基于机器学习算法的帕金森病患者脑深部电刺激术后认知状态改善的预测模型构建与验证-东南大学学报医学版

基于机器学习算法的帕金森病患者脑深部电刺激术后认知状态改善的预测模型构建与验证

单位：秦皇岛市第一医院康复科, 河北秦皇岛 066000

分类号：R742.5;R195.1

出版年·卷·期（页码）：2026·45·第三期（393-405）

摘要：

目的: 探讨基于机器学习(ML)算法构建帕金森病(PD)患者脑深部电刺激(DBS)术后认知状态改善预测模型的可行性与效能。方法: 前瞻性选取2022年7月至2024年7月期间于本院行DBS治疗的315例PD患者为研究对象。采用R软件将所有数据集按7∶3比例随机划分为训练集(n=219)与验证集(n=96)。以术后1年简易精神状态检查(MMSE)评分改善率≥12%定义为认知状态改善。基于训练集采用逻辑回归(LR)、决策树(DT)、极端梯度提升(XGBoost)和随机森林(RF)构建PD患者DBS术后认知状态改善的预测模型。基于验证集采用DeLong检验评估各模型的曲线下面积(AUC),并采用决策曲线分析(DCA)评估最优模型的临床净收益。最后,应用沙普利加和解释法(SHAP)对最优模型的预测结果进行可视化解释。结果: 模型性能评估结果显示,基于PD病程、术前MMSE评分、汉密尔顿焦虑量表(HAMA)评分、汉密尔顿抑郁量表(HAMD)评分、术后颅内气体体积构建的RF算法模型在验证集上预测PD患者DBS术后认知状态改善的AUC为0.923,优于LR(0.629)、DT(0.822)以及XGBoost模型(0.921)。RF算法模型训练集和验证集均在阈值为0.1时获得最大净收益(训练集0.263 8,验证集0.269 7)。SHAP可解释性分析显示,术后颅内气体体积、HAMA评分、术前MMSE评分、HAMD评分、PD病程依次是影响模型预测的重要特征。结论: RF模型在预测PD患者DBS术后认知状态改善方面具有最优的判别效能与临床实用性,且术后颅内气体体积、HAMA评分、术前MMSE评分、HAMD评分、PD病程依次是其预测的关键因子。

Objective: To investigate the feasibility and performance of machine learning(ML) algorithms in constructing a predictive model for cognitive improvement following deep brain stimulation(DBS) surgery in patients with Parkinson disease(PD). Methods: A prospective cohort of 315 PD patients who underwent DBS at our hospital between July 2022 and July 2024 was enrolled. Using R software, the dataset was randomly partitioned into a training set(n=219) and a validation set(n=96) at a 7∶3 ratio. Cognitive improvement was defined as an increase of ≥12% in the Mini-Mental State Examination(MMSE) score at one year postoperatively. Four ML models: Logistic regression(LR), decision tree(DT), extreme gradient boosting(XGBoost), and random forest(RF) were developed using the training set to predict post-DBS cognitive improvement. Model performance on the validation set was evaluated by comparing the area under the receiver operating characteristic curve(AUC) using the DeLong test. Decision curve analysis(DCA) was employed to assess the clinical net benefit of the optimal model. Finally, Shapley additive explanations(SHAP) were used to provide a visual interpretation of the optimal model's predictions. Results: Performance evaluation revealed that the RF model constructed using five predictors[PD duration, preoperative MMSE score, Hamilton Anxiety Rating Scale(HAMA) score, Hamilton Depression Rating Scale(HAMD) score, and postoperative intracranial gas volume achieved the highest AUC of 0.923 on the validation set, outperforming the LR(0.629), DT(0.822), and XGBoost(0.921) models. DCA indicated that the RF model yielded the maximum net clinical benefit at a threshold probability of 0.1(net benefit: 0.263 8 in the training set and 0.269 7 in the validation set). SHAP interpretability analysis identified postoperative intracranial gas volume, HAMA score, preoperative MMSE score, HAMD score, and PD duration as the top five features influencing the model's predictions, in descending order of importance. Conclusion: The RF model demonstrates superior discriminative ability and clinical utility for predicting cognitive improvement after DBS in PD patients. Postoperative intracranial gas volume, HAMA score, preoperative MMSE score, HAMD score, and PD duration are the key predictive factors.

