Integrating Microbiological, Biochemical, and Clinical Laboratory Data with Machine Learning for Early Prediction of Antimicrobial Resistance

Shams Nahi Albeladi
Chemistry, Ministry of National Guard-Health Affairs, King Abdulaziz Medical City

Nouf Essam Owaidah
Clinical Laboratory, National Guard Hospital, King Abdulaziz Medical City

Ahmad Rafie Alluhibie
Medical Laboratory, National Guard Hospital, king Abdulaziz Medical City

Maha Mohammed Etwadi
Microbiology, National Guard Hospital, King Abdulaziz Medical City

Shahad Talal Karkashan
Biochemistry, Ministry of National Guard-Health Affairs, King Abdulaziz Medical City

Mashael Mohammed Alneami
Biochemistry, National Guard Hospital, king Abdulaziz Medical city

Rayan Hussien Nafea
Medical Laboratory, National Guard Hospital, King Abdulaziz Medical City

Hashem Mohammed Alfotewi
Laboratory Medicine, King Abdulaziz Medical City, National Guard, King Abdulaziz Medical City, Jeddah

Ranem Adham Jan
Microbiology, National Guard Health Affairs, King Abdulaziz Medical City, Jeddah

Abstract

Background: The global crisis of antimicrobial resistance is now the number one perceived emerging threat due to the slowness and late definitive nature of conventional, culture based antimicrobial susceptibility testing (AST); laboratory or clinical information that is already commonly acquired routinely can be translated by machine learning (ML) into an early estimation of resistance risk.

Objective: To provide a systematic narrative on modern ML approaches for prediction of AMR, and construction of a functional, multimodal approach integrating microbiological, biochemical/hematological and clinical laboratory data for early risk stratification in actual clinical workflow.

Methods: An integrative, evidence-synthesis systematic review was undertaken with peer reviewed manuscripts and recent systematic review publications spanning from 2020 through 2026, focusing upon the data modality(ies), machine-learning methods, time of prediction, discrimination capacity and the attainability of the prediction models for use with clinical care workflow in empiric infection management; patient-level data was not made available, so results have focused upon synthesized evidence as opposed to fabrication of patient data to facilitate a novel analytical outcome or presentation of clinical trial like evidence. Results: Recent clinical experience demonstrated clinically meaningful predictive capabilities from electronically medical data including LIS data, MALDI-TOF mass spectrometry data, and whole genome sequences. Between a 2022 systematic review and meta-analysis of 33 studies, an pooled AUC was computed of 0.82, and a 2025 systematic review on several high priority pathogens showed gradient boosted trees, random forest and XG Boost to be among the top performing predictive algorithms. Between a few examples from various investigations of MALDI-TOF for prediction purposes it seemed likely the AUROC performance for individual targeted pathogens may achieve 0.74 – 0.80 while genomic prediction models for specific drug resistance phenotype performed as well as required to reach higher discrimination. 2026 empirical evidence showed that that routine clinical, biochemical and other laboratory measures can accurately predict early risk even before definitive AST results can be obtained.

Conclusion: Instead of the replace culture-based AST with models which carry many logistic and validation barriers to translate into clinical workflow for patient care, instead, adding a calibrated explanation machine learning early warning layer to current AST processes using microbiology, biochemical/hematological/other clinical lab factors and priori susceptibility information will effectively inform empiric antibiotic selection, subsequent antibiotic need for optimization or withdrawal of drugs to achieve better patient care or infection control, and promote optimization of overall antimicrobial therapy stewardship.

Keywords: Antimicrobial Resistance ,Machine Learning ,Clinical Laborator ,Microbiology ,Antimicrobial Susceptibility Testing ,Maldi-Tof; Whole-Genome Sequencing