Uncovering predictors of low hippocampal volume: Evidence from a large-scale machine-learning-based study in the UK Biobank.

INTRODUCTION: Hippocampal atrophy is an established biomarker for conversion from the normal ageing process to developing cognitive impairment and dementia. This study used a novel hypothesis-free machine-learning approach, to uncover potential risk factors of lower hippocampal volume using information from the world's largest brain imaging study. METHODS: A combination of machine learning and conventional statistical methods were used to identify predictors of low hippocampal volume. We run gradient boosting decision tree modelling including 2891 input features measured before magnetic resonance imaging assessments (median 9.2 years, range 4.2-13.8 years) using data from 42,152 dementia-free UK Biobank participants. Logistic regression analyses were run on 87 factors identified as important for prediction based on Shapley values. False discovery rate adjusted P-value <0.05 was used to declare statistical significance. RESULTS: Older age, male sex, greater height, and whole-body fat free mass were the main predictors of low hippocampal volume with the model also identifying associations with lung function and lifestyle factors including smoking, physical activity, and coffee intake (corrected P<0.05 for all). Red blood cell count and several red blood cell indices such as haemoglobin concentration, mean corpuscular haemoglobin, mean corpuscular volume, mean reticulocyte volume, mean sphered cell volume, and red blood cell distribution width were among many biomarkers associated with low hippocampal volume. CONCLUSION: Lifestyles, physical measures, and biomarkers may affect hippocampal volume, with many of the characteristics potentially reflecting oxygen supply to the brain. Further studies are required to establish causality and clinical relevance of these findings.

Phenome-wide association study of ovarian cancer identifies common comorbidities and reveals shared genetics with complex diseases and biomarkers.

BACKGROUND: Ovarian cancer (OC) is commonly diagnosed among older women who have comorbidities. This hypothesis-free phenome-wide association study (PheWAS) aimed to identify comorbidities associated with OC, as well as traits that share a genetic architecture with OC. METHODS: We used data from 181,203 white British female UK Biobank participants and analysed OC and OC subtype-specific genetic risk scores (OC-GRS) for an association with 889 diseases and 43 other traits. We conducted PheWAS and colocalization analyses for individual variants to identify evidence for shared genetic architecture. RESULTS: The OC-GRS was associated with 10 diseases, and the clear cell OC-GRS was associated with five diseases at the FDR threshold (p = 5.6 × 10-4 ). Mendelian randomizaiton analysis (MR) provided robust evidence for the association of OC with higher risk of "secondary malignant neoplasm of digestive systems" (OR 1.64, 95% CI 1.33, 2.02), "ascites" (1.48, 95% CI 1.17, 1.86), "chronic airway obstruction" (1.17, 95% CI 1.07, 1.29), and "abnormal findings on examination of the lung" (1.51, 95% CI 1.22, 1.87). Analyses of lung spirometry measures provided further support for compromised respiratory function. PheWAS on individual OC variants identified five genetic variants associated with other diseases, and seven variants associated with biomarkers (all, p ≤ 4.5 × 10-8 ). Colocalization analysis identified rs4449583 (from TERT locus) as the shared causal variant for OC and seborrheic keratosis. CONCLUSIONS: OC is associated with digestive and respiratory comorbidities. Several variants affecting OC risk were associated with other diseases and biomarkers, with this study identifying a novel genetic locus shared between OC and skin conditions.

Hypothesis-free discovery of novel cancer predictors using machine learning.

BACKGROUND: Cancer is a leading cause of morbidity and mortality worldwide, and better understanding of the risk factors could enhance prevention. METHODS: We conducted a hypothesis-free analysis combining machine learning and statistical approaches to identify cancer risk factors from 2828 potential predictors captured at baseline. There were 459,169 UK Biobank participants free from cancer at baseline and 48,671 new cancer cases during the 10-year follow-up. Logistic regression models adjusted for age, sex, ethnicity, education, material deprivation, smoking, alcohol intake, body mass index and skin colour (as a proxy for sun sensitivity) were used for obtaining adjusted odds ratios, with continuous predictors presented using quintiles (Q). RESULTS: In addition to smoking, older age and male sex, positively associating features included several anthropometric characteristics, whole body water mass, pulse, hypertension and biomarkers such as urinary microalbumin (Q5 vs. Q1 OR 1.16, 95% CI = 1.13-1.19), C-reactive protein (Q5 vs. Q1 OR 1.20, 95% CI = 1.16-1.24) and red blood cell distribution width (Q5 vs. Q1 OR 1.18, 95% CI = 1.14-1.21), among others. High-density lipoprotein cholesterol (Q5 vs. Q1 OR 0.84, 95% CI = 0.81-0.87) and albumin (Q5 vs. Q1 OR 0.84, 95% CI = 0.81-0.87) were inversely associated with cancer. In sex-stratified analyses, higher testosterone increased the risk in females but not in males (Q5 vs. Q1 ORfemales 1.23, 95% CI = 1.17-1.30). Phosphate was associated with a lower risk in females but a higher risk in males (Q5 vs. Q1 ORfemales 0.94, 95% CI = 0.90-0.99 vs. ORmales 1.09, 95% CI 1.04-1.15). CONCLUSIONS: This hypothesis-free analysis suggests personal characteristics, metabolic biomarkers, physical measures and smoking as important predictors of cancer risk, with further studies needed to confirm causality and clinical relevance.

Combining machine learning and conventional statistical approaches for risk factor discovery in a large cohort study.

We present a simple and efficient hypothesis-free machine learning pipeline for risk factor discovery that accounts for non-linearity and interaction in large biomedical databases with minimal variable pre-processing. In this study, mortality models were built using gradient boosting decision trees (GBDT) and important predictors were identified using a Shapley values-based feature attribution method, SHAP values. Cox models controlled for false discovery rate were used for confounder adjustment, interpretability, and further validation. The pipeline was tested using information from 502,506 UK Biobank participants, aged 37-73 years at recruitment and followed over seven years for mortality registrations. From the 11,639 predictors included in GBDT, 193 potential risk factors had SHAP values ≥ 0.05, passed the correlation test, and were selected for further modelling. Of the total variable importance summed up, 60% was directly health related, and baseline characteristics, sociodemographics, and lifestyle factors each contributed about 10%. Cox models adjusted for baseline characteristics, showed evidence for an association with mortality for 166 out of the 193 predictors. These included mostly well-known risk factors (e.g., age, sex, ethnicity, education, material deprivation, smoking, physical activity, self-rated health, BMI, and many disease outcomes). For 19 predictors we saw evidence for an association in the unadjusted but not adjusted analyses, suggesting bias by confounding. Our GBDT-SHAP pipeline was able to identify relevant predictors 'hidden' within thousands of variables, providing an efficient and pragmatic solution for the first stage of hypothesis free risk factor identification.