Dong Bosi, He Mengqiao, Ji Shuming, Yang Wenjie, Hong Qiulei, Tang Yusha, Peng Anjiao, Chen Lei
Department of Neurology, West China Hospital, Sichuan University, No. 37 Guo Xue Xiang, Chengdu, 610041, Sichuan, China.
Department of Clinical Research Management, West China Hospital, Sichuan University, Chengdu, 610044, Sichuan, China.
Sci Rep. 2025 Mar 5;15(1):7700. doi: 10.1038/s41598-025-92764-5.
Cognitive impairment is a common health problem. However, it is often ignored in primary healthcare due to complex examination. To develop biomarkers through easily accessible blood tests, we conducted a cross-sectional analysis including 2806 healthy Chinese adults aged over 60 years old who finished the cognition assessment though Mini-Mental State Examination as well as blood routine and biochemical examination from four communities in China between July 2020 and September 2021. The blood biomarkers commonly selected by RFE and LASSO with cross-validation in the training dataset were levels of hemoglobin, high density lipoprotein, alkaline phosphatase, direct bilirubin, globulin, creatinine, magnesium, and calcium. We developed XGBoost, logistic regression and SVM models with cross-validation in the training dataset and then evaluated their performance by the receiver operating characteristic curves, F1 score and decision curve analysis in the test dataset. The accuracies of the eight biomarkers in XGBoost, logistic regression and SVM models were 0.880 (0.846-0.915), 0.851 (0.812-0.889), and 0.852 (0.814-0.890) with confirmed clinical utility to predict cognitive impairment separately. Cognitive impairment can be predicted based on blood routine and biochemical examination with good discrimination, which could be helpful in detection and intervention at primary care.
认知障碍是一个常见的健康问题。然而，由于检查复杂，它在基层医疗保健中常常被忽视。为了通过易于进行的血液检测开发生物标志物，我们进行了一项横断面分析，纳入了2806名60岁以上的中国健康成年人，这些人于2020年7月至2021年9月期间在中国四个社区完成了简易精神状态检查以及血常规和生化检查。在训练数据集中通过带有交叉验证的递归特征消除法（RFE）和最小绝对收缩和选择算子法（LASSO）共同选择的血液生物标志物有血红蛋白水平、高密度脂蛋白、碱性磷酸酶、直接胆红素、球蛋白、肌酐、镁和钙。我们在训练数据集中开发了带有交叉验证的极端梯度提升（XGBoost）、逻辑回归和支持向量机（SVM）模型，然后在测试数据集中通过受试者工作特征曲线、F1分数和决策曲线分析评估它们的性能。XGBoost、逻辑回归和SVM模型中这八种生物标志物的预测认知障碍的准确率分别为0.880（0.846至0.915）、0.851（0.812至0.889）和0.852（0.814至0.890），均具有经确认的临床效用。基于血常规和生化检查可以较好地区分预测认知障碍，这有助于基层医疗保健中的检测和干预。
