Machine learning analysis of the UK Biobank reveals prognostic and diagnostic immune biomarkers for polyneuropathy and neuropathic pain in diabetes

Abstract

AimsWe assessed the health data of 11,047 people with diabetes in the UK Biobank to rank 329 risk factors for diabetic polyneuropathy (DPN) and DPN with chronic neuropathic pain without a priori assumption. MethodsThe Integrated Disease Explanation and Risk Scoring (IDEARS) platform applies machine learning algorithms to multimodal data to determine individual disease risk, and rank risk factor importance using mean SHapley Additive exPlanations (SHAP) score. ResultsIDEARS models showed discriminative performances with AUC > 0.64. Lower socioeconomic status, being overweight, poor overall health, cystatin C, HbA1C, and immune activation marker, C-reactive protein (CRP), predict DPN risk. Neutrophils and monocytes were higher in males and lymphocytes lower in females with diabetes that develop DPN. Neutrophil-to-Lymphocyte Ratio (NLR) was increased and IGF-1 levels decreased in people with type 2 diabetes that later develop DPN. CRP was significantly elevated in those with DPN and chronic neuropathic pain compared to DPN without pain. ConclusionsLifestyle factors and blood biomarkers predict the later development of DPN and may relate to DPN pathomechanisms. Our results are consistent with DPN as a disease involving systemic inflammation. We advocate for the use of these biomarkers clinically to predict future DPN risk and improve early diagnosis.