期刊文献+
共找到2篇文章
< 1 >
每页显示 20 50 100
基于机器学习算法探讨糖尿病视网膜病变的风险因素 被引量:13
1
作者 宋亚男 武惠韬 +5 位作者 应俊 李琬悦 陈康 刘铁城 张卯年 张颖 《解放军医学院学报》 CAS 北大核心 2021年第9期906-912,992,共8页
背景糖尿病视网膜病变(diabetic retinopathy,DR)是糖尿病患者主要并发症之一,其病程进行性发展可致视功能损伤甚至失明。探索影响DR进展的临床因素对糖尿病患者预防、控制和管理DR具有重要意义。目的通过机器学习算法和沙普利可加性特... 背景糖尿病视网膜病变(diabetic retinopathy,DR)是糖尿病患者主要并发症之一,其病程进行性发展可致视功能损伤甚至失明。探索影响DR进展的临床因素对糖尿病患者预防、控制和管理DR具有重要意义。目的通过机器学习算法和沙普利可加性特征解释方法(SHAP)分析探讨2型糖尿病患者并发DR的风险因素。方法回顾性分析“国家人口与健康科学数据共享平台”公布的“解放军总医院糖尿病并发症预警数据集”3000例2型糖尿病患者的临床资料,对58项观察变量在无DR并发症(non diabetic retinopathy,NDR)患者和并发DR患者两组组间进行基线分析以及差异性检验;评判XGBoost、随机森林、logistic回归三种机器学习算法,采用递归特征消除(RFE)和XGBoost机器学习算法选取最优模型预测变量,并对变量特征权重值排序;应用SHAP方法对模型的风险因子进行解释分析。结果DR组的高血压症(收缩压/舒张压)、糖化血红蛋白、血脂水平(总胆固醇、低密度脂蛋白)、脑卒中、肾病(血尿素、血肌酐、血尿酸)、肾衰、下肢动脉病变等并发比例或指标水平高于NDR组(P<0.05),而年龄、冠心病、心肌梗死、高脂血症、动脉粥样硬化症等低于NDR组(P<0.05)。XGBoost较其他模型表现更佳,模型中排在前十位的重要区分特征为肾病、冠心病、下肢动脉病变、身高、其他肿瘤、糖化血红蛋白、血尿素、血清白蛋白、肾衰、高脂血症。SHAP集成散点图解释XGBoost模型中变量的重要性依次为糖化血红蛋白(0.59)、肾病(0.44)、血尿素(0.32)、下肢动脉病变(0.25),四项的SHAP值>0且绝对值均高。同时SHAP值分布呈现明显分类,即DR的显著危险因素。糖化血红蛋白、肾病、血尿素对DR病程影响呈现潜在交互关系,且血尿素>5 mmol/L时DR风险显著升高。结论XGBoost算法和SHAP模型可用于预测糖尿病患者DR的风险因素及解释特征变量交互关系,提示糖化血红蛋白、合并肾病、血尿素水平对DR这一2型糖尿病微血管并发症的高风险预测性。 展开更多
关键词 糖尿病视网膜病变 机器学习 危险因素分析 2型糖尿病 糖尿病并发症
下载PDF
A new classification for epidemiological study of me- chanical eye injuries 被引量:1
2
作者 Xiao Jianhe zhang maonian +4 位作者 Li Shiyang Jiang Caihui Jiang Hua zhang Ying Qiu Huaiyu 《Chinese Journal of Traumatology》 CAS CSCD 2014年第1期35-37,共3页
Objective: Considering the difficulty in classifying some cases with eye trauma by Birmingham Eye Trauma Terminology (BETT) in our epidemiological study, we introduce a new classification for epidemiological study ... Objective: Considering the difficulty in classifying some cases with eye trauma by Birmingham Eye Trauma Terminology (BETT) in our epidemiological study, we introduce a new classification for epidemiological study of mechanical eye injuries based on BETT. Methods: A retrospective investigation was carried out in 31 hospitals from January 2005 to December 2010. All medical records of inpatients with eye injuries were reviewed. A total of l0 718 patients (11 227 eyes) were diagnosed as mechanical eye injuries. All mechanical eye injuries were tried to be classified using BETT. While some eye injuries were difficult to categorize. We recorded the injury type and case number. A new classification based on BETT was also used for the same project. Results: Of 10 718 patients (11 227 eyes) with me- chanical eye injuries, the following cases cannot be classi- fied by BETT: 1 488 patients (1 559 eyes) with merely orbitalor ocular adnexa injury, 1 961 (2 054) globe injuries associ- ated with orbital or ocular adnexa injury, 271 (284) ocular surface foreign body (OSFB) or ocular wall foreign body (OWFB), 77 (89) contusion, 9 (11) lamellar laceration asso- ciated with OSFB or OWFB, 29 (30) rupture associated with OSFB, OWFB or intraocular foreign body and 60 (62) lace- ration associated with OSFB or OWFB. While according to our new classification, all eye injuries can be categorized without any difficulty. Conclusion: Difficulty in classifying some eye injuries in epidemiological study by BETT brings some trouble to our study, which can be solved by our new eye injury clas- sification to some extent. It is hoped that other ophthal- mologists present better ones to make the classification more perfect. 展开更多
关键词 Eye injuries CLASSIFICATION EPIDEMIOLOGY
原文传递
上一页 1 下一页 到第
使用帮助 返回顶部