基于证素辨证原理的2型糖尿病视网膜病变预测模型构建

doi:10.12122/j.issn.1673-4254.2026.10.15

摘要/Abstract

摘要：

目的通过引入证素辨证原理，结合机器学习技术，从而构建2型糖尿病视网膜病变（T2DR）的预测模型，为其发生的综合性、精准化预测提供思路和方法。方法基于横断面调查资料，按照病例报告表和四诊信息表，以T2DR、2型糖尿病（T2DM）患者为研究对象，采集其状态参数并区分差异性，以此形成具有中医特色的T2DR发生潜在预测因子集合；再使用LASSO回归分析从中筛选出对T2DR发生最具预测价值的状态特征；最后使用逻辑回归（LR）、人工神经网络（ANN）、极致梯度提升（XGBoost）、随机森林（RF）、支持向量机（SVM）、K近邻（KNN）6种机器学习技术，构建能够综合反映T2DR发生的预测模型，并进行模型内部验证和评估各预测特征的重要性。结果本研究共纳入T2DM组376例，T2DR组273例，通过对两组筛选出有统计学意义的状态因素进行LASSO分析，发现肝证素、肺证素、阳亢证素、血虚证素、热证素、2型糖尿病性肾病（T2DKD）、脾证素、阴虚证素、血瘀证素、2型糖尿病性周围神经病（T2DPN）、气滞证素、肾证素、2型糖尿病性周围血管病（T2DPVD）、气虚证素、筋骨证素、津亏证素、心率、腰臀比（WHR）、饮食量情况、痰证素、血尿素氮（BUN）、尿素（Urea）、T2DM病程、血红蛋白（HGB）、丙氨酸氨基转移酶（ALT）、天冬氨酸氨基转移酶（AST）、臀围对T2DR发生最具预测价值。通过LR、ANN、XGBoost、RF、SVM、KNN六种ML技术构建T2DR发生的预测建模。区分度评估：采用准确率、精确率、召回率、特异度、假正例率、F1分数、ROC曲线、AUC（95% CI）、Delong检验和500次Bootstrap方法，发现LR、KNN模型具有良好的区分度。校准度评估：采用GiViTI校准带、Platt scaling方法，发现LR模型的概率输出在全局中具有高度可靠性和校准稳健性；采用十组校准曲线、Brier score、等渗回归方法，发现KNN模型的概率预测属于良好水平。临床实用性评估：采用DCA评估LR、KNN模型在训练及验证人群中的临床效用，发现LR与KNN模型在广泛的阈值概率范围内均可提供正的临床净效益，且LR模型在整个范围内的净效益曲线均位于KNN模型之上，尤其在验证集中表现明显，表明其在指导个体化决策方面具有更高的潜在价值。因此，本研究将LR模型作为最优预测模型。结论融入证素因素利用机器学习技术构建的预测模型，可以有效发挥中医药预测疾病的治未病优势，从而实现对T2DM人群易发T2DR的综合性预测，辅助临床决策。

关键词: 中医, 证素, 2型糖尿病视网膜病变, 机器学习, 预测模型

Abstract:

Objective To construct a prediction model for occurrence of type 2 diabetic retinopathy (T2DR) by combining the principle of differentiation of syndrome elements with machine learning (ML). Methods Cross-sectional data were collected from 273 patients with T2DR and 376 T2DM patients for comparison of routine test results and parameters of syndrome elements in traditional Chinese medicine (TCM) identify the candidate TCM indicators for predicting T2DR occurrence. LASSO regression analysis was used to select the syndrome features with the highest predictive value. Six ML algorithms (LR, ANN, XGBoost, RF, SVM and KNN) were used to construct the prediction models, followed by internal validation of the model and evaluation of the importance of each predictive feature. Results LASSO analysis identified the liver, lung, and yang excess syndrome elements, among others, as the most predictive features for T2DR occurrence. Among the 6 ML models, LR and KNN models showed good discrimination power. The LR model yielded highly reliable and robust probability output in global calibration, and the KNN model had good performance for probability predictions. Both the LR and KNN models provided positive net clinical benefits over a wide range of threshold probabilities, but the net benefit curve of the LR model was above that of the KNN model in the whole range, especially in the validation set. Conclusion The ML-based prediction models (especially the ML model) combining the features of syndrome elements effectively use the advantage of TCM for predicting disease and allow comprehensive prediction of T2DR in T2DM patients to assist clinical decision-making for early interventions.

Key words: Traditional Chinese medicine, syndrome elements, type 2 diabetic retinopathy, machine learning, prediction model

乔世杰, 游进顺, 王昶, 熊红萍, 李书楠. 基于证素辨证原理的2型糖尿病视网膜病变预测模型构建[J]. 南方医科大学学报, 2026, 46(10): 2456-2470.

