Bayesian analysis of influencing factors for the misclassification of blood glucose stability
-
摘要:
目的 应用贝叶斯Logistic回归分析法探讨糖尿病患者血糖稳定性存在错误分类时的影响因素,并与传统Logistic回归分析法相比较,为临床工作者提供一种结局指标存在错误分类时的研究方法。 方法 采用分层抽样的方法选取江苏省糖尿病并发症监测点上的患者,以糖化血红蛋白(glycosylated hemoglobin, HbA1c)为“金标准”给出问卷调查血糖稳定性的灵敏度与特异度的专家先验,应用贝叶斯方法和传统方法进行单因素与多因素分析,回归模型的选择以偏差信息准则(deviance information criterion, DIC)或赤池信息准则(Akaike information criterion, AIC)小且参数最少为标准。 结果 调查对象共509人,年龄为(62.32±9.29)岁,贝叶斯多因素Logistic回归分析得出糖尿病病程短(OR=0.299,95% CI: 0.093~0.840)、文化程度高(OR=0.348,95% CI: 0.130~0.914)为血糖稳定性的保护因素,食用腌制食物的频率高(OR=4.310,95% CI: 1.369~22.131)、BMI异常(OR=2.912,95% CI: 1.196~9.865)为血糖稳定性的危险因素。传统Logistic回归分析也得出了同样的结果,但每周食用腌制食物的频率和BMI的OR(95% CI)值被低估,分别为1.671(1.100~2.538)、1.650(1.115~2.440)。 结论 贝叶斯方法和传统Logistic回归分析法均得出病程、文化程度、食用腌制食物的频率以及BMI是影响血糖稳定性的因素,且贝叶斯方法可以校正因变量分类错误,得到的结果更可信。 Abstract:Objective The Bayesian Logistic method was used to explore the influencing factors of the misclassification of blood glucose stability in diabetic patients, and compared with the traditional Logistic method, to provide a research method with misclassification of outcome for the clinicians. Methods The stratified sampling method was used to select patients at the monitoring sites of diabetes complications in Jiangsu Province. The expert before the sensitivity and specificity of blood glucose stability was given using glycosylated hemoglobin as the gold standard. Univariate and multivariate analyses were carried out using the Bayesian and traditional methods. The selection of the regression model was based on the minimum deviance information criterion or Akaike information criterion and the least parameters. Results A total of 509 subjects were investigated, with an age of (62.32±9.29) years old. Bayesian Logistic regression analysis showed that short course of diabetes (OR=0.299, 95% CI: 0.093-0.840) and high education level (OR=0.348, 95% CI: 0.130-0.914) were protective factors for blood glucose stability. High frequency of consumption of preserved food (OR=4.310, 95% CI: 1.369-22.131) and abnormal BMI (OR=2.912, 95% CI: 1.196-9.865) were risk factors for blood glucose stability. Traditional Logistic regression analysis also reached the same result, but the OR (95% CI) values of preserved food consumption and BMI were underestimated, which were 1.671 (1.100-2.538) and 1.650 (1.115-2.440), respectively. Conclusions The Bayesian and the traditional Logistic regression method concluded that the course of the disease, educational level, frequency of consumption of preserved food and BMI were the factors affecting blood glucose stability. The Bayesian method could correct the classification error of dependent variables, and the results obtained were more reliable. -
