Machine Learning Modelling for Prospective Parkinson’s Disease - the importance of inflammatory biomarkers and IGF1 - UKB study

Abstract INTRODUCTION We use the world-leading UK Biobank dataset (UKB) with over 500,000 participants and >10,000 variables across multiple data modalities to determine a ranking of candidate risk factors for Parkinson’s Disease (PD) without a priori assumption. METHODS The Integrated Disease Explanation and Risk Scoring platform (IDEARS) applies machine learning algorithms to multi-modal personalised health data to determine individual disease risk and interpret the most important risk factors using mean SHAP score. IDEARS is applied to the UKB to determine the risk factors for PD. RESULTS IDEARS showed an improved discriminative performance (AUC=0.744) compared to a model using known risk factors (0.727). Age and gender had the highest mean SHAP scores. IGF-1, bilirubin, neutrophil/lymphocyte ratio (NLR, an inflammatory marker) and frailty factors were also ranked highly. A further investigation into IGF-1, bilirubin, AST:ALT and NLR showed elevated levels either in the period prior to diagnosis or at the point of diagnosis. DISCUSSION The IDEARS model outperformed an approach looking only at agreed risk factors despite making no a priori assumptions. Novel PD risk biomarkers including elevated IGF-1 and NLR are likely to play a role in disease mechanism. This panel of biomarkers may be used clinically to predict future PD risk, improve early diagnosis and to understand disease mechanism.