Shijie QIAO, Jinshun YOU, Chang WANG, Hongping XIONG, Shunan LI. Construction of predictive models for type 2 diabetic retinopathy based on the principle of differentiation of syndrome elements[J]. Journal of Southern Medical University, 2026, 46(10): 2456-2470.

图/表 14

参考文献 36

[1]	中华医学会眼科学分会眼底病学组, 中国医师协会眼科医师分会眼底病学组, 许迅, 等. 我国糖尿病视网膜病变临床诊疗指南(2022年): 基于循证医学修订[J]. 中华眼底病杂志, 2023, 39(2): 99-124.
[2]	Tan GS, Cheung N, Simó R, et al. Diabetic macular oedema[J]. Lancet Diabetes Endocrinol, 2017, 5(2): 143-55. doi：10.1016/s2213-8587(16)30052-3
[3]	Solomon SD, Chew E, Duh EJ, et al. Diabetic retinopathy: a position statement by the American diabetes association[J]. Diabetes Care, 2017, 40(3): 412-8. doi：10.2337/dc16-2641
[4]	刘梦伟, 鲍新颖, 王烁, 等. 中医药治疗糖尿病视网膜病变的研究进展[J].中医临床研究, 2022, 14(36): 45-47. doi：10.3969/j.issn.1674-7860.2022.36.012
[5]	李灿东, 纪立金, 鲁玉辉, 等. 论中医健康认知理论的逻辑起点[J]. 中华中医药杂志, 2011, 26(1): 109-11.
[6]	中国医师协会中西医结合医师分会内分泌和代谢病学专业委员会. 糖尿病视网膜病变病证结合诊疗指南(2021-09-24)[J]. 世界中医药, 2021, 16(22): 3270-7.
[7]	褚海琳, 任永霞. 糖尿病视网膜病变患者就诊延迟情况及就诊时间的影响因素分析[J]. 中国全科医学, 2022, 25(12): 1470-4. doi：10.12114/j.issn.1007-9572.2021.02.078
[8]	Saeidian J, Mahmoudi T, Riazi-Esfahani H, et al. Automated assessment of the smoothness of retinal layers in optical coherence tomography images using a machine learning algorithm[J]. BMC Med Imaging, 2023, 23(1): 21. doi：10.1186/s12880-023-00976-w
[9]	董旭辉, 陈德凤, 李蓓, 等. 2型糖尿病患者发生视网膜病变风险预测模型的系统评价[J]. 中华现代护理杂志, 2024, 30(5): 644-50.
[10]	武巾媛, 安洪庆. 糖尿病并发视网膜病变的关键影响因素分析和风险预测研究[J]. 现代预防医学, 2024, 51(3): 557-63.
[11]	刘曦阳, 张伟, 赵若冰, 等. 2型糖尿病视网膜病变风险预测模型的系统评价[J]. 中华现代护理杂志, 2023, 29(32): 4429-36.
[12]	van der Heijden AA, Nijpels G, Badloe F, et al. Prediction models for development of retinopathy in people with type 2 diabetes: systematic review and external validation in a Dutch primary care setting[J]. Diabetologia, 2020, 63(6): 1110-9. doi：10.1007/s00125-020-05134-3
[13]	Usman TM, Saheed YK, Nsang A, et al. A systematic literature review of machine learning based risk prediction models for diabetic retinopathy progression[J]. Artif Intell Med, 2023, 143: 102617. doi：10.1016/j.artmed.2023.102617
[14]	朱文锋. 证素辨证学[M]. 北京: 人民卫生出版社, 2008.
[15]	李灿东, 甘慧娟, 鲁玉辉, 等. 基于证素辨证原理的健康状态辨识研究[J]. 中华中医药杂志, 2011, 26(4): 754-7.
[16]	张怡, 杨波, 李茂, 等. 2018—2022年成都地区体检人群2型糖尿病流行病学特征及影响因素分析[J]. 现代预防医学, 2024, 51(17): 3157-62, 3179. doi：10.20043/j.cnki.MPM.202401474
[17]	陈玉冰. 2型糖尿病视网膜病与肾病相关关系及危险因素研究[D]. 石河子: 石河子大学, 2022.
[18]	中华医学会糖尿病学分会. 中国2型糖尿病防治指南(基层版)[J]. 中华全科医师杂志, 2013, 12(8): 675-96. doi：10.3760/cma.j.issn.1671-7368.2013.08.038
[19]	International Diabetes Federation.IDF Diabetes Atlas,11th edition[EB/OL].(2025-04-07)[2026-01-25]..
[20]	佚名编著. 黄帝内经·素问[M]. 北京: 中国医药科技出版社, 2023: 23, 34.
[21]	赵芳. 中重度非增殖期糖尿病视网膜病变中医证素与眼底改变及miRNA的研究[D]. 北京: 中国中医科学院, 2022.
[22]	周莹, 刘军彤, 杨宇峰, 等. 从“因虚致瘀”理论探讨糖尿病视网膜病变病机[J]. 中国中医基础医学杂志, 2022, 28(8): 1231-3, 1329.
[23]	张玉娴, 王丽英, 苏伟娟, 等. 惜视饮治疗血虚络阻型糖尿病视网膜病变的临床研究[J]. 中医药通报, 2020, 19(5): 47-9, 43.
[24]	田代华, 刘更生整理. 灵枢经[M]. 北京: 人民卫生出版社, 2023: 59, 79, 91.
[25]	高文柱,校注.《诸病源候论》校注[M].北京:学苑出版社,2018: 290.
[26]	熊萌, 曾梅艳, 宋厚盼, 等. 基于现代文献分析的糖尿病视网膜病变中医证型及证素分布探讨[J]. 广州中医药大学学报, 2023, 40(3): 761-8.
[27]	曾筱婷, 滕月, 俞晓艺, 等. 真实世界中不同中医证型增殖型糖尿病视网膜病变患者行抗血管内皮生长因子联合玻璃体切除术治疗的疗效分析[J]. 广州中医药大学学报, 2024, 41(4): 906-16.
[28]	段俊国，等.国际中医临床实践指南糖尿病视网膜病变[J]. 世界中医药, 2025, 20(14): 2427-31.
[29]	段俊国, 毕宏生. 中西医结合眼科学[M]. 4版. 北京: 中国中医药出版社, 2021: 203-4.
[30]	李书楠. 基于证素原理的中医眼底望诊方法建立及2型糖尿病视网膜病变的临床验证研究[D]. 长沙: 湖南中医药大学, 2022.
[31]	翟丽萍, 王旭, 王谦, 等. 清化瘀热方通过IL-33/ST2通路缓解糖尿病视网膜病变大鼠中视网膜细胞炎症反应[J]. 南昌大学学报: 医学版, 2024, 64(6): 33-41, 封3.
[32]	仝小林. 糖尿病中医药临床循证实践指南: 2016版[M]. 北京: 科学出版社, 2016: 141.
[33]	吴红云, 袁振轩, 陈芳, 等. 2型糖尿病视网膜病变患者中医证型分布及其与血清炎症因子及脂联素水平的相关性研究[J]. 广州中医药大学学报, 2024, 41(6): 1371-9.
[34]	谢学军, 王明芳. 糖尿病视网膜病变的病机探讨[J]. 中国中医眼科杂志, 1995, 5(2): 86-8.
[35]	卫晓琴. 基于“目诊仪”对糖尿病视网膜病变中医证候分布规律研究及相关因素分析[D]. 成都: 成都中医药大学, 2023.
[36]	张莹. 芪贞降糖颗粒治疗糖尿病性视网膜病变非增殖期肝肾阴虚证临床研究[J]. 中医学报, 2017, 32(12): 2354-8.

