Construction and Validation of a Risk Prediction Model for Cerebral Small Vessel Disease Progression Based on Multimodal MRI Combined with Serum sRAGE
-
摘要:
目的 构建多模态MRI联合血清可溶性晚期糖基化终末产物受体(soluble receptor for advanced glycation end products,sRAGE)的血管性脑白质病变(white matter lesions,WML)进展风险预测模型,并验证其预测效能。 方法 采用回顾性队列研究设计,纳入2020年1月至2023年10月邯郸市第一医院确诊的WML患者330例,按7∶3比例随机分为建模集(n = 231)和验证集(n = 99)。收集患者一般资料、基础疾病、实验室指标(含血清sRAGE)及多模态磁共振成像(magnetic resonance imaging,MRI)指标(Fazekas评分、FLAIR病变体积等)共18项候选预测因子,以随访2年内病变是否进展为结局事件。经LASSO回归筛选核心因子,采用多因素Logistic回归构建列线图预测模型,通过受试者工作特征(receiver operating characteristic,ROC)曲线、Hosmer-Lemeshow检验及决策曲线分析(decision curve analysis,DCA)分别评估模型的区分度、校准度及临床适用性。 结果 建模集和验证集病变进展率分别为21.6%(50/231)和22.2%(22/99)。LASSO回归筛选出5项核心预测因子,年龄≥70岁、糖尿病、血清sRAGE水平降低、Fazekas评分≥3分、FLAIR病变体积≥5 mL。多因素Logistic回归显示,年龄≥70岁、糖尿病、Fazekas评分≥3分、血清sRAGE水平降低、FLAIR病变体积≥5 mL为WML进展的独立危险因素(P < 0.05)。建模集ROC曲线下面积(area under the curve,AUC)为0.912(95%CI:0.875~0.949),灵敏度为0.840,特异度为0.884,约登指数为0.724;验证集AUC为0.885(95%CI:0.821~0.949),灵敏度为0.818,特异度为0.869,约登指数为0.687。Hosmer-Lemeshow检验显示建模集(χ2 = 8.762,P = 0.363)与验证集(χ2 = 9.541,P = 0.308)模型校准度均良好;DCA曲线提示模型在临床决策阈值范围内具有高净获益。 结论 基于多模态MRI联合血清sRAGE构建的预测模型能有效识别WML进展高危患者,具有优异的预测性能和临床适用性,可为临床个体化管理和早期干预决策提供直观的量化参考依据。 Abstract:Objective To construct and validate a predictive model for the progression risk of vascular white matter lesions (WML) based on multimodal MRI combined with serum soluble receptor for advanced glycation end products (sRAGE). Methods A retrospective cohort study design was employed. A total of 330 patients diagnosed with WML diagnosed at the First Hospital of Handan from January 2020 to October 2023 were enrolled and randomly divided into a modeling set (n = 231) and a validation set (n = 99) in a 7∶3 ratio. Eighteen candidate predictive factors were collected, including general data, underlying diseases, laboratory indicators (including serum sRAGE), and multimodal MRI parameters (Fazekas score, FLAIR lesion volume, etc.). The outcome event was defined as lesion progression within 2 years of follow-up. Core factors were selected using LASSO regression, and a nomogram prediction model was constructed using multivariable logistic regression. The discriminative ability, calibration, and clinical applicability of the model were assessed using receiver operating characteristic (ROC) curves, Hosmer-Lemeshow test, and decision curve analysis (DCA), respectively. Results The lesion progression rates in the modeling and validation sets were 21.6% (50/231) and 22.2% (22/99), respectively. LASSO regression identified five core predictors: age ≥70 years, diabetes mellitus, decreased serum sRAGE level, Fazekas score ≥3, and FLAIR lesion volume ≥5 mL. Multivariate logistic regression showed that age ≥70 years, diabetes mellitus, Fazekas score ≥3, decreased serum sRAGE level, and FLAIR lesion volume ≥5 mL were independent risk factors for WML progression (all P < 0.05). In the modeling set, the area under the ROC curve (AUC) was 0.912 (95%CI: 0.875~0.949), with sensitivity of 0.840, specificity of 0.884, and Youden index of 0.724. In the validation set, the AUC was 0.885 (95%CI: 0.821~0.949), with sensitivity of 0.818, specificity of 0.869, and Youden index of 0.687. The Hosmer-Lemeshow test showed good calibration in both modeling set (χ2 = 8.762, P = 0.363) and validation set (χ2 = 9.541, P = 0.308). The DCA curve demonstrated that the model provides high net benefit within the clinical decision threshold range. Conclusion The prediction model constructed based on multimodal MRI combined with serum sRAGE can effectively identify patients at high risk for WML progression, exhibiting excellent predictive performance and clinical applicability. It provides an intuitive quantitative reference for individualized clinical management and early intervention decision-making. -
Key words:
- Vascular white matter lesion /
- MRI /
- sRAGE /
- Influencing factors /
- Nomogram
-
表 1 建模集与验证集基线资料比较[M(P25,P75),n(%)]
Table 1. Comparison of baseline characteristics between modeling set and validation set[M(P25,P75),n(%)]
项目 建模集(n = 231) 验证集(n = 99) χ2/Z P 性别 0.012 0.913 男 127(55.0) 54(54.5) 女 104(45.0) 45(45.5) 年龄(岁) 68.0(62.0,75.0) 67.0(61.0,74.0) −0.582 0.560 BMI(kg/m2) 25.3(23.1,27.8) 25.1(22.9,27.6) −0.321 0.748 吸烟史 0.098 0.754 有 89(38.5) 38(38.4) 无 142(61.5) 61(61.6) 饮酒史 0.156 0.693 有 76(32.9) 31(31.3) 无 155(67.1) 68(68.7) 教育程度 0.524 0.770 小学及以下 65(28.1) 28(28.3) 中学 112(48.5) 47(47.5) 大学及以上 54(23.4) 24(24.2) 多病共存 0.043 0.835 是 107(46.3) 45(45.5) 否 124(53.7) 54(54.5) 高血压 0.112 0.737 有 156(67.5) 66(66.7) 无 75(32.5) 33(33.3) 糖尿病 0.078 0.780 有 92(39.8) 40(40.4) 无 139(60.2) 59(59.6) 高血脂 0.215 0.643 有 109(47.2) 43(43.4) 无 122(52.8) 56(56.6) 冠心病 0.035 0.851 有 68(29.4) 29(29.3) 无 163(70.6) 70(70.7) 血清sRAGE(ng/mL) 1.82(1.35,2.41) 1.78(1.32,2.38) −0.413 0.680 空腹血糖(mmol/L) 5.8(5.2,6.7) 5.7(5.1,6.6) −0.532 0.595 糖化血红蛋白(%) 5.9(5.6,6.4) 5.8(5.5,6.3) −0.396 0.692 同型半胱氨酸(μmol/L) 14.2(11.5,17.8) 13.9(11.2,17.5) −0.367 0.713 C反应蛋白(mg/L) 3.1(1.2,5.8) 2.9(1.1,5.6) −0.284 0.776 MMSE评分(分) 26.0(23.0,28.0) 25.5(22.5,27.5) −0.618 0.537 MoCA评分(分) 22.0(19.0,25.0) 21.5(18.5,24.5) −0.489 0.625 Barthel指数(分) 90.0(80.0,95.0) 88.0(78.0,95.0) −0.721 0.471 Fazekas评分(分) 3.0(2.0,4.0) 3.0(2.0,4.0) −0.294 0.769 Fazekas评分(分) 0.187 0.910 0~1 35(15.2) 15(15.2) 2~3 118(51.1) 51(51.5) 4~6 78(33.8) 33(33.3) ADC值(×10−3 mm2/s) 1.18(1.05,1.32) 1.17(1.04,1.31) −0.216 0.829 FLAIR病变体积(mL) 4.2(2.1,7.8) 4.0(2.0,7.5) −0.385 0.700 SWI微出血灶数量(个) 0.254 0.881 0 146(63.2) 62(62.6) 1~3 65(28.1) 28(28.3) ≥4 20(8.7) 9(9.1) CBF值[mL/(100 g·min)] 32.5(28.6,36.8) 32.2(28.3,36.5) −0.452 0.651 rCBF 0.82(0.71,0.93) 0.81(0.70,0.92) −0.317 0.751 表 2 建模集患者WML进展影响因素的单因素分析[M(P25,P75),n(%)]
Table 2. Univariate analysis of factors influencing WML progression in the modeling set[M(P25,P75),n(%)]
项目 非进展组(n = 181) 进展组(n = 50) χ2/Z P 性别 1.025 0.311 男 92(50.8) 35(70.0) 女 89(49.2) 15(30.0) 年龄(岁) 66.0(61.0,73.0) 75.0(68.0,80.0) −5.872 < 0.001* 32.641 < 0.001* 50~59 40(22.1) 2(4.0) 60~69 86(47.5) 9(18.0) ≥70 55(30.4) 39(78.0) BMI(kg/m2) 25.4(23.2,27.9) 24.9(22.8,27.5) −0.763 0.445 多病共存 18.326 < 0.001* 是 72(39.8) 35(70.0) 否 109(60.2) 15(30.0) 高血压 0.957 0.328 有 118(65.2) 38(76.0) 无 63(34.8) 12(24.0) 糖尿病 12.458 < 0.001* 有 62(34.3) 30(60.0) 无 119(65.7) 20(40.0) 高血脂 1.873 0.171 有 82(45.3) 27(54.0) 无 99(54.7) 23(46.0) 血清sRAGE(ng/mL) 2.05(1.52,2.58) 1.28(0.95,1.63) −6.941 < 0.001* 空腹血糖(mmol/L) 5.7(5.2,6.5) 6.1(5.5,7.3) −1.825 0.068 糖化血红蛋白(%) 5.8(5.5,6.3) 6.1(5.7,6.7) −1.934 0.053 同型半胱氨酸(μmol/L) 13.5(11.2,16.8) 17.6(14.5,21.3) −4.726 < 0.001* C反应蛋白(mg/L) 3.0(1.1,5.6) 3.5(1.5,6.2) −1.284 0.200 MMSE评分(分) 26.5(23.5,28.0) 24.0(21.0,26.5) −2.987 0.003* MoCA评分(分) 22.5(19.5,25.0) 20.0(17.0,22.5) −3.864 < 0.001* Fazekas评分(分) 2.5(2.0,3.0) 4.0(3.0,5.0) −7.215 < 0.001* Fazekas评分(分) 48.932 < 0.001* 0~1 32(17.7) 3(6.0) 2~3 101(55.8) 17(34.0) 4~6 48(26.5) 30(60.0) ADC值(×10−³ mm2/s) 1.15(1.03,1.28) 1.26(1.12,1.41) −3.658 < 0.001* FLAIR病变体积(mL) 3.5(1.8,6.2) 8.7(5.3,12.4) −6.329 < 0.001* FLAIR病变体积(mL) 36.782 < 0.001* < 3 78(43.1) 5(10.0) 3~10 82(45.3) 21(42.0) > 10 21(11.6) 24(48.0) SWI微出血灶数量(个) 14.863 < 0.001* 0 125(69.1) 21(42.0) 1~3 45(24.9) 20(40.0) ≥4 11(6.1) 9(18.0) CBF值(mL/(100 g·min) 33.8(29.5,37.6) 29.2(25.4,32.8) −4.983 < 0.001* *P < 0.05。 表 3 建模集多因素Logistic回归分析结果
Table 3. Multivariate Logistic regression analysis of modeling set
项目 B SE Wald χ2 P OR 95%CI 常量 −5.236 0.518 102.537 < 0.001* 0.005 − 年龄≥70岁 1.452 0.386 14.329 < 0.001* 4.263 2.158~8.425 糖尿病 1.055 0.368 8.367 0.002* 2.871 1.453~5.674 血清sRAGE降低 1.044 0.276 14.283 < 0.001* 1.352 1.198~1.625 Fazekas评分≥3分 1.366 0.372 13.451 < 0.001* 3.917 1.986~7.725 FLAIR病变体积≥5 mL 1.148 0.364 10.023 0.001* 3.154 1.608~6.185 *P < 0.05。 -
[1] 中华医学会神经病学分会, 中华医学会神经病学分会脑血管病学组, 黄一宁, 等. 中国脑小血管病诊治指南2020[J]. 中华神经科杂志, 2022, 55(8): 807-818. [2] 吕小涵, 成斯琪, 陈杏雨, 等. 脑白质病变与帕金森病[J]. 中华行为医学与脑科学杂志, 2022, 31(4): 320-325. doi: 10.13406/j.cnki.cyxb.003502 [3] 洪文娟, 黄洁, 王从过, 等. 老年脑白质病变与阿尔茨海默病痴呆的关联研究[J]. 中华解剖与临床杂志, 2021, 26(2): 197-203. doi: 10.3760/cma.j.cn101202-20201030-00375 [4] 奚继明, 左莉, 冯丽, 等. 老年脑白质病变患者120例的认知功能障碍状况及其相关性研究[J]. 