基于数据挖掘技术的老年口腔癌患者围术期并发症发生概率评估系统的建立

Establishment of an evaluation system based on data mining tech aimed for predicting the rate of perioperative complications in old oral cancer patients

目的:开发一种具有较强临床实用性的老年口腔癌患者围术期并发症发生概率评估系统,使得相关并发症发生概率的评估更加直观与高效。方法:根据APACHEⅡ以及POSSUM评分体系,结合临床实际,确定核心数据项目。采用数据挖掘方法,分析各项临床数据之间潜在的逻辑关系,并以CAG为平台,开发评估系统。结果:历时7个月,整理收集并回顾性录入513个病例的临床数据,建立包含49个临床录入项目的 50×513维数据集(数据库)。经过一系列测试后,采用随机森林算法作为核心算法,建立预测评估模型,进而开发出预测系统(软件),为系统设置自我学习功能,并计划添加数据导出与网络化功能。结论:本评估系统在老年口腔癌患者围术期并发症发生概率的预测中具有较高的临床实用性。

PURPOSE： To establish an evaluation system based on data mining tech aimed for predicting the rate of perioperative complications in old oral cancer patients with better clinical application. METHODS： Combined APACHE Ⅱ and POSSUM system as well as clinical reality, the key factors and the logical relationship were found. By using the CAG program, the evaluation system was established. RESULTS： Within the period of 7 months, the clinical data of 513 patients were collected, and an data sets sized as 50×513 with 49 key factors was finished. After some tests, the Radom Forest （RF） was chosen as the main algorithm of the model of the evaluation system （program）. The system had the ability for self-study, and is planed to add the function of data output and interact use. CONCLUSION： The evaluation system is good for clinical use to predict the rate of pefioperative complications in old oral cancer patients. Supported by Grant from Shanghai Municipal Bureau of Health （2009076） and the Fifth Innovation Plan for Students of Shanghai University （IAP5145）.