No.	Syndrome elements		T2DM(n=376)	T2DR(n=273)	χ²/Z/t	P
1	Yin deficiency	Severe	233	269	120.711	<0.01
		Non-severe	143	4
2	Phlegm	Severe	147	193	63.32	<0.01
		Non-severe	229	80
3	Liver	Severe	177	266	185.149	<0.01
		Non-severe	199	7
4	Heat	Severe	111	179	83.144	<0.01
		Non-severe	265	94
5	Kidney	Severe	130	195	85.929	<0.01
		Non-severe	246	78
6	Dampness	Severe	156	175	32.364	<0.01
		Non-severe	220	98
7	Yang deficiency	Severe	112	153	45.134	<0.01
		Non-severe	264	120
8	Blood deficiency	Severe	93	216	187.562	<0.01
		Non-severe	283	57
9	Blood stasis	Severe	52	101	47.133	<0.01
		Non-severe	324	172
10	Qi deficiency	Severe	45	128	98.639	<0.01
		Non-severe	331	145
11	Qi stagnation	Severe	42	90	46.379	<0.01
		Non-severe	334	183
12	Yang excess	Severe	4	65	86.122	<0.01
		Non-severe	372	208
13	Heart spirit	Severe	16	38	19.365	<0.01
		Non-severe	360	235
14	Spleen	Severe	5	28	26.144	<0.01
		Non-severe	371	245
15	Heart	Severe	1	10	10.954	<0.01
		Non-severe	375	263
16	Essence deficiency	Severe	0	6	8.341	<0.01
		Non-severe	376	267
17	Tendon-bone	Severe	11	43	34.106	<0.01
		Non-severe	365	230
18	Gallbladder	Severe	0	0	-	-
		Non-severe	376	273
19	Cold	Severe	2	10	8.544	<0.01
		Non-severe	374	263
20	Jin deficiency	Severe	0	3	4.151	<0.05
		Non-severe	376	270
21	Meridian	Severe	1	4	2.975	>0.05
		Non-severe	375	269
22	Stomach	Severe	2	5	2.504	>0.05
		Non-severe	374	268
23	Lung	Severe	2	33	41.397	<0.01
		Non-severe	374	240
24	Dryness	Severe	0	1	1.379	>0.05
		Non-severe	376	272

