Comment / Correspondence Beyond Trajectory: Limits of Metric Representations in Early Clinical Instability

This correspondence addresses a central limitation in current trajectory-based approaches to clinical instability. While recent developments in clinical AI move beyond static, state-based models toward dynamic representations of change, many systems rely on aggregated trajectory metrics to detect early deterioration. This represents an important conceptual advance, but introduces a critical problem. Early loss of stability in biological systems is not a single signal, but a structured pattern emerging across variability, recovery dynamics, and interactions between subsystems. Different underlying processes may produce similar trajectory signatures, particularly in early phases. As a result, a single trajectory metric may preserve sensitivity to change while obscuring structural heterogeneity. Systems may detect instability without distinguishing between qualitatively different forms of change. This paper argues that clinical decision systems must move beyond single-metric representations toward structured descriptions of instability, preserving distinctions between dynamic regimes. The goal is not only detection, but interpretation. This work is part of the Clinical AI Correspondence Series, focu