Accurate and timely identification of hospital outbreak clusters is crucial for preventing the spread of infections that have epidemic potential. While assessing pathogen similarity through whole genome sequencing (WGS) is considered the gold standard for outbreak detection, its high cost and lengthy turnaround time preclude routine implementation in clinical laboratories. We explore the utility of two rapid and cost-effective alternatives to WGS, matrix-assisted laser desorption ionization-time of flight (MALDI-TOF) mass spectrometry and antimicrobial resistance (AR) patterns. We develop a machine learning framework that extracts informative representations from MALDI-TOF spectra and AR patterns for outbreak detection and explore their fusion. Through multi-species analyses, we demonstrate that in some cases MALDI-TOF and AR have the potential to reduce reliance on WGS, enabling more accessible and rapid outbreak surveillance.
Disease outbreaks in hospitals can occur when pathogens spread unchecked among patients, medical equipment, and healthcare workers 1,2 . Identifying these outbreaks at their early stages is vital to preventing infections and limiting the resources needed to address widespread transmission 3 . To achieve this, healthcare institutions compile epidemiological data from the electronic health record and, separately, collect pathogen data from patient and environmental samples to detect and address potential outbreak clusters 4 . Consisting of samples grouped by pathogen similarity, these clusters are considered connected to recent transmission events 5,6 , enabling the tracking of pathogen dissemination within hospitals 7 .
Whole genome sequencing (WGS) stands as the gold standard for assessing pathogen similarity, currently providing the most information and highest discriminatory power 8 . The technology has been used to better understand the transmission of multiple pathogens, e.g., Pseudomonas aeruginosa (PSA) 9,10 and vancomycin-resistant Enterococcus faecium (VRE) 11,12 . Specifically, two cases belong to the same cluster if they are separated by single-nucleotide polymorphisms (SNPs) fewer than a specified threshold 13,14 . However, the cost and turnaround time required for WGS have prohibited its widespread deployment in clinical laboratories 15 .
To attain cost-effective and rapid outbreak detection that can be deployed with ease, we explore two existing alternatives to WGS: matrix-assisted laser desorption ionization-time of flight (MALDI-TOF) mass spectrometry and antimicrobial resistance (AR) patterns. MALDI-TOF is a routine tool for identifying bacterial species in clinical laboratories 16,17 and has been suggested as a possible substitute for WGS in outbreak cluster detection due to its affordability and rapid turnaround time 18 . It produces intensity spectra based on mass-to-charge ratios (m/z) of microbial proteins, creating a distinctive “fingerprint” for each microorganism. Subsequently, these spectra undergo analysis to identify outbreak clusters 19,20 . MALDI-TOF has shown promise in investigating the epidemiology of outbreaks associated with VRE 18 , Streptococcus pneumoniae (SPN) 21 , and methicillin resistant Staphylococcus aureus (MRSA) 22 . However, it has also been shown that MALDI-TOF disagreed with WGS-defined ground truth clusters for VRE 23 and Klebsiella pneumoniae (KLP) 24 outbreaks. Nevertheless, such studies are often limited in scope by examining small-scale data of only one single pathogen at a time. Furthermore, without eliciting crucial, relevant knowledge embedded in the MALDI-TOF spectra, their raw form may not be sufficient to produce clustering results with optimal similarity to WGS.
Leveraging data already collected by clinical laboratories, outbreak detection through analysis of AR patterns is attractive for its minimal resource burden 25 and near real-time speed 26 . It has witnessed success in detecting outbreak clusters associated with Shigella spp. 25 and VRE 27 . However, while AR often yields high sensitivity in outbreak detection, researchers noted a lack of specificity in the identified clusters 28 . Nevertheless, there has been a lack of research that compares the utility of AR with other candidate outbreak detection methods, such as MALDI-TOF.
To fill these gaps, we propose a machine learning-based framework that extracts representations of MALDI-TOF and AR data tailored to hospital outbreak cluster detection. Using a comprehensive dataset, we conduct a multi-species analysis that compares clustering results of MALDI-TOF and AR with those of WGS. We evaluate the utility of MALDI-TOF, AR, and their synergy, in species-agnostic and species-specific outbreak detection settings. We also demonstrate that in some cases MALDI-TOF and AR could potentially act as cost-effective and rapid alternatives by reducing the need for WGS in outbreak detection.
Data source: Our dataset is de-identified and proprietary, obtained from a large non-profit research and academic hospital. It consists of 4921 isolates spanning 17 bacterial species (Table 1), collected from 10/2021 to 10/2024 duri ng routine surveillance and were used to identify active outbreaks. Each isolate contains a raw MALDI-TOF spectrum and a phenotypic antimicrobial resistance profile. For each pair of isolates under the same species, we collect their SNP distance derived from WGS experiments. Ground truth outbreak clusters are determined by performing hierarchical clustering (with complete linkage) on the complete SNP distance matrix with a cutoff distance of 15. This threshold reflects the standard practice adopted by the hospital infection control team that provided our proprietary data, serving as their routine reference point for detecting emerging outbreak clusters. In our experiments, we perform 4-fold cross-validation on isolates, grouped by clusters: we preserve the percentage of samples of each
This content is AI-processed based on open access ArXiv data.