure and contributed to the design and execution of the data processing tools. MLH and MT drafted the manuscript, All authors contributed to the editing of the final manuscript. Acknowledgments: Authors would like to acknowledge data collection teams at the Philippine General Hospital Manila, Philippines, Soetomo General Hospital, Surabaya, Indonesia, Hasan Sadikin General Hospital, Bandung Indonesia and Aga Khan University Hospital, Pakistan for their efforts in data collection for which this data processing system has been engineered. Competing interests: Authors have no competing interests to declare. References United Nations Sustainable Development Goals. Online. Available from: https://sdgs.un.org/goals UN General Assembly High-Level Meeting on antimicrobial resistance 2024. Online: United Nations; 2024.https://www.who.int/news-room/events/detail/2024/09/26/default-calendar/un-general-assembly-high-level-meeting-on-antimicrobial-resistance-2024 WHONET: WHO Collaborating Centre for Surveillance of Antimicrobial Resistance. 2025 [cited 2025 14 October]. Available from: https://whonet.org/ WHONET microbiology laboratory database software. 2025. Available from: https://whonet.org/ Global antimicrobial resistance and use surveillance system (‎GLASS)‎ report: 2022. World Health Organization Online: World Health Organization; 2022.https://www.who.int/publications/i/item/9789240062702 GLASS manual for antimicrobial resistance surveillance in common bacteria causing human infection. Online; 2023 31 August 2023.https://www.who.int/publications/i/item/9789240076600 Holt KE, Carey ME, Chandler C, Cross JH, Dyson ZA, Furnham N, et al. Tools and challenges in the use of routine clinical data for antimicrobial resistance surveillance. npj Antimicrobials and Resistance 2025;3(1).10.1038/s44259-025-00105-3 van Doorn HR, Ashley EA, Turner P. Case-based surveillance of antimicrobial resistance in the ACORN (A Clinically Oriented Antimicrobial Resistance Surveillance Network) study. JAC-Antimicrobial Resistance 2020;2(1).10.1093/jacamr/dlaa018 Turner P, Ashley EA, Celhay OJ, Douangnouvong A, Hamers RL, Ling CL, et al. ACORN (A Clinically-Oriented Antimicrobial Resistance Surveillance Network): a pilot protocol for case based antimicrobial resistance surveillance. Wellcome Open Research 2020;5:13.10.12688/wellcomeopenres.15681.2 ACORN2 protocol: A Clinically-Oriented Antimicrobial Resistance Surveillance Network – Phase 2. Online2023 [cited 2024]. Available from: https://acornamr.net/docs/ACORN2_protocol_V2.0.pdf Iskandar K, Molinier L, Hallit S, Sartelli M, Hardcastle TC, Haque M, et al. Surveillance of antimicrobial resistance in low- and middle-income countries: a scattered picture. Antimicrobial Resistance & Infection Control 2021;10(1).10.1186/s13756-021-00931-w Hay SI, Rao PC, Dolecek C, Day NPJ, Stergachis A, Lopez AD, et al. Measuring and mapping the global burden of antimicrobial resistance. BMC Medicine 2018;16(1).10.1186/s12916-018-1073-z Seale AC, Gordon NC, Islam J, Peacock SJ, Scott JAG. AMR Surveillance in low and middle-income settings - A roadmap for participation in the Global Antimicrobial Surveillance System (GLASS). Wellcome Open Research 2017;2:92.10.12688/wellcomeopenres.12527.1 Defining Artificial Intelligence. Online: National Aeronautics and Space Administration (NASA); 2025 [cited 2025 11 November]. Available from: https://www.nasa.gov/what-is-artificial-intelligence/ Arnold A, McLellan S, Stokes JM. How AI can help us beat AMR. npj Antimicrobials and Resistance 2025;3(1).10.1038/s44259-025-00085-4 Kasse GE, Cosh SM, Humphries J, Islam MS. Leveraging artificial intelligence for One Health: opportunities and challenges in tackling antimicrobial resistance - scoping review. One Health Outlook 