Nuffield Department of Population Health (NDPH), University of Oxford, Oxford, UK
Nuffield Department of Population Health (NDPH), University of Oxford, Oxford, UK.
J Epidemiol Community Health. 2023 Dec;77(12):802-808. doi: 10.1136/jech-2023-220801. Epub 2023 Sep 12.
Metabolic profiling (the extensive measurement of circulating metabolites across multiple biological pathways) is increasingly employed in clinical care. However, there is little evidence on the benefit of metabolic profiling as compared with established atherosclerotic cardiovascular disease (CVD) risk scores.
UK Biobank is a prospective study of 0.5 million participants, aged 40-69 at recruitment. Analyses were restricted to 74 780 participants with metabolic profiling (measured using nuclear magnetic resonance) and without CVD at baseline. Cox regression was used to compare model performance before and after addition of metabolites to QRISK3 (an established CVD risk score used in primary care in England); analyses derived three models, with metabolites selected by association significance or by employing two different machine learning approaches.
We identified 5097 incident CVD events within the 10-year follow-up. Harrell's C-index of QRISK3 was 0.750 (95% CI 0.739 to 0.763) for women and 0.706 (95% CI 0.696 to 0.716) for men. Adding selected metabolites did not significantly improve measures of discrimination in women (Harrell's C-index of three models are 0.759 (0.747 to 0.772), 0.759 (0.746 to 0.770) and 0.759 (0.748 to 0.771), respectively) or men (0.710 (0.701 to 0.720), 0.710 (0.700 to 0.719) and 0.710 (0.701 to 0.719), respectively), and neither did it improve reclassification or calibration.
This large-scale study applied both conventional and machine learning approaches to assess the potential benefit of metabolic profiling to well-established CVD risk scores. However, there was no evidence that metabolic profiling improved CVD risk prediction in this population.
代谢组学（对多个生物途径中的循环代谢物进行广泛测量）在临床护理中越来越多地被采用。然而，与已建立的动脉粥样硬化性心血管疾病（CVD）风险评分相比，代谢组学的益处证据甚少。
英国生物库是一项针对 50 万名年龄在 40-69 岁的参与者的前瞻性研究。分析仅限于基线时无 CVD 的 74780 名具有代谢组学（使用磁共振进行测量）的参与者。Cox 回归用于比较在 QRISK3（一种在英格兰初级保健中使用的已建立的 CVD 风险评分）中添加代谢物前后的模型性能；分析得出了三个模型，通过关联显著性选择代谢物，或通过采用两种不同的机器学习方法选择代谢物。
在 10 年的随访中，我们确定了 5097 例 CVD 事件。QRISK3 的 Harrell C 指数在女性中为 0.750（95%CI 0.739 至 0.763），在男性中为 0.706（95%CI 0.696 至 0.716）。添加选定的代谢物并未显著提高女性的判别力指标（三个模型的 Harrell C 指数分别为 0.759（0.747 至 0.772）、0.759（0.746 至 0.770）和 0.759（0.748 至 0.771））或男性（0.710（0.701 至 0.720）、0.710（0.700 至 0.719）和 0.710（0.701 至 0.719）），也没有改善重新分类或校准。
这项大规模研究应用了传统和机器学习方法来评估代谢组学对既定 CVD 风险评分的潜在益处。然而，没有证据表明代谢组学改善了该人群的 CVD 风险预测。
