Chinese General Practice ›› 2023, Vol. 26 ›› Issue (32): 4013-4019.DOI: 10.12114/j.issn.1007-9572.2023.0175

Special Issue: 内分泌代谢性疾病最新文章合集

• Article • Previous Articles     Next Articles

Construction of Recurrence Risk Prediction Model for Diabetic Foot Ulcer on the Basis of Logistic Regression, Support Vector Machine and BP Neural Network Model

  

  1. Department of Burn Plastic Surgery, General Hospital of Ningxia Medical University, Yinchuan 750001, China
  • Received:2023-03-22 Revised:2023-05-23 Published:2023-11-15 Online:2023-05-31
  • Contact: MA Qiang

糖尿病足溃疡复发风险预测模型的构建:基于Logistic回归和支持向量机及BP神经网络模型

  

  1. 750001 宁夏回族自治区银川市,宁夏医科大学总医院烧伤整形美容科
  • 通讯作者: 马强
  • 作者简介:
    作者贡献:张娟提出研究思路,设计研究方案,研究的实施与可行性分析,包括复发风险预测模型建模思路,数据收集,文章的构思和论文撰写;张娟、李海芬、李小曼、姚苗、马惠珍负责筛选研究对象,数据采集和整理,构建预测模型统计学处理,结果的分析与解释,图表的设计和制作;马强负责论文的修订,负责文章的质量控制及审校,提供研究经费及材料支持,对文章整体负责,监督管理。
  • 基金资助:
    宁夏自然科学基金项目(2022AAC03515); 宁夏回族自治区重点研发计划项目(2020BEG03025)

Abstract:

Background

The rates of first and multiple recurrence of diabetic foot ulcers (DFUs) are increasing annually worldwide, and the risk of early recurrence is higher than the distant recurrence. There are numerous risk factors for DFUs recurrence, and there is a lack of systematic screening. Therefore, there is a need to explore the risk factors for DFUs recurrence in order to identify high-risk population of recurrence at an early stage.

Objective

To explore the predictive value of Logistic regression (LR), support vector machine (SVM), BP neural network model (BPNN) in the recurrence risk of DFUs.

Methods

From January 2020 to October 2021, a total of patients with DFUs attending the Department of Burn Plastic Surgery, Endocrinology and Wound Ostomy Outpatient Department in General Hospital of Ningxia Medical University were selected as the research objects and divided into the recurrence group (n=116, 29.7%) and non-recurrence group (n=274, 70.3%) according to the recurrence of DFUs within 1 year after discharge. General information was collected and compared between the two groups of patients, including sociodemographic characteristics, medical history assessment and clinical case information. The Diabetes Foot Self-care Behavior Scale (DFSBS) was used to assess the self-management behavior of diabetes foot in patients and chronic diseases risk perception questionnaire was used to assess the risk perception level of DFUs among patients. Multivariable Logistic regression analysis was used to explore the influencing factors of DFUs recurrence in patients within 1 year after discharge. The patients were divided into training and test sets according to the ratio of 7 to 3, the LR, SVM and BPNN recurrence risk prediction models were developed based on Logistic regression variable screening strategy. The receiver operating characteristic (ROC) curves of each model were plotted to predict the recurrence risk of DFUs.

Results

There were significant differences in BMI, living alone, duration of diabetes, history of smoking, history of alcohol consumption, history of involved toe amputation, classification of diabetic foot ulcers, ankle-brachial index, glycated hemoglobin, sole ulcer, toe involvement, walking impairment, osteomyelitis, multidrug-resistant bacteria infection, diabetic peripheral neuropathy, lower limb atherosclerosis, self-management behavior of diabetes foot, level of risk perception in both groups of DFUs patients (P<0.05). Multivariable Logistic regression analysis showed that BMI〔OR=0.394, 95%CI (0.285, 0.546), P<0.001〕, duration of diabetes〔OR=1.635, 95%CI (1.303, 2.051), P<0.001〕, history of smoking〔OR=0.186, 95%CI (0.080, 0.434), P<0.001〕, classification of diabetic foot ulcers〔OR=2.139, 95%CI (1.133, 4.038), P=0.019〕, glycated hemoglobin〔OR=2.289, 95%CI (1.485, 3.528), P<0.001〕, sole ulcer〔OR=3.148, 95%CI (1.344, 7.373), P=0.008〕, self-management behavior of diabetes foot〔OR=0.744, 95%CI (0.673, 0.822), P<0.001〕and level of risk perception〔OR=0.892, 95%CI (0.845, 0.942), P<0.001〕were influencing factors of the recurrence of DFUs within 1 year (P<0.05). The accuracy rates of LR, SVM and BPNN models to predict the recurrence risk of DFUs in the test sets were 82.43%, 94.87% and 87.17%, with AUCs of 0.843, 0.937 and 0.820, respectively. There were significant differences in AUC of ROC curves of LR, SVM and BPNN recurrence risk prediction models of DFUs (Z=2.741, P<0.05) ; the AUC of ROC curve of SVM recurrence risk prediction model was higher than the LR and BPNN models (Z=5.937, P=0.013; Z=3.946, P<0.001) .

Conclusion

SVM model can predict the recurrence risk of DFUs patients within 1 year after discharge with good accuracy rate, sensitivity, specificity, AUC and other indicators, which is the relative optimal model. It is recommended to further promote and apply the prediction model to verify its effectiveness.

Key words: Diabetes mellitus, Foot ulcer, Diabetic foot, Recurrence, Logistic models, Support vector machine, Back propagation neural network, Root cause analysis

摘要:

背景

全球范围内糖尿病足溃疡(DFUs)首次复发与再次复发率逐年上升,且早期复发风险高于远期风险。导致DFUs复发的风险因素较多,目前缺乏系统的筛选,因此需要探索DFUs复发的危险因素,以便早期识别复发高危人群。

目的

探讨Logistic回归、支持向量机(SVM)和BP神经网络(BPNN)模型在DFUs复发风险中的预测价值。

方法

选取2020年1月—2021年10月在宁夏医科大学总医院烧伤整形美容科、内分泌科和伤口造口门诊就诊的DFUs患者390例作为开发模型的研究对象。根据患者出院后1年内DFUs是否复发分为复发组116例(29.7%)和非复发组274例(70.3%)。收集两组患者的一般资料包括社会人口学特征、病史评估和临床病例资料并进行比较,采用糖尿病足部自我管理行为量表(DFSBS)评估患者糖尿病足部自我管理行为,采用慢性病风险感知问卷评估患者DFUs风险感知水平。采用多因素Logistic回归分析探讨DFUs患者出院后1年内DFUs复发的影响因素;将患者按照7∶3划分为训练集和测试集,运用Logistic回归变量筛选策略,分别建立Logistic回归、SVM和BPNN模型;绘制各模型预测DFUs复发风险的受试者工作特征(ROC)曲线。

结果

两组DFUs患者BMI、独居、糖尿病病程、吸烟史、饮酒史、受累足趾截肢史、足溃疡分级、踝肱指数、糖化血红蛋白、溃疡位置在脚底、足趾受累、足部存在行走障碍、骨髓炎、多重耐药菌感染、糖尿病周围神经病变、下肢动脉粥样硬化、足部自我管理行为、DFUs风险感知水平比较,差异均有统计学意义(P<0.05)。多因素Logistic回归分析结果显示,BMI〔OR=0.394,95%CI(0.285,0.546),P<0.001〕、糖尿病病程〔OR=1.635,95%CI(1.303,2.051),P<0.001〕、吸烟史〔OR=0.186,95%CI(0.080,0.434),P<0.001〕、足溃疡分级〔OR=2.139,95%CI(1.133,4.038),P=0.019〕、糖化血红蛋白〔OR=2.289,95%CI(1.485,3.528),P<0.001〕、溃疡位置在脚底〔OR=3.148,95%CI(1.344,7.373),P=0.008〕、足部自我管理行为〔OR=0.744,95%CI(0.673,0.822),P<0.001〕和DFUs风险感知水平〔OR=0.892,95%CI(0.845,0.942),P<0.001〕是DFUs患者1年内DFUs复发的影响因素。Logistic回归、SVM和BPNN模型在测试集中预测DFUs复发风险的正确率分别82.43%、94.87%、87.17%,ROC曲线下面积(AUC)分别为0.843、0.937、0.820。Logistic回归、SVM和BPNN模型预测DFUs复发风险的ROC曲线AUC比较,差异有统计学意义(Z=2.741,P<0.05);SVM模型预测DFUs复发风险的ROC曲线AUC高于Logistic回归和BPNN模型(Z=5.937,P=0.013;Z=3.946,P<0.001)。

结论

SVM模型预测DFUs患者出院后1年内DFUs复发风险的正确率、灵敏度、特异度、AUC等指标均较好,为相对最优的模型,建议进一步推广应用以验证预测模型的效能。

关键词: 糖尿病, 足溃疡, 糖尿病足, 复发, Logistic模型, 支持向量机模型, BP神经网络模型, 影响因素分析