针对大型医院即时检测（POCT）血糖设备数量庞大、质量均质化管理难及隐蔽性风险难识别的痛点，验证“风险导向抽样+技术评估”策略在识别系统性风险及指导管理决策中的价值。

建立四方联动机制，引入公共卫生“风险预警”理念，实施风险导向抽样，从北京大学第三医院850余台POCT血糖仪中筛选出115台（约13.5%）具有“饱和运行”或“设备性能微衰减”特征的设备作为风险导向设备，以全自动生化分析仪作为参比，采用5个浓度梯度的新鲜静脉血进行现场比对，Bland-Altman法评估偏倚，进行风险溯源与干预。

3种品牌血糖仪与生化分析仪的相关性良好（r＞0.99），但在生化分析仪测定值13.1 mmol/L处，经Bland-Altman分析显示，品牌Q和H存在系统性负偏倚（bias分别为-0.403 mmol/L、-0.688 mmol/L），相对偏差范围分别达-16.79%～-7.63% 和-17.56%～-9.16%，局部点位突破了常规±15%误差限。尽管依据常规标准3种品牌的总体合格率均为100%，但该策略成功识别出隐匿的浓度依赖性偏倚。数据分析揭示，在相同实验工况下，不同品牌检测系统在高浓度区间呈现出显著的响应特征不一致性。

本策略证实了风险导向抽样策略在有限的医疗资源投入下，能有效地甄别出常规合格率评价所掩盖的、特定品牌设备在高浓度区间系统性负偏倚特征，研究结果为医疗机构对POCT血糖设备实施动态校准（纠正系统性偏倚）、制定科学的淘汰标准、实施靶向临床督导（规避潜在的误差累积）提供了管理策略的循证支撑。

To address the challenges posed by large number of the point of care testing (POCT) equipment for blood glucose in large-scale hospitals, homogeneous quality management and concealed risks, this paper verified the practical value of 'isk-oriented sampling combined with technical evaluation' strategy for detecting systematic risks and guiding management decisions.

Four-party collaboration mechanism was established, risk-oriented sampling was conducted and public health "risk early warning" theory was introduced. From over 850 POCT glucose meters throughout the hospital, 115 devices (around 13.5%) exhibiting characteristic of "saturation operation" or "minor performance degradation" were selected as risk-oriented devices. Using the fully automatic biochemical analyzer as the reference method, fresh venous blood samples at five different concentration gradients were tested for on-site comparison. Bias was assessed using Bland-Altman analysis, followed by risk tracing and intervention.

The results showed that correlation coefficients (r) of the three brands of glucose meters and the biochemical analyzer were all above 0.99. However, at a biochemical analyzer measurement value of 13.1 mmol/L, Bland-Altman analysis results showed systematic negative bias in Brand Q and H (bias of -0.403 mmol/L and -0.688 mmol/L respectively) and the relative bias ranges were -16.79% to -7.63% for Brand Q and -17.56% to -9.16% for Brand H, with a few points exceeding the conventional ±15% error limit. Although an overall pass rate of 100% was achieved for all the three brands based on conventional standard, this strategy successfully identified the concealed concentration-dependent bias. The data analysis revealed that under the same experimental conditions, detection systems of different brands showed significant response inconsistency at high concentration ranges.

The strategy validated that with limited healthcare resource, risk-oriented sampling was able to effectively identify the systematic negative bias at high concentration ranges for devices with specific brands, which is concealed by conventional pass rate evaluation. These findings provide an evidence-based foundation for management strategy on implementing dynamic calibration of blood glucose POCT devices (to correct systematic drift), establishing of scientific retirement criteria, and implementing of targeted clinical supervision (to mitigate potential error accumulation).