上消化道高风险患者智能随访系统的开发与验证

Development and validation of a rule-based automatic surveillance system for upper gastrointestinal high-risk patients

摘要:

目的：构建自动识别上消化道高风险患者并分配随访间隔的智能随访系统，并验证其能力。方法：回顾性收集武汉大学人民医院2021年1月至2021年10月的23 035例行胃镜检查患者的内镜和病理报告，其中1月至8月的17 934例患者为训练集，9月至10月的5 101例患者为测试集，系统首先提取内镜和病理报告中的关键词，根据关键词自动识别上消化道高风险患者，并将患者分类为7个风险等级，随后依照指南为患者分配标准化随访间隔。测试集内可分配随访间隔的患者中，有189例住院患者可以从医疗记录中获取管床医师给出的随访间隔，以专家医师根据指南基于内镜和病理报告给出的随访间隔为金标准，用于比较不同科室医师与系统分配随访间隔的准确率。用简单随机抽样抽取67例用于探究智能随访系统对3名内镜医师分配随访间隔的辅助作用。结果：智能随访系统识别上消化道高风险患者的总体准确率为99.94%（5 098/5 101），对正确纳入的患者分配随访间隔的总体准确率为100.00%（534/534）。智能随访系统在分配随访间隔上的准确率显著优于全体来自不同科室的医师[98.94%(187/189) 比 35.45%(67/189)，X2=118.01，P＜0.001]，也优于消化内科医师[100.00%(117/117) 比 24.79%(29/117)，X2=86.01，P＜0.001]。在智能随访系统辅助下，3名内镜医师对67例患者分配随访间隔的准确率得到显著提升[55.22%(111/201) 比 22.39%(45/201)，X2=58.68，P＜0.001]。结论：智能随访系统可准确识别上消化道高风险患者并根据风险等级分配相应的随访间隔，有助于减轻医师随访工作量，提高患者的复查率。

Abstract:

Objective To develop an automatic surveillance system to identify upper gastrointestinal high-risk patients and assign surveillance intervals. Methods The endoscopic and pathological reports of 23,035 patients undergoing endoscopy from Renmin Hospital of Wuhan University were collected retrospectively between January to October 2021, of which 17,934 patients from January to August were used as the training set, and 5,101 patients from September to October as the test set. Firstly, keywords in the endoscopic and pathological reports were extracted based on the rules, according to which the high-risk patients were automatically identified and classified into 7 risk levels. Then the standardized surveillance intervals were assigned based on the guideline. Of the patients that could be assigned with surveillance intervals within the test set, 189 were hospitalized and the surveillance intervals given by physicians could be obtained from the electronic health records. We compared the system’s accuracy of assigning surveillance intervals with that of physicians from different departments, using the gold standard of guideline-based surveillance intervals assigned by expert physicians based on endoscopic and pathological reports. Then 67 cases were randomly selected by simple random sampling to evaluated the adjunctive effect of the system in assigning surveillance intervals among 3 endoscopists. Results The overall accuracy of the automatic surveillance system in identifying upper gastrointestinal high-risk patients was 99.94% (5098/5101), and that of assigning surveillance intervals to patients correctly included was 100.00% (534/534). The automatic surveillance system achieved significantly better performance compared with all physicians from different departments [98.94% (187/189) vs 35.45% (67/189), X2=118.01, P<0.001] as well as physicians from gastroenterology [100.00% (117/117) vs 24.79% (29/117)，X2=86.01，P＜0.001]. With the assistance of the automatic surveillance system, the endoscopists’ accuracy of assigning surveillance intervals to 67 patients was significantly improved [55.22% (111/201) vs 22.39% (45/201), X2=58.68, P <0.001]. Conclusion The automatic surveillance system can accurately identify upper gastrointestinal high-risk patients and assign surveillance intervals according to the risk level, which can help reduce the workload of doctors and improve the follow-up rate of patients.