Most medical models assume that average disease burden predicts average outcomes — but this mathematical framework challenges that assumption in a way that matters for cancer, infection, and chronic disease management. When internal biological states fluctuate unpredictably, the shape of the relationship between those states and mortality risk turns out to be just as important as the states themselves.
Published in PNAS, this theoretical work examines how within-host population variability — think fluctuating cancer clone counts or pathogen loads — translates into host-level survival outcomes. The core argument centers on the curvature of what the authors call the "hazard map": the mathematical function linking internal biological state to instantaneous mortality risk. When that curve is convex (accelerating upward), Jensen's inequality dictates that variability around a given mean state increases expected hazard beyond what the mean alone would predict. Conversely, concave curvature can make variability protective. The framework is deterministic in its logic but has broad implications for stochastic biological systems where variance is unavoidable.
This is a conceptually significant contribution because it formalizes an intuition that clinicians have long held but rarely quantified: that two patients with identical average tumor burdens or pathogen loads may have very different prognoses depending on how much those values fluctuate over time. The work draws on Jensen's inequality, a classical mathematical result, and applies it rigorously to survival analysis in a biological context — an application that has been underexplored. The key limitation is that this remains a theoretical framework; empirical validation across specific cancer types or infectious diseases has not yet been provided here. Whether real-world hazard maps are consistently convex in clinically relevant ranges is an open empirical question. If confirmed, this could shift how oncologists and infectious disease specialists interpret dynamic biomarker trajectories, moving attention from mean values toward variance as an independent risk signal. Incremental as a standalone paper, but potentially foundational if it spurs empirical testing.