Key words:
- Bayesian method /
- Blood glucose stability /
- Influencing factors
-
表 1 人口学特征和临床信息
Table 1. Demographic characteristics and clinical information
变量 血糖不稳定[n(%)] 血糖稳定[n(%)] 贝叶斯Logistic法
OR(95% CI)值传统Logistic法
OR(95% CI)值人口学特征 年龄(岁) ≤50 29(8.33) 25(15.53) 0.303(0.082~1.348) 0.405(0.205~0.801) a >50~<61 103(29.60) 49(30.44) 0.957(0.274~6.488) 0.734(0.427~1.264) 61~<71 133(38.22) 58(36.02) 1.208(0.340~8.662) 0.801(0.475~1.352) 71~80 83(23.85) 29(18.01) 1.000 1.000 性别 男 151(43.39) 79(49.07) 1.000 1.000 女 197(56.61) 82(50.93) 2.733(0.759~10.590) 1.257(0.864~1.828) 地区 城市 109(31.32) 66(40.99) 1.000 1.000 农村 239(68.68) 95(59.01) 3.886(1.105~15.040) a 1.523(1.034~2.244) a 婚姻 已婚 309(88.79) 145(90.06) 1.393(0.283~4.560) 0.874(0.473~1.616) 其他 39(11.21) 16(9.94) 1.000 1.000 人口学特征 文化程度 初中及以下 298(85.63) 121(75.16) 1.000 1.000 高中及以上 50(14.37) 40(24.84) 0.404(0.127~0.842) a 0.508(0.318~0.809) a 医保类型 职工医保 57(16.38) 43(26.71) 0.403(0.137~0.850)a 0.537(0.343~0.843) a 其他 291(83.62) 118(73.29) 1.000 1.000 BMI 正常 128(36.78) 81(50.31) 1.000 1.000 不正常 220(63.22) 80(49.69) 5.015(1.440~19.160) a 1.740(1.192~2.540) a 生活方式及饮食习惯 吸烟 从不 241(69.25) 114(70.81) 1.000 1.000 曾经 46(13.22) 21(13.04) 1.489(0.364~11.775) 1.036(0.591~1.818) 现在 61(17.53) 26(16.15) 1.717(0.441~12.833) 1.110(0.666~1.849) 饮酒 从不 244(70.11) 116(72.05) 1.000 1.000 曾经 34(9.77) 15(9.32) 1.609(0.353~14.041) 1.078(0.565~2.057) 现在 70(20.11) 30(18.63) 1.687(0.469~11.834) 1.109(0.685~1.795) 体力活动水平 高 7(2.01) 5(3.11) 1.000 1.000 中 87(25.00) 40(24.84) 2.149(0.423~13.667) 1.554(0.465~5.195) 低 254(72.99) 116(72.05) 2.105(0.445~11.519) 1.564(0.486~5.032) 睡眠质量 一般及以下 90(25.86) 26(16.15) 1.000 1.000 较好 71(20.40) 33(20.50) 0.408(0.136~1.111) 0.530(0.322~0.872) a 好 187(53.74) 102(63.35) 0.733(0.198~5.244) 0.622(0.341~1.133) 每周有几天吃腌制食物(d) E1 214(61.49) 117(72.67) 1.000 1.000 ≥1 134(38.51) 44(27.33) 5.267(1.252~21.690) a 1.665(1.107~2.504) a 每周有几天吃甜品(d) < 1 339(97.41) 150(93.17) 1.000 1.000 ≥1 9(2.59) 11(6.83) 0.437(0.079~1.538) 0.362(0.147~0.892) a 疾病相关因素 病程(年) <6 78(22.41) 55(34.16) 0.338(0.104~0.969) a 0.508(0.294~0.877) a 6~<12 121(34.77) 54(33.54) 0.984(0.285~5.720) 0.802(0.471~1.365) 12~<17 68(19.54) 23(14.29) 2.166(0.464~18.011) 1.058(0.561~1.998) 17~36 81(23.28) 29(18.01) 1.000 1.000 是否有并发症 有 240(68.97) 99(61.49) 3.243(0.917~12.560) 1.392(0.942~2.056) 无 108(31.03) 62(38.51) 1.000 1.000 家族史 无 165(47.41) 76(47.20) 1.000 1.000 有 174(50.00) 78(48.45) 1.224(0.418~7.022) 0.973(0.665~1.425) 不清楚 9(2.59) 7(4.35) 0.709(0.123~7.606) 0.576(0.207~1.603) 用药情况 无 9(2.59) 8(4.97) 1.000 1.000 磺脲类 151(43.39) 82(50.93) 1.348(0.315~5.023) 1.637(0.609~4.404) 胰岛素 41(11.78) 20(12.42) 1.954(0.392~12.379) 1.822(0.611~5.432) 两者均有 147(42.24) 51(31.68) 4.402(0.940~27.440) 2.562(0.939~6.994) 注:a OR值的95% CI不包括1.000,说明该变量有统计学意义。 表 2 血糖稳定性的多因素分析
Table 2. Multivariate analysis of blood glucose stability
变量 贝叶斯Logistic法
OR(95% CI)值传统Logistic法
OR(95% CI)值病程(年) <6 0.299(0.093~0.840) a 0.471(0.268~0.825) a 6~<12 0.642(0.197~2.129) 0.762(0.442~1.314) 12~<17 2.035(0.444~16.151) 0.988(0.515~1.896) 17~36 1.000 1.000 教育水平 初中及以下 1.000 1.000 高中及以上 0.348(0.130~0.914) a 0.538(0.333~0.870) a 每周有几天吃腌制食物(d) < 1 1.000 1.000 ≥1 4.310(1.369~22.131) a 1.671(1.100~2.538) a BMI 正常 1.000 1.000 不正常 2.912(1.196~9.865) a 1.650(1.115~2.440) a 注:a OR值的95% CI不包括1.000,说明该变量有统计学意义。 -
[1] 陈赛萍. 探讨糖尿病患者糖化血红蛋白检验的应用[J]. 糖尿病新世界, 2020, 23(18): 46-47, 50. DOI: 10.16658/j.cnki.1672-4062.2020.18.046.Chen SP. Explore the application of glycosylated hemoglobin test in diabetic patients[J]. Diabetes New World, 2020, 23(18): 46-47, 50. DOI: 10.16658/j.cnki.1672-4062.2020.18.046. [2] 关合华, 郜亚楠. 对糖化血红蛋白与血脂检测在2型糖尿病检测中的临床意义进行探究[J]. 实用糖尿病杂志, 2021, 17(1): 33. https://www.cnki.com.cn/Article/CJFDTOTAL-LNSY202101033.htmGuan HH, Gao YN. Explore the clinical significance of glycosylated hemoglobin and blood lipids in the detection of type 2 diabetes[J]. Journal of Practical Diabetology, 2021, 17(1): 33. https://www.cnki.com.cn/Article/CJFDTOTAL-LNSY202101033.htm [3] 中华医学会糖尿病学分会. 中国2型糖尿病防治指南(2020年版)[J]. 中华内分泌代谢杂志, 2021, 37(4): 311-398. DOI: 10.3760/cma.j.cn311282-20210304-00142.Chinese Diabetes Society. Guideline for the prevention and treatment of type 2 diabetes mellitus in China (2020 edition)[J]. Chin J Endocrinol Metab, 2021, 37(4): 311-398. DOI: 10.3760/cma.j.cn311282-20210304-00142. [4] 王懿倩. 即时检验血糖仪指尖血及静脉全血血糖的检测效果探讨[J]. 心理月刊, 2018(9): 197-198. DOI: 10.19738/j.cnki.psy.2018.09.178.Wang YQ. Discussion on the detection effect of blood glucose meter fingertip blood and venous whole blood blood glucose[J]. Psychological Monthly, 2018(9): 197-198. DOI: 10.19738/j.cnki.psy.2018.09.178. [5] 李海, 李延兵. 《葡萄糖目标范围内时间国际共识》解读[J]. 中华糖尿病杂志, 2019, 11(12): 825-827. DOI: 10.3760/cma.j.issn.1674-5809.2019.12.012.Li H, Li YB. International consensus on time in range[J]. Chin J Diabetes Mellitus, 2019, 11(12): 825-827. DOI: 10.3760/cma.j.issn.1674-5809.2019.12.012. [6] Liu Y, Johnson WO, Gold EB, et al. Bayesian analysis of risk factors for anovulation[J]. Stat Med, 2004, 23(12): 1901-1919. DOI: 10.1002/sim.1773. [7] Mcinturff P, Johnson WO, Cowling D, et al. Modelling risk when binary outcomes are subject to error[J]. Stat Med, 2004, 23(7): 1095-1109. DOI: 10.1002/sim.1656. [8] Magder LS, Hughes JP. Logistic regression when the outcome is measured with uncertainty[J]. Am J Epidemiol, 1997, 146(2): 195-203. DOI: 10.1093/oxfordjournals.aje.a009251. [9] 王泽锋, 史代敏. 