Development and validation of machine learning models with blood-based digital biomarkers for Alzheimer's disease diagnosis: a multicohort diagnostic study.
{"title":"Development and validation of machine learning models with blood-based digital biomarkers for Alzheimer's disease diagnosis: a multicohort diagnostic study.","authors":"Bin Jiao, Ziyu Ouyang, Xuewen Xiao, Cong Zhang, Tianyan Xu, Qijie Yang, Yuan Zhu, Yiliang Liu, Xixi Liu, Yafang Zhou, Xinxin Liao, Shilin Luo, Beisha Tang, Zhigang Li, Lu Shen","doi":"10.1016/j.eclinm.2025.103142","DOIUrl":null,"url":null,"abstract":"<p><strong>Background: </strong>Alzheimer's disease (AD) involves complex alterations in biological pathways, making comprehensive blood biomarkers crucial for accurate and earlier diagnosis. However, the cost-effectiveness and operational complexity of method using blood-based biomarkers significantly limit its availability in clinical practice.</p><p><strong>Methods: </strong>We developed low-cost, convenient machine learning-based with digital biomarkers (MLDB) using plasma spectra data to detect AD or mild cognitive impairment (MCI) from healthy controls (HCs) and discriminate AD from different types of neurodegenerative diseases. Retrospective data were gathered for 1324 individuals, including 293 with amyloid beta positive AD, 151 with mild cognitive impairment (MCI), 106 with Lewy body dementia (DLB), 106 with frontotemporal dementia (FTD), 135 with progressive supranuclear palsy (PSP) and 533 healthy controls (HCs) between July 2017 and August 2023.</p><p><strong>Findings: </strong>Random forest classifier and feature selection procedures were used to select digital biomarkers. MLDB achieved area under the curves (AUCs) of 0.92 (AD vs. HC, Sensitivity 88.2%, specificity 84.1%), 0.89 (MCI vs. HC, Sensitivity 88.8%, specificity 86.4%), 0.83 (AD vs. DLB, Sensitivity 77.2%, specificity 74.6%), 0.80 (AD vs. FTD, sensitivity 74.2%, specificity 72.4%), and 0.93 (AD vs. PSP, sensitivity 76.1%, specificity 75.7%). Digital biomarkers distinguishing AD from HC were negatively correlated with plasma p-tau217 (<i>r</i> = -0.22, <i>p</i> < 0.05) and glial fibrillary acidic protein (GFAP) (<i>r</i> = -0.09, <i>p</i> < 0.05).</p><p><strong>Interpretation: </strong>The ATR-FTIR (Attenuated Total Reflectance-Fourier Transform Infrared) plasma spectra features can identify AD-related pathological changes. These spectral features serve as digital biomarkers, providing valuable support in the early screening and diagnosis of AD.</p><p><strong>Funding: </strong>The National Natural Science Foundation of China, STI2030-Major Projects, National Key R&D Program of China, Outstanding Youth Fund of Hunan Provincial Natural Science Foundation, Hunan Health Commission Grant, Science and Technology Major Project of Hunan Province, Hunan Innovative Province Construction Project, Grant of National Clinical Research Center for Geriatric Disorders, Xiangya Hospital and Postdoctoral Fellowship Program of CPSF.</p>","PeriodicalId":11393,"journal":{"name":"EClinicalMedicine","volume":"81 ","pages":"103142"},"PeriodicalIF":9.6000,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11925590/pdf/","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"EClinicalMedicine","FirstCategoryId":"3","ListUrlMain":"https://doi.org/10.1016/j.eclinm.2025.103142","RegionNum":1,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2025/3/1 0:00:00","PubModel":"eCollection","JCR":"Q1","JCRName":"MEDICINE, GENERAL & INTERNAL","Score":null,"Total":0}
引用次数: 0
Abstract
Background: Alzheimer's disease (AD) involves complex alterations in biological pathways, making comprehensive blood biomarkers crucial for accurate and earlier diagnosis. However, the cost-effectiveness and operational complexity of method using blood-based biomarkers significantly limit its availability in clinical practice.
