人工智能电话随访在高血压随访管理中的应用

The application of artificial intelligence telephone follow-up in the management of hypertension follow-up

目的评价人工智能电话随访(以下简称AI随访)的接通采集情况与识别准确性,为应用于高血压随访管理提供可行性依据。方法于2019年4月18日至5月9日,对打浦桥街道社区卫生服务中心的4 486例在管高血压患者进行AI随访,分析总体信息采集率及完整信息采集率;并在有效接通的2 476例患者中随机抽取10%,由调查员重听原始录音并记录抽取的248例患者的随访情况,进一步分析AI识别的准确性。采用SPSS 22.0软件进行χ^(2)检验和趋势χ^(2)检验。结果总体信息采集率为53.5%,随患者年龄的增长而升高,<60、60~69、70~79、≥80岁组分别为46.5%、53.9%、55.9%和54.1%;工作日的采集率(55.1%)高于双休日(47.8%),差异均有统计学意义(P<0.05)。完整信息采集率为45.2%,随患者年龄的升高而降低(P<0.05),<60、60~69、70~79、≥80岁组分别为52.9%、46.7%、46.3%和35.5%。各题的采集率随序次增长而下降,差异有统计学意义(P<0.01)。AI识别总体准确性为90.2%,且不同性别、年龄组、拨打时间对应的准确率差异均无统计学意义(P>0.05)。结论 AI随访具有一定的信息采集效率和较高的识别准确性,应用于社区慢性病随访能降低社区医生工作量,在大规模慢性病人群随访中具有良好的应用前景。

Objective To evaluate the collection efficiency and identification accuracy of artificial intelligence telephone follow-up(AI follow-up),and provide the feasible basis for application in the management of hypertension follow-up. Methods A total of4 486 hypertension patients in management at Dapuqiao Community Health Service Center were followed up by artificial intelligence telephone(AI) from April 18 to May 9 in 2019,the overall information collection rate and the complete information collection rate were analyzed. Then 10% of the 2 476 patients who were effectively connected were selected randomly. The investigators listened the original recordings again and recorded the content of the selected 248 patients for further analysis of the accuracy of AI recognition. The χ^(2)test and trend χ^(2)test were used to analyze the data. The used software was SPSS 22.0. Results The overall information collection rate was 53.5%,which increased with age(the collection rates of <60,60-69,70-79,≥80 years old groups were 46.5%,53.9%,55.9% and 54.1%,respectively). And the collection rate on working days was higher than that on weekends(55.1% and 47.8%,respectively),P <0.05. The complete information collection rate was 45.2%,which decreased with age(the collection rates of <60,60-69,70-79,≥80 years old groups were 52.9%,46.7%,46.3% and 35.5%,respectively),P<0.01. The overall accuracy of AI recognition was 90.2%,and there were not significant differences between groups with different gender,age and follow-up time(P>0.05). Conclusion The AI telephone follow-up system has information collection efficiency to a certain extent and high recognition accuracy. The application of AI follow-up can reduce the workload of community doctors and have a good application prospect for large-scale follow-up of chronic diseases.