解放军预防医学杂志, 2019, 37(12): 79-80. doi: 10.13704/j.cnki.jyyx.2019.12.084 [5] Tayyib N A, Ramaiah P, Alshahrani S H, et al. Soluble receptor for advanced glycation end products (sRAGE) is associated with obesity rates: A systematic review and meta-analysis of cross-sectional study[J]. BMC Endocr Disord, 2023, 23(1): 275. doi: 10.1186/s12902-023-01520-1 [6] Eke-Gungor H, Sagiroglu B, Can-Coskun Z N, et al. Serum sRAGE levels in children with atopic dermatitis: A prospective study[J]. Pdia, 2023, 40(6): 766-771. doi: 10.5114/ada.2023.133585 [7] van Smeden M, Moons K G, de Groot J A, et al. Sample size for binary logistic prediction models: Beyond events per variable criteria[J]. Stat Methods Med Res, 2019, 28(8): 2455-2474. doi: 10.1177/0962280218784726 [8] 李妍. 甲状腺激素水平及血管性危险因素与脑白质病变关系的研究[D]. 大连: 大连医科大学, 2023. [9] 中国研究型医院学会脑小血管病专业委员会《中国脑小血管病诊治专家共识》编写组. 中国脑小血管病诊治专家共识2021[J]. 中国卒中杂志, 2021, 16(7): 716-726. doi: 10.3969/j.issn.1673-5765.2021.07.013 [10] 中国高血压防治指南修订委员会, 高血压联盟(中国, 中华医学会心血管病学分会中国医师协会高血压专业委员会, 等. 中国高血压防治指南(2018年修订版)[J]. 中国心血管杂志, 2019, 24(1): 24-56. [11] 中华医学会糖尿病学分会. 中国2型糖尿病防治指南(2017年版)[J]. 中华糖尿病杂志, 2018, 10(1): 4-67. [12] Folstein M F, Folstein S E, McHugh P R. "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician[J]. J Psychiatr Res, 1975, 12(3): 189-198. [13] Nasreddine Z S, Phillips N A, Bédirian V, et al. The Montreal cognitive assessment, MoCA: A brief screening tool for mild cognitive impairment[J]. J Am Geriatr Soc, 2005, 53(4): 695-699. doi: 10.1111/j.1532-5415.2005.53221.x [14] Mahoney F I, Barthel D W. Functional evaluation: The barthel index[J]. Md State Med J, 1965, 14: 61-65. [15] Rankin J. Cerebral vascular accidents in patients over the age of 60: II. prognosis[J]. Scott Med J, 1957, 2(5): 200-215. doi: 10.1177/003693305700200504 [16] Farrell B, Godwin J, Richards S, et al. The United Kingdom transient ischaemic attack (UK-TIA) aspirin trial: Final results[J]. J Neurol Neurosurg Psychiatry, 1991, 54(12): 1044-1054. [17] van Swieten J C, Koudstaal P J, Visser M C, et al. Interobserver agreement for the assessment of handicap in stroke patients[J]. Stroke, 1988, 19(5): 604-607. doi: 10.1161/01.str.19.11.1448 [18] 罗倩, 王秀红, 袁平乔, 等. 基于机器学习的4种老年脑白质病变患者衰弱风险预测模型的构建与比较[J]. 护士进修杂志, 2025, 40(13): 1383-1388. doi: 10.16821/j.cnki.hsjx.2025.13.007 [19] 金莹, 毛莉, 韦永珠. 老年糖尿病患者缺血性脑白质病变的危险因素分析[J]. 中华老年心脑血管病杂志, 2025, 27(5): 659-662. doi: 10.3969/j.issn.1009-0126.2025.05.024 [20] 程健, 闵汇刚, 杨娟, 等. 老年2型糖尿病患者脑白质病变严重程度相关因素分析[J]. 西部医学, 2024, 36(10): 1460-1464. doi: 10.3969/j.issn.1672-3511.2024.10.009 [21] 张慧, 刘春艳, 韦丹. S100A6经RAGE介导影响肥胖儿童血管内皮损伤的机制研究[J]. 中国儿童保健杂志, 2020, 28(2): 135-138. [22] 杨升华, 韩雪洁, 胡宏宇, 等. 可溶性晚期糖基化终末产物受体与心脑动脉粥样硬化的相关性[J]. 中华医学杂志, 2019, 99(39): 3062-3067. [23] 胡慧宇. HMGB1和sRAGE在动脉瘤性蛛网膜下腔出血患者脑脊液中的表达及其临床意义[D]. 太原: 山西医科大学, 2023. [24] Shu Z Y, Shao Y, Xu Y Y, et al. Radiomics nomogram based on MRI for predicting white matter hyperintensity progression in elderly adults[J]. J Magn Reson Imag, 2020, 51(2): 535-546. doi: 10.1002/jmri.26813 [25] 钦佩, 马建华. 脑白质变性患者认知障碍的相关因素及列线图模型构建[J]. 中国实用神经疾病杂志, 2025, 28(9): 1099-1104. -
下载: