A Neuro-Symbolic AI Approach to Personal Health Risk Assessment and Immune Age Characterisation using Common Blood Markers

We introduce a simulated digital model that learns a person's baseline blood health over time. Using an adaptive learning algorithm, the model provides a risk assessment score that compares an individual's chronological age with an estimation of biological age based on common immune-relevant markers used in current clinical practice. We demonstrate its efficacy on real and synthetic data from medically relevant cases, extreme cases, and empirical blood cell count data from 100K data records in the Centers for Disease Control and Prevention's National Health and Nutrition Examination Survey (CDC NHANES) that spans 13 years. We find that the score is informative in distinguishing healthy individuals from those with diseases, both self-reported and as manifested via abnormal blood test results, providing an entry-level score for patient triaging. The risk assessment score is not a machine learning black-box approach but can interact with ML and DL approaches to help guide, control the attention given to specific features, and assign proper explainable weight to an otherwise transparent adaptive learning algorithm. This approach may allow fast and scalable deployment to personalised, sensitive, and predictive derivative indexes within digital medicine, without the need for a new test, assay, or prospective sampling, unlike other biological ageing-related scores and methods. It demonstrates the potential of clinical informatics and deep medicine in digital healthcare as drivers of innovation in preventive patient care.