No.	Syndrome elements		T2DM(n=376)	T2DR(n=273)	χ²/Z/t	P
1	Yin deficiency	Severe	233	269	120.711	<0.01
		Non-severe	143	4
2	Phlegm	Severe	147	193	63.32	<0.01
		Non-severe	229	80
3	Liver	Severe	177	266	185.149	<0.01
		Non-severe	199	7
4	Heat	Severe	111	179	83.144	<0.01
		Non-severe	265	94
5	Kidney	Severe	130	195	85.929	<0.01
		Non-severe	246	78
6	Dampness	Severe	156	175	32.364	<0.01
		Non-severe	220	98
7	Yang deficiency	Severe	112	153	45.134	<0.01
		Non-severe	264	120
8	Blood deficiency	Severe	93	216	187.562	<0.01
		Non-severe	283	57
9	Blood stasis	Severe	52	101	47.133	<0.01
		Non-severe	324	172
10	Qi deficiency	Severe	45	128	98.639	<0.01
		Non-severe	331	145
11	Qi stagnation	Severe	42	90	46.379	<0.01
		Non-severe	334	183
12	Yang excess	Severe	4	65	86.122	<0.01
		Non-severe	372	208
13	Heart spirit	Severe	16	38	19.365	<0.01
		Non-severe	360	235
14	Spleen	Severe	5	28	26.144	<0.01
		Non-severe	371	245
15	Heart	Severe	1	10	10.954	<0.01
		Non-severe	375	263
16	Essence deficiency	Severe	0	6	8.341	<0.01
		Non-severe	376	267
17	Tendon-bone	Severe	11	43	34.106	<0.01
		Non-severe	365	230
18	Gallbladder	Severe	0	0	-	-
		Non-severe	376	273
19	Cold	Severe	2	10	8.544	<0.01
		Non-severe	374	263
20	Jin deficiency	Severe	0	3	4.151	<0.05
		Non-severe	376	270
21	Meridian	Severe	1	4	2.975	>0.05
		Non-severe	375	269
22	Stomach	Severe	2	5	2.504	>0.05
		Non-severe	374	268
23	Lung	Severe	2	33	41.397	<0.01
		Non-severe	374	240
24	Dryness	Severe	0	1	1.379	>0.05
		Non-severe	376	272

