PLOS ONE · 9 authors, 3 centres
This summary was generated by AI from a single paper. It has not been reviewed by a clinician and is not clinical advice. Verify against the source before acting on it.
Using machine learning on UK Biobank data from 334,062 participants, the study identified elevated insulin-like growth factor 1 (IGF-1) and neutrophil-to-lymphocyte ratio (NLR) as novel biomarkers that predict Parkinson's disease risk up to 10 years before diagnosis. These findings suggest that systemic inflammation and dysregulated IGF-1 signaling may play key roles in PD pathogenesis, offering potential targets for early diagnosis and preventive therapies.
**Background:** Parkinson's disease (PD) is the second most common neurodegenerative disease, affecting over 6 million people worldwide with a 3-fold increase in prevalence over the last 30 years. Identifying blood-borne biomarkers and environmental risk factors is critical for early diagnosis and intervention, as neurodegenerative processes begin years before clinical diagnosis. Previous studies have used univariate approaches or logistic regression, which cannot model interactions among many variables. The UK Biobank, the largest deeply phenotyped epidemiological study globally, provides an opportunity to apply machine learning to explore a wide range of risk factors without a priori assumptions.
**Methods:** The study used the Integrated Disease Explanation and Risk Scoring platform (IDEARS), which combines XGBoost classification with SHAP feature importance. From 502,253 UK Biobank participants, 334,062 met inclusion criteria (aged 50-70 at baseline, no PD diagnosis within 2 years of baseline, not deceased within 10 years from non-PD causes). Among these, 2,719 developed PD (ICD10 G20) during a mean follow-up of 8.0 years. The model analyzed 1,753 variables including blood assays, clinical measurements, ICD10 codes, and questionnaire data. Three model types were compared: XGBoost (mean AUC 0.67), Support Vector Machines (0.66), random forest (0.65), and logistic regression (0.59). XGBoost was selected for subsequent analysis. Models were trained on 80% of data and validated on 20% holdout, with 100 resamples. Gender-stratified analyses were also performed.
**Key Results:** The IDEARS model (using all variables) achieved a mean AUC of 0.672, significantly higher than the model using only established risk factors (AUC 0.620, P<0.001). Male gender was the highest-ranked risk factor. The top novel biomarkers included: elevated serum IGF-1 (3rd overall), neutrophil/lymphocyte ratio (15th), and neutrophil percentage (18th). Frailty-related features ranked highly: self-reported overall health rating (2nd), usual walking pace (4th), number of treatments/medications (5th), and hand grip strength (27th). Protective factors included elevated total cholesterol (16th), C-reactive protein (7th), and lymphocyte count (10th). In gender-stratified analyses, IGF-1 was the top feature in males (1st) and 3rd in females. NLR was elevated in both sexes up to 10 years before diagnosis: in males, NLR was 2.72±1.32 at -10 to -5 years vs. 2.48±1.25 in non-PD (P<0.001); in females, 2.37±2.10 vs. 2.19±1.08 (P<0.001). Ibuprofen use showed a protective effect, most pronounced at high NLR levels. Urate levels were significantly reduced in males at all time points (e.g., 343.00±74.22 at -10 to -5 years vs. 354.63±70.96, P<0.001). Total cholesterol was reduced in males at -10 to -5 years (5.19±1.11 vs. 5.47±1.13, P<0.001). Frailty indicators such as total ICD10 diagnoses and grip strength were significantly worse in PD cases up to 10 years before diagnosis (all P<0.001).
**Clinical Implications:** This study identifies IGF-1 and NLR as novel, robust biomarkers that predict PD risk up to a decade before clinical diagnosis, particularly in males. The findings support the hypothesis that PD may be a central manifestation of systemic inflammatory disease, with peripheral inflammation and dysregulated IGF-1 signaling playing key roles. The protective effect of ibuprofen, especially in individuals with high NLR, suggests anti-inflammatory interventions may reduce PD risk. The biomarker panel (IGF-1, NLR, AST:ALT ratio, urate, cholesterol) could be used clinically for risk stratification and early diagnosis, enabling timely lifestyle modifications or future neuroprotective therapies. Frailty indicators such as reduced grip strength and walking speed are simple clinical measures that may aid in identifying at-risk individuals. The study's limitations include its observational nature (no causality inferred) and the predominantly white European cohort, which may limit generalizability. Future research should explore causal relationships using Bayesian networks and validate these biomarkers in diverse populations.