Stable Longitudinal Screening of Latent Physiological Dysregulation from Psychometric Data Using Machine Learning

Area of Science:

• Psychosomatic Medicine
• Computational Biology
• Population Health

Background:

• Chronic stress links psychosocial factors to health via physiological dysregulation.
• Current identification methods are invasive or resource-intensive.
• Scalable, non-invasive screening is needed for early detection.

Purpose of the Study:

• Evaluate high-dimensional psychometric data for non-invasive screening of physiological dysregulation.
• Develop a framework for scalable population health tools from longitudinal data.

Main Methods:

• Utilized longitudinal data from Midlife in the United States (MIDUS) Waves 2 and 3.
• Defined physiological targets across inflammatory, metabolic, and neuroendocrine domains using allostatic load.
• Employed a teacher-ranking-pruning-student pipeline for dimensionality reduction and knowledge distillation.

Main Results:

• Reduced predictor dimensionality significantly without performance loss.
• Achieved area under the receiver operating characteristic curve up to 0.78.
• Demonstrated substantial precision-recall lift over baseline prevalence.

Conclusions:

• Psychometric data can support scalable, non-invasive screening for latent physiological dysregulation.
• The developed framework generalizes longitudinal data into deployable population health tools.
• Enables population-level health screening using only survey data at inference.