|
| 基于SMOTE算法和决策树的绝经后骨质疏松性骨折分类模型建构 |
| The construction of postmenopausal osteoporosis model based on SMOTE and strategy tree method |
| |
| DOI：10.3969/j.issn.1006-7108.2019.01.001 |
| 中文关键词: 骨质疏松性骨折 风险评估 SMOTE过抽样 决策树模型 |
| 英文关键词:osteoporotic fracture risk evaluation SMOTE strategy tree model |
| 基金项目:基金项目:国家自然科学基金面上项目(81373885);北京市中医药科技发展资金项目(JJ2015-57) |
|
| 摘要点击次数: 1416 |
| 全文下载次数: 624 |
| 中文摘要: |
| 目的 构建符合北京、上海两地40~65岁女性人口学特征的危险因素和中医症状相结合的骨质疏松性骨折早期风险预测工具。方法 本研究采用注册登记式研究的方法，于2009年3-8月在北京市东城区及上海市徐汇区收集的1 823例40~65岁女性骨质疏松症高危人群的危险因素及中医症状信息，进行连续3年的登记观察。采用SMOTE过抽样算法平衡数据，基于决策树模型筛选与骨质疏松症骨折有关的危险因素及中医症状，并建立骨质疏松性骨折风险评估工具。结果 本研究选择C4.5算法作为预测模型建立工具。首先筛选出对绝经后骨质疏松性骨折高危患者发生脆性骨折的危险因素，然后建立预测模型。由于样本量较小，在节点的设置中采用交叉验证，Mode选用Expert，修剪纯度设为75，采用全局修剪。根据此生长和修剪规则，所建立分类树模型共包括5层，共19个结点，共筛选出6个解释变量。各指标按重要程度从大到小依次为骨密度、目眩、肉类、生产次数、视物模糊和乏力。经过逐层各影响因素的分类，最终骨折人群比例占13%。对该预测模型预测概率绘制受试者工作特征曲线，结果显示曲线下面积为0.871(95 %CI =0.8226~0.9211。结论 初步建立了基于北京、上海人口学特征40~65岁女性骨质疏松性骨折分类模型。 |
| 英文摘要: |
| Objective To construct a risk predictive tool of early osteoporotic fractures according to the characteristics of 40-65-year-old females and the Chinese medical syndromes. Methods Data of the risk factors and Chinese medical syndromes of 1 823 40-65-year-old women, who were in East District Beijing and Xuhui District Shanghai, were collected using registration method.The observation continued for 3 years. The data were balanced using SMOTE over stratified method. The risk factors and Chinese medical syndromes were screened base on the strategy tree model. The risk evaluation tool was established. Results The C4.5 calculation method was used as tool to establish the prediction model. The risk factors of fragile fractures in high risk patients were firstly screened out. The predictive model was then established. Due to the small sample size, cross-validation was adopted in the censor setting. Expert was selected in Mode. The purity of trimming was set to 75，and overall trimming was adopted. According to the growth and trimming rules，the stratified tree model established included 5 layers，19 censor points，and 6 explanatory variables. The parameters were bone mineral density, dizziness，meat, number of productions，blurred vision, and fatigue, in order of importance. After the stratification of each influencing factor by layer, the proportion of the final fracture population accounted for 13%. The prediction probability of the prediction model was used to draw the subject's working characteristic curve, and the result showed that the area under the curve was 0. 871(95% CI=0. 8226-0. 9211).Conclusion Based on the demographic characteristics of Beijing and Shanghai, the stratification model of osteoporotic fractures in women aged 40 to 65 years was established. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |