Application of machine learning algorithm in diabetes risk prediction of physical examination population
-
摘要:
目的 探索Logistic回归分析模型和LightGBM(light gradient boosting machine)算法对体检人群未来罹患糖尿病的预测效果及影响因素。 方法 选取2003年8月-2019年4月在南方医院健康管理中心多次进行团体参检的36 292例非糖尿病人员，分层随机选取70%样本，以首次体检的性别、年龄、BMI、腰围、心率、收缩压、舒张压、空腹血糖等34项指标作为自变量，以相对首次体检时间的5年内是否罹患糖尿病为因变量，基于Logistic回归分析模型和LightGBM算法分别建立糖尿病预测模型。将预测模型应用于剩余30%样本，并使用受试者工作特征(receiver operating characteristic, ROC)曲线下面积(area under curve, AUC)进行预测效果的评价。 结果 Logistic回归分析模型和LightGBM算法模型的AUC分别为0.906和0.910，在最佳临界点上，Logistic回归分析模型的灵敏度和特异度分别为81.5%和84.3%，LightGBM(light gradient boosting machine)算法模型的灵敏度和特异度分别为81.6%和85.2%。 结论 Logistic回归分析模型和LightGBM算法模型对体检人群的未来糖尿病患病风险均有较好的预测效果。 -
关键词:
- 糖尿病 /
- 体检 /
- Logistic回归分析模型 /
- LightGBM模型
Abstract:Objective To explore the predictive effect and influencing factors of Logistic regression analysis model and Light GBM algorithm on the development of diabetes in the physical examination population. Methods A total of 36 292 subjects without diabetes were selected from the Health Management Center of Nanfang Hospital from August 2003 to April 2019. We ramdomly selected 70% samples by stratification to construct trainingset. The independent variables were 34 indicators including gender, age, body mass index (BMI), waist circumference, heart rate, systolic blood pressure, diastolic blood pressure, and fasting blood glucose in the first physical examination. We defined the dependent variable as developing diabetes within 5 years from the first physical examination.Logistic regression analysis model and LightGBM (light gradient boosting machine) algorithm was uesd to establish diabetes prediction models, respectively. The prediction model was applied to the remaining 30% samples and the area under the receiver operating characteristic (ROC) curve (AUC) was used to evaluate the prediction effect. Results The AUC of the Logistic regression algorithm model was 0.906, while the AUC of the LightGBM analysis model was 0.910. At the optimal critical point, the sensitivity and specificity of the Logistic regression analysis model were 81.5% and 84.3%, respectively. And the sensitivity and specificity of the LightGBM analysis model were 81.6% and 85.2%, respectively. Conclusion The Logistic regression algorithm model and LightGBM algorithm model have good prediction effect on the development of diabetes in the physical examination population. -
表 1 体检人群的基本情况分析(x±s)
Table 1. Analysis of the basic characteristics of medical examination population (x±s)
变量 5年内未患糖尿病 5年内罹患糖尿病 P值 性别[n(%)] < 0.001 男 20 332(91.72) 1 835(8.28) 女 13 716(97.10) 409(2.90) 年龄(岁) 38.96±11.73 49.68±12.95 < 0.001 BMI(kg/m2) 23.17±3.14 26.02±3.32 < 0.001 腰围(cm) 78.03±9.46 87.26±9.38 < 0.001 心率(次/分) 79.14±11.84 80.26±11.67 < 0.001 收缩压(mm Hg) 119.24±15.58 132.62±17.94 < 0.001 舒张压(mm Hg) 73.62±10.92 81.62±12.77 < 0.001 空腹血糖(mmol/L) 4.70±0.53 5.66±0.80 < 0.001 尿酸(0.001 mmol/L) 353.83±98.58 406.90±95.30 < 0.001 总胆固醇(mmol/L) 5.13±0.98 5.52±1.09 < 0.001 甘油三酯(mmol/L) 1.49±1.34 2.57±2.37 < 0.001 低密度脂蛋白胆固醇(mmol/L) 2.77±0.78 3.06±0.85 < 0.001 高密度脂蛋白胆固醇(mmol/L) 1.58±0.43 1.31±0.35 < 0.001 丙氨酸氨基转移酶(U/L) 22.32±20.68 33.46±26.04 < 0.001 天门冬氨酸氨基转移酶(U/L) 25.4±12.55 30.49±15.40 < 0.001 肌酐(0.001 mmol/L) 72.27±16.69 76.18±16.15 < 0.001 尿素(mmol/L) 4.58±1.06 4.95±1.27 < 0.001 白细胞总数(109/L) 6.48±1.60 7.25±1.75 < 0.001 中性粒细胞总数(109/L) 3.66±1.20 4.11±1.27 < 0.001 红细胞总数(1012/L) 4.90±0.55 5.06±0.55 < 0.001 单核细胞总数(109/L) 0.37±0.13 0.42±0.15 < 0.001 淋巴细胞总数(109/L) 2.24±0.60 2.46±0.69 < 0.001 嗜碱细胞总数(109/L) 0.03±0.02 0.03±0.02 < 0.001 嗜酸细胞总数(109/L) 0.19±0.17 0.23±0.19 < 0.001 血小板总数(109/L) 243.84±53.50 245.08±56.83 0.314 血红蛋白(g/L) 143.52±15.27 149.25±14.13 < 0.001 红细胞平均血红蛋白量(pg) 29.46±2.74 29.70±2.95 < 0.001 红细胞平均血红蛋白浓度(g/L) 333.07±10.61 334.91±10.99 < 0.001 红细胞比容(%) 43.00±40.00 45.00±40.00 < 0.001 红细胞平均容积(fL) 88.36±7.00 88.57±7.48 0.192 大型血小板比值(%) 13.13±1.46 15.14±5.76 < 0.001 平均血小板容积(fL) 10.82±0.83 10.78±0.84 0.034 红细胞分布宽度-CV(%) 13.08±1.12 13.22±1.19 < 0.001 红细胞分布宽度-SD(fL) 41.69±2.83 42.12±3.06 < 0.001 
下载: 导出CSV
表 2 多因素Logistic回归分析模型分析结果
Table 2. Analysis results of multi-factor Logistic regression analysis model
变量 β值 Wald值 OR(95% CI)值 P值 年龄(岁) 0.021 72.441 1.021(1.016~1.026) < 0.001 丙氨酸氨基转移酶(U/L) 0.007 53.143 1.007(1.005~1.009) < 0.001 中性粒细胞总数(109/L) 0.131 35.125 1.140(1.091~1.190) < 0.001 红细胞平均血红蛋白量(pg) 0.184 6.316 1.202(1.040~1.386) 0.012 红细胞比容(%) -7.870 3.474 0.000(0.000~1.611) 0.062 淋巴细胞总数(109/L) 0.204 21.558 1.226(1.125~1.337) < 0.001 红细胞总数(1012/L) 0.805 4.870 2.236(1.088~4.548) 0.027 嗜碱细胞总数(109/L) 2.101 2.238 8.174(0.498~123.787) 0.135 红细胞平均血红蛋白浓度(g/L) -0.014 4.595 0.986(0.973~0.999) 0.032 嗜酸细胞总数(109/L) 0.298 3.948 1.347(0.998~1.798) 0.047 收缩压(mm Hg) 0.008 22.336 1.008(1.005~1.012) < 0.001 大型血小板比值(%) 0.160 265.812 1.173(1.151~1.196) < 0.001 肌酐(0.001 mmol/L) -0.009 23.236 0.991(0.987~0.994) < 0.001 空腹血糖(mmol/L) 1.933 1 927.141 6.913(6.345~7.541) < 0.001 总胆固醇(mmol/L) 0.430 102.369 1.537(1.414~1.670) < 0.001 低密度脂蛋白胆固醇(mmol/L) -0.499 88.919 0.607(0.547~0.673) < 0.001 高密度脂蛋白胆固醇(mmol/L) -0.948 123.818 0.387(0.328~0.457) < 0.001 腰围(cm) 0.047 177.602 1.048(1.041~1.056) < 0.001 尿素(mmol/L) 0.045 3.153 1.046(0.995~1.099) 0.076
下载: 导出CSV
表 3 Logistic回归分析模型预测模型在测试集上分类结果[n(%)]
Table 3. Logistic regression prediction model classification results on the test set [n(%)]
预测结果 实际结果 合计 5年内罹患糖尿病 5年内未患糖尿病 5年内罹患糖尿病 563(81.48) 1 600(15.69) 2 163(19.87) 5年内未患糖尿病 128(18.52) 8 597(84.31) 8 725(80.13) 合计 691(100.00) 10 197(100.00) 10 888(100.00)
下载: 导出CSV
表 4 LightGBM预测模型在测试集上分类结果[n(%)]
Table 4. LightGBM prediction model classification results on the test set [n(%)]
预测结果 实际结果 合计 5年内罹患糖尿病 5年内未患糖尿病 5年内罹患糖尿病 564(81.62) 1 512(14.83) 2 076(19.07) 5年内未患糖尿病 127(18.38) 8 685(85.17) 8 812(80.93) 合计 691(100.00) 10 197(100.00) 10 888(100.00)
下载: 导出CSV
-
[1] Cho NH, Shaw JE, Karuranga S, et al. IDF diabetes atlas: global estimates of diabetes prevalence for 2017 and projections for 2045[J]. Diabetes Res Clin Pract, 2018, 138: 271-281. DOI: 10.1016/j.diabres.2018.02.023. [2] Wei Y, Wang J, Han X, et al. Metabolically healthy obesity increased diabetes incidence in a middle‐aged and elderly Chinese population[J]. Diabetes Metab Res Rev, 2019, 36(1). DOI: 10.1002/dmrr.3202. [3] Peer N, Balakrishna Y, Durao S. Screening for type 2 diabetes mellitus[J]. Cochrane Database Syst Rev, 2020, (5): 1465-1858. DOI: 10.1002/14651858.cd005266. [4] Ekoe J, Goldenberg R, Katz P. Screening for diabetes in adults[J]. Can J Diabetes, 2018, 42: S16-S19. DOI: 10.1016/j.jcjd.2017.10.004. [5] 覃伟, 高敏, 沈莹, 等. 基于机器学习算法的2型糖尿病患者3个月血糖预测[J]. 中华疾病控制杂志, 2019, 23(11): 1313-1317. DOI: 10.16462/j.cnki.zhjbkz.2019.11.003.Qin W, Gao M, Shen Y, et al. Prediction of 3-mouth glycemic control in type 2 diabetes mellitus based on machine learning algorithm[J]. Chin J Dis Control Prev, 2019, 23(11): 1313-1317. DOI: 10.16462/j.cnki.zhjbkz.2019.11.003. [6] Mahboob AT, Iqbal MA, Ali Y, et al. A model for early prediction of diabetes[J]. Inform Med Unlocked, 2019, 16: 100204. DOI: 10.1016/j.imu.2019.100204. [7] Dinh A, Miertschin S, Young A, et al. A data-driven approach to predicting diabetes and cardiovascular disease with machine learning[J]. BMC Med Inform Decis Mak, 2019, 19(1): 211. DOI: s12911-019-0918-5. [8] Mamuda M, Sathasivam S. Predicting the survival of diabetes using neural network[J]. AIP Conf Proc, 2017, 1870(1). DOI: 10.1063/1.4995878. [9] 吴文军, 徐蕾, 黄丽梅, 等. 2016-2017年上海市社区糖尿病高危人群筛查结果分析[J]. 中华疾病控制杂志, 2019, 23(7): 802-806. DOI: 10.16462/j.cnki.zhjbkz.2019.07.012.Wu WJ, Xu L, Huang LM, et al. Analysis of diabetes screening in high-risk population in different communities in Shanghai from 2016 to 2017[J]. Chin J Dis Control Prev, 2019, 23(7): 802-806. DOI: 10.16462/j.cnki.zhjbkz.2019.07.012. [10] Lynam AL, Dennis JM, Owen KR, et al. Logistic regression has similar performance to optimised machine learning algorithms in a clinical setting: application to the discrimination between type 1 and type 2 diabetes in young adults[J]. Diagn Progn Res, 2020, 4(1): 6. DOI: 10.1186/s41512-020-00075-2. [11] Ke G, Meng Q, Finley T, et al. LightGBM: a highly efficient gradient boosting decision tree[J]. Adv Neural Inf Process Syst, 2017: 3146-3154. doi: 10.5555/3294996.3295074 [12] 中华医学会, 中华医学会杂志社, 中华医学会全科医学分会, 等. 2型糖尿病基层诊疗指南(实践版·2019)[J]. 中华全科医师杂志, 2019, 18(9): 810-818. DOI:10.3760/cma.j.issn.1671-7368. 2019.09.003.Chinese Medical Association, Chinese Medical Journals Publishing House, Chinese Society of General Practice, et al. Guideline for primary care of type 2 diabetes: practice version(2019)[J]. Chin J Gen Pract, 2019, 18(9): 810-818. DOI: 10.3760/cma.j.issn.1671-7368.2019.09.003. [13] Shuai, Y, Zheng Y, Huang H, et al. Hybrid Software Obsolescence Evaluation Model Based on PCA-SVM-GridSearchCV[C] // 2018 IEEE 9th International Conference on Software Engineering and Service Science (ICSESS) IEEE. Beijing: IEEE, 2018: 449-453. [14] Mugeni R, Aduwo JY, Briker SM, et al. A Review of diabetes prediction equations in African descent populations[J]. Front Endocrinol (Lausanne), 2019, 10: 663. DOI: 10.3389/fendo.2019.00663. [15] Xie Z, Nikolayeva O, Luo J, et al. Building risk prediction models for type 2 diabetes using machine learning techniques[J]. Prev Chronic Dis, 2019, 16: E130. DOI: 10.5888/pcd16.190109. [16] Chen X, Wu Z, Chen Y, et al. Risk score model of type 2 diabetes prediction for rural Chinese adults: the rural deqing cohort study[J]. J Endocrinol Invest, 2017, 40(10): 1115-1123. DOI: 10.1007/s40618-017-0680-4. [17] Wang K, Gong M, Xie S, et al. Nomogram prediction for the 3-year risk of type 2 diabetes in healthy mainland China residents[J]. EPMA J, 2019, 10(3): 227-237. DOI: 10.1007/s13167-019-00181-2. [18] Yatsuya H, Li Y, Hirakawa Y, et al. A point system for predicting 10-year risk of developing type 2 diabetes mellitus in Japanese men: aichi workers' cohort study[J]. J Epidemiol, 2018, 28(8): 347-352. DOI: 10.2188/jea.je20170048. [19] Zhang Y, Zhou X, Wang Q, et al. Quality of reporting of multivariable logistic regression models in Chinese clinical medical journals[J]. Medicine (Baltimore), 2017, 96(21): e6972. DOI:10.1097/md.000000000 0006972. [20] Maddatu J, Anderson-Baucum E, Evans-Molina C. Smoking and the risk of type 2 diabetes[J]. Transl Res, 2017, 184: 101-107. DOI: 10.1016/j.trsl.2017.02.004. [21] Holst C, Becker U, Jørgensen ME, et al. Alcohol drinking patterns and risk of diabetes: a cohort study of 70, 551 men and women from the general Danish population[J]. Diabetologia, 2017, 60(10): 1941-1950. DOI: 10.1007/s00125-017-4359-3. [22] Sun D, Zhou T, Heianza Y, et al. Type 2 diabetes and hypertension : a study on bidirectional causality[J]. Circ Res, 2019, 124(6): 930-937. DOI: 10.1161/circresaha.118.314487. [23] Schnurr TM, Jakupović H, Carrasquilla GD, et al. Obesity, unfavourable lifestyle and genetic risk of type 2 diabetes: a case-cohort study[J]. Diabetologia, 2020, 63(7): 1324-1332. DOI: 10.1007/s00125-020-05140-5. [24] Kolb H, Martin S. Environmental/lifestyle factors in the pathogenesis and prevention of type 2 diabetes[J]. BMC Med, 2017, 15(1). DOI: 10.1186/s12916-017-0901-x. -