参考文献：

[1] 郝洁,刘昌佳,唐进松,等.帕金森病患者血清Sirtuin1、SGK1水平与病情分期及认知障碍的关系[J].东南大学学报(医学版),2025,44(4):586-592.
[2] 蔡竹清,唐春梅,曾敏.帕金森患者抑郁及认知功能障碍的临床特点及影响因素分析[J].实用医院临床杂志,2023,20(3):125-129.
[3] KAWABATA K,DJAMSHIDIAN A,BAGARINAO E,et al.Cognitive dysfunction in de novo Parkinson disease:Remitting vs.progressive cognitive impairment[J].Park Relat Disord,2024,120:105984.
[4] 包彧雯,王滢鹏,万彬,等.帕金森病脑深部电刺激术联合药物治疗的卫生经济学评估[J].现代医学,2023,51(10):1448-1452.
[5] VALK-GEUKE M E H,GEURTSEN G J,DE BIE R M A,et al.The effect of deep brain stimulation on sleep and cognition in patients with Parkinson's disease[J].J Sleep Res,2026,35(2):e70193.
[6] CHANG B,NI C,ZHANG W,et al.Nomogram to predict cognitive state improvement after deep brain stimulation for Parkinson's disease[J].Brain Sci,2022,12(6):759.
[7] HUANG Y,MAO Y,XU L,et al.Exploring risk factors for cervical lymph node metastasis in papillary thyroid microcarcinoma:construction of a novel population-based predictive model[J].BMC Endocr Disord,2022,22(1):269.
[8] 中华医学会神经病学分会帕金森病及运动障碍学组,中国医师协会神经内科医师分会帕金森病及运动障碍专业.中国帕金森病的诊断标准(2016版)[J].中华神经科杂志,2016,49(4):268-271.
[9] 林晓红.脑深部电刺激术对帕金森病患者认知功能障碍的影响及预后研究[D].济南:山东大学,2023.
[10] PAN H,WANG X,ZHANG X,et al.Investigating the association of pan-immune-inflammation value,systemic immune-inflammation index,and neutrophil-to-lymphocyte ratio with pain in Parkinson's disease[J].Front Neurol,2025,16:1682964.
[11] JIA X,WANG Z,HUANG F,et al.A comparison of the Mini-Mental State Examination(MMSE) with the Montreal Cognitive Assessment(MoCA) for mild cognitive impairment screening in Chinese middle-aged and older population:a cross-sectional study[J].BMC Psychiatry,2021,21(1):485.
[12] HE Y,CHENG S,YANG L,et al.Associations between plasma markers and symptoms of anxiety and depression in patients with breast cancer[J].BMC Psychiatry,2024,24(1):678.
[13] LI B,ZHANG C,CHEN W,et al.Amygdala volume abnormalities and cognitive impairment in major depressive disorder and bipolar disorder II[J].BMC Psychiatry,2025,25(1):839.
[14] YANG W,ZHANG Q,GAO C,et al.Effect and clinical mechanism exploration of acupuncture intervention for chemotherapy-related cognitive impairment(CRCI) in triple-negative breast cancer:study protocol for a randomized controlled trial[J].Front Neurol,2025,16:1565040.
[15] 朱宏,董克礼,吴岳,等.补肾活血法对阿尔茨海默病患者认知功能改善的影响[J].中国老年学杂志,2010,30(11):1493-1495.
[16] FARZANEHFAR P,WOODROW H,HORNE M.Sensor measurements can characterize fluctuations and wearing off in Parkinson's disease and guide therapy to improve motor,non-motor and quality of life scores[J].Front Aging Neurosci,2022,14:852992.
[17] LEIMBACH F,ATKINSON-CLEMENT C,SOCORRO P,et al.The effects of subthalamic nucleus deep brain stimulation in Parkinson's disease on associative learning of verbal and non-verbal information by trial and error or with corrective feedback[J].J Park Dis,2022,12(3):885-896.
[18] BOURILHON J,MULLIE Y,OLIVIER C,et al.Stimulation of the pedunculopontine and cuneiform nuclei for freezing of gait and falls in Parkinson disease:Cross-over single-blinded study and long-term follow-up[J].Park Relat Disord,2022,96:13-17.
[19] 范建华,王培娟,吴郁琼.老年帕金森病患者发生认知障碍的影响因素研究[J].临床心身疾病杂志,2016,22(6):94-96.
[20] MYERS P S,O'DONNELL J L,JACKSON J J,et al.Proteinopathy and longitudinal cognitive decline in Parkinson disease[J].Neurology,2022,99(1):e66-e76.
[21] FUKAYA C,WATANABE M,KOBAYASHI K,et al.Predictive factors for long-term outcome of subthalamic nucleus deep brain stimulation for Parkinson's disease[J].Neurol Med Chir(Tokyo),2017,57(4):166-171.
[22] LIU C,LI H,FENG S,et al.Alterations in structural and functional magnetic resonance imaging associated with cognitive function in patients with treatment-nave first-episode major depressive disorder[J].Prog Neuropsychopharmacol Biol Psychiatry,2025,139:111367.
[23] LUO B,ZOU Y,YAN J,et al.Altered cognitive networks connectivity in Parkinson's disease during the microlesion period after deep brain stimulation[J].CNS Neurosci Ther,2024,30(12):e70184.
[24] 谢思宁,蒋忱冠,李享佳卉,等.帕金森病患者脑深部电刺激术后神经精神功能障碍发生率及其影响因素分析:一项前瞻性队列研究[J].协和医学杂志,2025,16(2):300-306.
[25] BRAAK H,DEL TREDICI K,RüB U,et al.Staging of brain pathology related to sporadic Parkinson's disease[J].Neurobiol Aging,2003,24(2):197-211.
[26] ZHOU C,GUO T,BAI X,et al.Locus coeruleus degeneration is associated with disorganized functional topology in Parkinson's disease[J].Neuroimage Clin,2021,32:102873.
[27] CAREY G,GÖRMEZOLU M,DE JONG J J A,et al.Neuroimaging of anxiety in Parkinson's disease:a systematic review[J].Mov Disord,2021,36(2):327-339.
[28] YANG N,JU Y,REN J,et al.Prevalence and affective correlates of subjective cognitive decline in patients with de novo Parkinson's disease[J].Acta Neurol Scand,2022,146(3):276-282.
[29] PRANGE S,KLINGER H,LAURENCIN C,et al.Depression in patients with Parkinson's disease:current understanding of its neurobiology and implications for treatment[J].Drugs Aging,2022,39(6):417-439.

服务与反馈：

【文章下载】【发表评论】【查看评论】【加入收藏】

提示：您还未登录，请登录！点此登录