依据大数据分析探索如何有效选择重点监管对象和质控指标，尝试建立依据数据分析精准定位质控管理方向和策略的科学质控管理体系。

对2018年四川省155家哨点医院超声医学科的25个质控指标进行多元统计分析。首先对质控指标采用Pearson相关性分析，探索指标之间是否存在共线性；其次用主成分分析法将多元质控数据投影到主平面，进行异常值探索并采用最小协方差聚类判别（MCD）算法给出统计差异的95%可信区间；最终根据主成分的得分图得到医院的聚类信息以及根据主成分的载荷图发现对探测聚类贡献度最大的质控指标集。

MCD算法可以明确指出在同级医疗机构中超声质控指标异常分布的医院。如华西二院，其第二主成分明显高于其他三级甲等医院，结合载荷图发现其低年资医师人数和硕士以上学历人员远高于其他医院，表示其人才储备最丰富。其中有6家三甲医院、7家三乙医院、6家二甲医院、6家二乙及其他医院得分异常，为重点监管对象。载荷图中的主要差异指标进行成分分析显示：年龄25~35岁的医师占比、住院医师占比、学士学位占比是三甲医院质控评价差异的主要原因；超声检查中门诊患者占比、超声检查中住院患者占比、年龄36~45岁的医师占比和病理符合率是三乙医院质控评价差异的主要原因；学士学位医师占比、学士以下学历医师占比、年龄>45岁的医师占比和超声科医患比是二甲医院质控评价差异的主要原因；年龄35~45岁的医师占比和门诊占比是二乙及其他等级医院质控差异的主要原因。

基于大数据分析各级医疗机构超声质控数据可以精准定位质控管理对象和质控监管具体指标。

To explore how to effectively select key hospitals that need quality control management and quality control indicators, and try to establish a quality control management system based on big data analysis to accurately determine the direction and strategy of quality control management.

In this study, 25 quality control indicators of ultrasonic medicine department of 155 sentinel hospitals in Sichuan Province were analyzed by multivariate statistical analysis. First, Pearson correlation analysis was used to explore whether there was collinearity between the indicators. Then, multivariate quality control data were projected onto the main plane to explore the outliers, and the 95% confidence interval of statistical differences was given by using the minimum covariance determinant (MCD) algorithm. Finally, according to the score chart and load chart, the quality control index set which had the greatest contribution to the hospital abnormal clustering was detected.

Using the MCD algorithm, we can clearly identify hospitals with abnormal distribution of ultrasonic quality control indicators among the same level medical institutions. For example, it was found that in West China Second University Hospital & Sichuan University, the numbers of junior doctors and those with master's degree or above were higher than those of other hospitals based on the MCD algorithm combined with the load chart, indicating that talent reserve in this hospital was the most abundant. The analysis of the components of the main different indexes in the load chart showed that the main reasons for the quality control difference in the tertiary class A hospitals were the proportion of doctors aged 25-35 years, the proportion of residents, and the proportion of doctors with a bachelor's degree. The main reasons for the quality control difference in the tertiary class B hospitals were the proportion of outpatients, the proportion of inpatients, the proportion of doctors aged 36-45 years, and the rate of coincidence with pathological results. The main reasons for the quality control difference in the secondary class A hospitals were the proportion of doctors with a bachelor's degree, the proportion of doctors with an education level below bachelor's degree, the proportion of doctors aged >45 years, the proportion of doctors, and the staff/patient ratio in ultrasound department. The main reasons for the quality control difference in the secondary class B or lower hospitals were the proportion of doctors aged from 35 to 45 years and the proportion of outpatients.

Based on big data analysis of ultrasonic quality control data among different hospitals, we can accurately determine the hospitals that need quality control management and specific quality control indicators.