From species identification to AI-enabled outbreak triage: MALDI-TOF MS for early warning in hospitals

Purpose: Healthcare-associated infections (HAIs) remain a major clinical and economic burden, and early detection of transmission clusters is essential for timely intervention. Whole-genome sequencing (WGS) is the reference standard for outbreak confirmation, but its cost, turnaround time, and operational requirements limit its use as an always-on first-line surveillance tool. Because matrix-assisted laser desorption/ionization time-of-flight mass spectrometry (MALDI-TOF MS) spectra are already generated during routine species identification, they may provide a rapid triage layer before confirmatory sequencing. Methods: We synthesize current evidence for artificial intelligence (AI)-based MALDI-TOF MS analysis in within-species typing, strain relatedness inference, and outbreak-oriented surveillance. We organize the literature into three methodological pathways: peak-centric biomarker typing, supervised subtype classification, and representation learning. Results: The available evidence indicates that AI-based models can extract within-species signals from MALDI-TOF MS spectra that are relevant to typing and outbreak surveillance, but performance remains context-dependent. Peak-based approaches offer interpretability but are vulnerable to spectral instability and limited portability. Supervised classifiers can perform well for narrow predefined targets, but depend on fixed label spaces and may degrade when local epidemiology changes. Representation-learning approaches are less clinically validated for typing, but may offer the most scalable route toward adaptable, similarity-based early-warning systems. Conclusions: MALDI-TOF MS should not be viewed as a replacement for WGS, but as a locally calibrated decision-support tool that prioritizes suspicious isolates for sequencing and infection prevention and control (IPC) review, similar to a screening tool. Future studies should move beyond retrospective accuracy estimates and evaluate prospective clinical utility, including time gained, sequencing saved, false-alert burden, portability to new laboratories, and performance after minimal local calibration.