No.	Indicators		T2DM (n=376)	T2DR (n=273)	χ²/Z/t	P
1	Gender	Male	216	137	3.364	>0.05
		Female	160	136
2	Age (years)		60.54±13.02	61.33±12.41	-0.784	>0.05
3	Residence	Urban	230	166	0.009	>0.05
		Rural	146	107
4	Occupation	Employed	166	107	1.593	>0.05
		Unemployed	210	166
5	Marital status	Married	313	222	0.405	>0.05
		Unmarried	63	51
6	Education level	Illiterate	24	38	-3.682	<0.01
		Primary school	119	95
		Junior high school	107	78
		Senior high school/technical secondary school/vocational high school	83	44
		College/associate degree	43	17
		Master's degree or above	0	1
7	Household income (CNY/month)	≤1000	28	24	-1.569	>0.05
		1001-2000	35	38
		2001-5000	113	79
		≥5001	200	132
8	Living arrangement	Living with others	330	244	0.402	>0.05
		Living alone	46	29
9	Religious belief	None	224	155	0.51	>0.05
		Yes	152	118
10	History of insulin therapy	None	288	156	27.697	<0.01
		Yes	88	117
11	Smoking history	None	280	211	0.684	>0.05
		Yes	96	62
12	Smoking quantity	Non-smoker	280	211	-0.746	>0.05
		<1 pack/day	48	27
		1-2 packs/day	36	29
		>2 packs/day	12	6
13	Drinking history	None	287	221	1.987	>0.05
		Yes	89	52
14	Diet:time	Irregular	147	122	2.039	>0.05
		Regular	229	151
15	Diet:dietary intake	Overeating	224	133	7.532	<0.01
		Controlled	152	140
16	Diet:category-Staple food	Rice-based	297	223	1.129	>0.05
		Wheat-based	17	9
		Mixed	60	39
		Other	2	2
17	Diet:category-non-staple food	Vegetable-based	220	173	4.599	>0.05
		White meat-based	109	65
		Red meat-based	3	6
		Balanced	44	29
18	Diet:Snacking status	None	167	126	0.193	>0.05
		Yes	209	147
19	Dietary preference	None	43	24	1.195	>0.05
		Yes	333	249
20	Physical labor	Light	290	201	-0.96	>0.05
		Moderate	77	67
		Heavy	9	5
21	Exercise habits and frequency	None	117	82	-0.266	>0.05
		Occasional (1-2 times/week)	16	9
		Sometimes (3-4 times/week)	7	12
		Frequently (5-6 times/week)	8	2
		Daily	228	168
22	Participation in T2DM education programs	None	366	269	1.069	>0.05
		Yes	10	4
23	Attention to health preservation and healthcare	Mostly ignored	300	230	-1.515	>0.05
		Moderate	53	33
		Highly valued	23	10
24	Regular blood glucose monitoring	None	306	198	7.149	<0.01
		Yes	70	75

No.	Indicators		T2DM (n=376)	T2DR (n=273)	χ²/Z/t	P
1	Gender	Male	216	137	3.364	>0.05
		Female	160	136
2	Age (years)		60.54±13.02	61.33±12.41	-0.784	>0.05
3	Residence	Urban	230	166	0.009	>0.05
		Rural	146	107
4	Occupation	Employed	166	107	1.593	>0.05
		Unemployed	210	166
5	Marital status	Married	313	222	0.405	>0.05
		Unmarried	63	51
6	Education level	Illiterate	24	38	-3.682	<0.01
		Primary school	119	95
		Junior high school	107	78
		Senior high school/technical secondary school/vocational high school	83	44
		College/associate degree	43	17
		Master's degree or above	0	1
7	Household income (CNY/month)	≤1000	28	24	-1.569	>0.05
		1001-2000	35	38
		2001-5000	113	79
		≥5001	200	132
8	Living arrangement	Living with others	330	244	0.402	>0.05
		Living alone	46	29
9	Religious belief	None	224	155	0.51	>0.05
		Yes	152	118
10	History of insulin therapy	None	288	156	27.697	<0.01
		Yes	88	117
11	Smoking history	None	280	211	0.684	>0.05
		Yes	96	62
12	Smoking quantity	Non-smoker	280	211	-0.746	>0.05
		<1 pack/day	48	27
		1-2 packs/day	36	29
		>2 packs/day	12	6
13	Drinking history	None	287	221	1.987	>0.05
		Yes	89	52
14	Diet:time	Irregular	147	122	2.039	>0.05
		Regular	229	151
15	Diet:dietary intake	Overeating	224	133	7.532	<0.01
		Controlled	152	140
16	Diet:category-Staple food	Rice-based	297	223	1.129	>0.05
		Wheat-based	17	9
		Mixed	60	39
		Other	2	2
17	Diet:category-non-staple food	Vegetable-based	220	173	4.599	>0.05
		White meat-based	109	65
		Red meat-based	3	6
		Balanced	44	29
18	Diet:Snacking status	None	167	126	0.193	>0.05
		Yes	209	147
19	Dietary preference	None	43	24	1.195	>0.05
		Yes	333	249
20	Physical labor	Light	290	201	-0.96	>0.05
		Moderate	77	67
		Heavy	9	5
21	Exercise habits and frequency	None	117	82	-0.266	>0.05
		Occasional (1-2 times/week)	16	9
		Sometimes (3-4 times/week)	7	12
		Frequently (5-6 times/week)	8	2
		Daily	228	168
22	Participation in T2DM education programs	None	366	269	1.069	>0.05
		Yes	10	4
23	Attention to health preservation and healthcare	Mostly ignored	300	230	-1.515	>0.05
		Moderate	53	33
		Highly valued	23	10
24	Regular blood glucose monitoring	None	306	198	7.149	<0.01
		Yes	70	75

Diseases	Training set	Validation set	χ²	P
T2DR (n=273)	192 (42.11)	81 (41.97)	0.001	>0.05
T2DM (n=376)	264 (57.89)	112 (58.03)	0.001	>0.05