Abstract:As we gain access to a greater depth and range of health-related information about individuals, three questions arise: (1) Can we build better models to predict individual-level risk of ill health? (2) How much data do we need to effectively predict ill health? (3) Are new methods required to process the added complexity that new forms of data bring? The aim of the study is to apply a machine learning approach to identify the relative contribution of personal, social, health-related, biomarker and genetic data as predictors of future health in individuals. Using longitudinal data from 6830 individuals in the UK from Understanding Society (2010-12 to 2015-17), the study compares the predictive performance of five types of measures: personal (e.g. age, sex), social (e.g. occupation, education), health-related (e.g. body weight, grip strength), biomarker (e.g. cholesterol, hormones) and genetic single nucleotide polymorphisms (SNPs). The predicted outcome variable was limiting long-term illness one and five years from baseline. Two machine learning approaches were used to build predictive models: deep learning via neural networks and XGBoost (gradient boosting decision trees). Model fit was compared to traditional logistic regression models. Results found that health-related measures had the strongest prediction of future health status, with genetic data performing poorly. Machine learning models only offered marginal improvements in model accuracy when compared to logistic regression models, but also performed well on other metrics e.g. neural networks were best on AUC and XGBoost on precision. The study suggests that increasing complexity of data and methods does not necessarily translate to improved understanding of the determinants of health or performance of predictive models of ill health.

Is your model predicting the past?

Past and Future: Backward and Forward Discounting

Backward discounting

What Will My Model Forget? Forecasting Forgotten Examples in Language Model Refinement

Predicting Performance During Tutoring with Models of Recent Performance

Performative Prediction: Past and Future

Electronic structure study by means of X-ray spectroscopy and theoretical calculations of the "ferric star" single molecule magnet

Evaluating the performance of personal, social, health-related, biomarker and genetic data for predicting an individuals future health using machine learning: A longitudinal analysis

Humans adaptively deploy forward and backward prediction

Latent-CF: A Simple Baseline for Reverse Counterfactual Explanations

Anticipating Performativity by Predicting from Predictions

Can Language Models Use Forecasting Strategies?

When regression coefficients change over time: A proposal

On the limits of algorithmic prediction across the globe

Why Has Predicting Downstream Capabilities of Frontier AI Models with Scale Remained Elusive?

Time Series Forecasting Models Copy the Past: How to Mitigate

Prediction scoring of data-driven discoveries for reproducible research

Interpretable machine learning approaches to prediction of chronic homelessness

Fair-by-design explainable models for prediction of recidivism

Understanding Prediction Discrepancies in Machine Learning Classifiers

Future Events as Backdoor Triggers: Investigating Temporal Vulnerabilities in LLMs