Methods: We developed low-cost, convenient machine learning-based with digital biomarkers (MLDB) using plasma spectra data to detect AD or mild cognitive impairment (MCI) from healthy controls (HCs) and discriminate AD from different types of neurodegenerative diseases. Retrospective data were gathered for 1324 individuals, including 293 with amyloid beta positive AD, 151 with mild cognitive impairment (MCI), 106 with Lewy body dementia (DLB), 106 with frontotemporal dementia (FTD), 135 with progressive supranuclear palsy (PSP) and 533 healthy controls (HCs) between July 2017 and August 2023.
Findings: Random forest classifier and feature selection procedures were used to select digital biomarkers. MLDB achieved area under the curves (AUCs) of 0.92 (AD vs. HC, Sensitivity 88.2%, specificity 84.1%), 0.89 (MCI vs. HC, Sensitivity 88.8%, specificity 86.4%), 0.83 (AD vs. DLB, Sensitivity 77.2%, specificity 74.6%), 0.80 (AD vs. FTD, sensitivity 74.2%, specificity 72.4%), and 0.93 (AD vs. PSP, sensitivity 76.1%, specificity 75.7%). Digital biomarkers distinguishing AD from HC were negatively correlated with plasma p-tau217 (r = -0.22, p < 0.05) and glial fibrillary acidic protein (GFAP) (r = -0.09, p < 0.05).
Interpretation: The ATR-FTIR (Attenuated Total Reflectance-Fourier Transform Infrared) plasma spectra features can identify AD-related pathological changes. These spectral features serve as digital biomarkers, providing valuable support in the early screening and diagnosis of AD.
Funding: The National Natural Science Foundation of China, STI2030-Major Projects, National Key R&D Program of China, Outstanding Youth Fund of Hunan Provincial Natural Science Foundation, Hunan Health Commission Grant, Science and Technology Major Project of Hunan Province, Hunan Innovative Province Construction Project, Grant of National Clinical Research Center for Geriatric Disorders, Xiangya Hospital and Postdoctoral Fellowship Program of CPSF.
背景：阿尔茨海默病（AD）涉及生物通路的复杂改变，因此全面的血液生物标志物对准确和早期诊断至关重要。然而，使用基于血液的生物标志物的方法的成本效益和操作复杂性极大地限制了其在临床实践中的可用性：我们利用血浆光谱数据开发了低成本、便捷的基于机器学习的数字生物标记物（MLDB），用于从健康对照（HCs）中检测出注意力缺失症或轻度认知障碍（MCI），并将注意力缺失症与不同类型的神经退行性疾病区分开来。研究人员在2017年7月至2023年8月期间收集了1324人的回顾性数据，其中包括293名淀粉样β阳性AD患者、151名轻度认知障碍患者、106名路易体痴呆患者、106名额颞叶痴呆患者、135名进行性核上性麻痹患者和533名健康对照者：随机森林分类器和特征选择程序用于选择数字生物标记物。MLDB的曲线下面积（AUC）分别为0.92（AD vs. HC，灵敏度88.2%，特异度84.1%）、0.89（MCI vs. HC，灵敏度88.8%，特异度86.4%）、0.83（AD vs. DLB，灵敏度77.2%，特异度74.6%）、0.80（AD vs. FTD，灵敏度74.2%，特异度72.4%）和0.93（AD vs. PSP，灵敏度76.1%，特异度75.7%）。区分 AD 和 HC 的数字生物标志物与血浆 p-tau217 呈负相关（r = -0.22，p r = -0.09，p 解释：ATR-FTIR（衰减全反射-傅立叶变换红外）血浆光谱特征可识别与AD相关的病理变化。这些光谱特征可作为数字生物标志物，为AD的早期筛查和诊断提供有价值的支持：国家自然科学基金、STI2030-重大项目、国家重点研发计划、湖南省自然科学基金杰出青年基金、湖南省卫生计生委资助项目、湖南省科技重大专项、湖南省创新型省份建设项目、湘雅医院国家老年疾病临床医学研究中心资助项目、中国公共卫生联合会博士后基金项目。
期刊介绍：
eClinicalMedicine is a gold open-access clinical journal designed to support frontline health professionals in addressing the complex and rapid health transitions affecting societies globally. The journal aims to assist practitioners in overcoming healthcare challenges across diverse communities, spanning diagnosis, treatment, prevention, and health promotion. Integrating disciplines from various specialties and life stages, it seeks to enhance health systems as fundamental institutions within societies. With a forward-thinking approach, eClinicalMedicine aims to redefine the future of healthcare.
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号