2025;7(1).10.1186/s42522-025-00170-8 Howard A, Reza N, Green PL, Yin M, Duffy E, Mwandumba HC, et al. Artificial intelligence and infectious diseases: tackling antimicrobial resistance, from personalised care to antibiotic discovery. The Lancet Infectious Diseases.10.1016/S1473-3099(25)00313-5 Harris PA, Taylor R, Minor BL, Elliott V, Fernandez M, O'Neal L, et al. The REDCap consortium: Building an international community of software platform partners. Journal of Biomedical Informatics 2019;95:103208.https://doi.org/10.1016/j.jbi.2019.103208 Dilling TJ. Artificial Intelligence Research: The Utility and Design of a Relational Database System. Adv Radiat Oncol 2020;5(6):1280-5.10.1016/j.adro.2020.06.027 Hossain B, Weber MW, Hamer DH, Hibberd PL, Ahmed ASMNU, Marzan M, et al. Classification of Blood Culture Isolates Into Contaminants and Pathogens on the Basis of Clinical and Laboratory Data. The Pediatric Infectious Disease Journal 2016;35(5):S52-S4.10.1097/inf.0000000000001107 AWaRe classification of antibiotics for evaluation and monitoring of use. Online2023 [cited 2025 13 March]. Available from: https://www.who.int/publications/i/item/WHO-MHP-HPS-EML-2023.04 Github. 2025 [cited 2025]. Available from: https://github.com/projects Sharland M, Zanichelli V, Ombajo LA, Bazira J, Cappello B, Chitatanga R, et al. The WHO essential medicines list AWaRe book: from a list to a quality improvement system. Clinical Microbiology and Infection 2022;28(12):1533-5.10.1016/j.cmi.2022.08.009 Harrison M, Harrison ML, Uy ME, Kasahara E, Caparas JJ, Duarte NE, et al. Diagnostic Challenges and Gram-Negative Pathogen Dominance in Early- and Late-Onset Neonatal Sepsis in Manila, Philippines: A Prospective Observational Study. Preprint Lancet 2025.Available at SSRN: https://ssrn.com/abstract=5400335 or http://dx.doi.org/10.2139/ssrn.5400335 R Core Team (2022). R: A language and environment for statistical computing. . Vienna, Austria. Available from: https://www.R-project.org/ Hossain B, Weber MW, Hamer DH, Hibberd PL, Ahmed ASMNU, Marzan M, et al. Classification of Blood Culture Isolates Into Contaminants and Pathogens on the Basis of Clinical and Laboratory Data. The Pediatric Infectious Disease Journal 2016;35(5) Naghavi M, Vollset SE, Ikuta KS, Swetschinski LR, Gray AP, Wool EE, et al. Global burden of bacterial antimicrobial resistance 1990–2021: a systematic analysis with forecasts to 2050. The Lancet 2024;404(10459):1199-226.10.1016/s0140-6736(24)01867-1 Granholm A, Alhazzani W, Møller MH. Use of the GRADE approach in systematic reviews and guidelines. British Journal of Anaesthesia 2019;123(5):554-9. Harrison M, Dickson B, Sharland M, Williams P. Beyond early- and late-onset neonatal sepsis definitions What are the current causes of neonatal sepsis globally? A systematic review and meta-analysis of the evidence. Pediatric Infectious Disease Journal 2024.DOI: 10.1097/INF.0000000000004485 Additional Declarations No competing interests reported. Supplementary Files SupplementaryS2Transform.docx SupplementaryS1ExtractModule.docx SupplementaryS3Load.docx SupplementaryS4QCReports.docx SupplementaryS5Pathogenicitytable.xlsx SupplementaryS6AntibioticandASTtable.xlsx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information

Related documents

• A socio-ecological System Dynamics model of antimicrobial use and resistance#120229
• Safety Monitoring of High-Risk Antibiotics Using Artificial Intelligence: A Narrative Review with Focus on Real-World Evidence.#409246
• Testing New Ways to Name Antimicrobial Resistance#448290
• Laboratory tools from the Fleming Fund antimicrobial resistance surveillance programme.#631167
• Code and data for: Composable probabilistic models can lower barriers to rigorous infectious disease modelling#231915