基于DIC准则的ASV模型和SV模型的实证比较[J]. 数量经济技术经济研究, 2007, 24(5): 134-141. DOI: 10.3969/j.issn.1000-3894.2007.05.015.Wang ZF, Shi DM. Empirical Comparison between ASV model and SV model based on DIC[J]. The Journal of Quantitative & Technical Economics, 2007, 24(5): 134-141. DOI: 10.3969/j.issn.1000-3894.2007.05.015. [10] 杨贵军, 孟杰, 王双喜. 基于赤池信息准则的分类回归决策树剪枝算法[J]. 计算机应用, 2014, 34(S2): 147-150. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJY2014S2042.htmYang GJ, Meng J, Wang SX. Pruning algorithm of classification and regression decision tree based on Akaike information criteria[J]. Journal of Computer Applications, 2014, 34(S2): 147-150. https://www.cnki.com.cn/Article/CJFDTOTAL-JSJY2014S2042.htm [11] 孙唯佳, 孙建琴, 彭景. 代谢当量在评估体力活动强度及健康效应中的应用[J]. 现代预防医学, 2010, 37(7): 1318-1320, 1323. https://www.cnki.com.cn/Article/CJFDTOTAL-XDYF201007053.htmSun WJ, Sun JQ, Peng J. Application of metabolism equivalent in appraisal of physical strength activity intensity and health effect[J]. Modern Prevent Med, 2010, 37(7): 1318-1320, 1323. https://www.cnki.com.cn/Article/CJFDTOTAL-XDYF201007053.htm [12] Vásárhelyi B. Measurement of hemoglobin A1c: analytical principles and their significance in clinical decision making[J]. Orv Hetil, 2016, 157(19): 753-757. DOI: 10.1556/650.2016.30442. [13] Shepard JG, Airee A, Dake AW, et al. Limitations of A1c Interpretation[J]. South Med J, 2015, 108(12): 724-729. DOI: 10.14423/SMJ.0000000000000381. [14] 李金宝. 即时检验血糖仪指尖血及静脉全血血糖的检测效果[J]. 医疗装备, 2017, 30(11): 36-37. https://www.cnki.com.cn/Article/CJFDTOTAL-YLZB201711022.htmLi JB. The detection effect of the blood glucose of fingertip blood and venous whole blood glucose[J]. Medical Equipmen, 2017, 30(11): 36-37. https://www.cnki.com.cn/Article/CJFDTOTAL-YLZB201711022.htm [15] 蒋云雯, 孙俐, 李晶, 等. 综合管理模式下糖尿病患者糖化血红蛋白达标结果及影响因素分析[J]. 中国慢性病预防与控制, 2020, 28(2): 121-125. DOI: 10.16386/j.cjpccd.issn.1004-6194.2020.02.009.Jiang YW, Sun L, Li J, et al. Analysis of the reaching standard and influencing factors of glycated hemoglobin in diabetes patients under comprehensive management[J]. Chin J Prev Contr Chron Dis, 2020, 28(2): 121-125. DOI: 10.16386/j.cjpccd.issn.1004-6194.2020.02.009. [16] 黄玉玲, 吴琳娜, 曾忠仪, 等. 老年2型糖尿病患者控糖效果影响因素研究[J]. 临床误诊误治, 2020, 33(1): 49-53. DOI: 10.3969/j.issn.1002-3429.2020.01.012.Huang YL, Wu LN, Zeng ZY, et al. Influencing factors of glucose control effects in elderly patients with type 2 diabetes mellitus[J]. Clinical Misdiagnosis & Mistherapy, 2020, 33(1): 49-53. DOI: 10.3969/j.issn.1002-3429.2020.01.012.