In clinical medicine, we are trained to look at the "average." We check if a patient’s mean blood sugar is too high, if their average blood pressure is stable, or if their weight falls within a healthy range. However, a growing body of research suggests that these static snapshots—measures of central tendency—ignore a far more potent predictor of health: volatility.
The core idea is that the "noise" in biological data—the minute-to-minute fluctuations, the day-to-day swings, and the "rollercoaster" effects—is not measurement error to be smoothed away. Instead, this variability is a direct signal of how well (or how poorly) a body’s regulatory systems are functioning.
From Measurement Noise to Biological Signal
Traditionally, longitudinal studies treat individual fluctuations around a mean trajectory as statistical noise. Researchers like Irena Chen and Michael R. Elliott have challenged this by demonstrating that individual-level residual variance is a critical biomarker for systemic dysregulation. In their work on the menopausal transition, they found that the instability of follicle-stimulating hormone, rather than just its absolute concentration, significantly predicts long-term bone mineral density loss.
This shift in perspective treats the "texture" of a signal as its primary informative feature. In the Time Series Topic Model (TSTM) developed by Suchi Saria and Daphne Koller, physiological signals are treated like a language. They argue that fine-grained residuals—the tiny fluctuations in a baby’s heart rate or oxygen levels—reveal specific "dynamic behaviors" that serve as more reliable indicators of health than absolute mean values.
Why Volatility Predicts Outcome
Why does a "swingy" signal predict a bad outcome? The intuition lies in homeostasis. A healthy biological system is a masterpiece of tight feedback loops. When a marker like blood glucose or blood pressure begins to fluctuate wildly, it signals that the underlying regulatory mechanisms are fraying.
For example, Jianbin Su and colleagues have shown that glycaemic variability—the magnitude of post-meal blood sugar swings—is a stronger independent risk factor for heart rhythm abnormalities and nerve damage than sustained high blood sugar alone. A "rollercoaster" of glucose levels causes more systemic stress than a steady, albeit high, level.
Similarly, in acute settings like the ICU, Dae Hyun Lee and Eric Horvitz found that the degree of fluctuation and atypia in vital signs are more significant predictors of mortality than the raw values themselves. A patient whose heart rate is "jumping" is often in more danger than one whose heart rate is consistently high.
Quantifying the Instability
To turn this intuition into a clinical tool, researchers use several mathematical strategies to capture the "second moment" (variance) of data:
- Simple Dispersion: Measures like the standard deviation (SD) or coefficient of variation (CV) of daily blood pressure readings have been used by Jinjun Ran to predict COVID-19 mortality, and by Lijuan Wang to differentiate between ischemic stroke subtypes.
- The "Gap" Metric: Siyu Liang and Amir Soltani have used the absolute difference between maximum and minimum values (e.g., the "sodium fluctuation" or "temperature gap") as proxies for homeostatic failure.
- Joint Modeling: Advanced statistical frameworks, such as those proposed by Sepehr Akhavan Masouleh and Daniel L. Gillen, use Gaussian processes to explicitly model within-subject volatility as a predictor for survival. This allows the model to distinguish between a patient’s general trend (the mean) and their stability (the variance).
- Distributional Regression: Instead of predicting a single number, Tong Chen and Sheng Wang use Neural ODEs to model how the entire distribution of gene expressions evolves, capturing shifts in variance that precede cancer.
Beyond Vitals: A Universal Signature
This principle of "volatility as signal" extends far beyond traditional vital signs.
In respiratory health, Davin Hill demonstrated that the reproducibility of effort in lung function tests—usually discarded as "bad data"—actually contains rich information about disease phenotypes. In metabolic health, Iulia Petria and Samuel Albuquerque found that weight cycling (the "yo-yo" effect) is a potent predictor of mortality in Type 1 diabetes, independent of BMI. Even in mental health, the volatility of emotional expression in social media posts serves as a primary indicator of psychological instability.
By shifting focus from the average to the fluctuation, these researchers are moving toward a more dynamic understanding of health—one where the body’s ability to remain steady in the face of change is the ultimate metric of resilience.