探讨指数加权移动均值法（EWMA）在血清电解质钾、钠、氯项目测定中质量控制的应用。

收集医院实验室信息系统2024年1月至2025年7月电解质钾、钠、氯的检验结果和室内质量控制（IQC）数据，截取近似正态分布范围内患者数据集，基于Python语言编写机器学习程序，采用Shapiro-Wilk检验进行数据正态分布检验，借用Box-Cox法等进行正态转化，完成参数调整、性能验证和数据可视化。统计分析纳入时段中每个设定步长内患者结果的EWMA估计值，计算累积变异系数（CV）并与其IQC的CV相比较；通过Z分数质量控制图将数据可视化，判断预警并分析预警可能原因。

患者检测结果截断浓度范围分别为：钾2.5~6.0 mmol/L、钠125~150 mmol/L、氯90~120 mmol/L；钾、钠、氯最适加权系数均为0.1；钾、钠最适步长为50，氯最适步长为60；纳入时段内钾、钠、氯的EWMA累积CV均小于IQC质量目标CV和累积CV；模型Z分数图共报警9次，钾、钠、氯分别报警了2、4、3次，其中真报警8次（真阳性率88.89%）、假报警1次（假阳性率11.11%）。EWMA模型对系统性误差检出率达100%（8/8），7次实现IQC失控前早期预警，而传统IQC对系统性误差检出率仅12.5%（1/8），对随机误差检出率为100%（11/11）。

本次借助患者数据建立的EWMA程序机器学习模型适用于小样本量实验室，可补充日常传统质量控制的局限性，与传统IQC联动实现全时段质量控制可视化并对血清钾、钠、氯项目分析性能的微小变化和多种误差进行早期预警。

To investigate the application of the exponentially weighted moving average (EWMA) method in the quality control of serum electrolyte assays (K⁺, Na+, Cl^-).

Laboratory test results and intra-laboratory quality control (IQC) data for K⁺, Na⁺, Cl^- from January 2024 to July 2025 were extracted from the hospital laboratory information system (LIS). Patient datasets within an approximately normal distribution range were selected. A machine learning model was developed using Python to perform normality testing by Shapiro-Wilk test, data normalization by Box-Cox transformation with the parameter selection principle of P>0.05 after transformation, parameter optimization, performance validation, and data visualization. The EWMA estimates for patient results were calculated at predefined intervals during the study period, and cumulative coefficients of variation (CV) were computed and compared with the CV targets of IQC. Data visualization was conducted via Z-score quality control charts to identify alerts and analyze potential causes of deviations.

The truncated concentration ranges for patient test results were as follows: K⁺ 2.5–6.0 mmol/L, Na⁺ 125–150 mmol/L, and Cl^- 90–120 mmol/L. The optimal weighting coefficients for K⁺, Na⁺ and Cl^- were all determined to be 0.1,with the optimal step size of 50 for K⁺ and Na⁺, and 60 for Cl^-. During the study period, the cumulative CVs of EWMA for K⁺, Na⁺ and Cl^- were consistently lower than the CV targets and cumulative CVs of IQC. A total of 9 alerts were triggered on the Z-score chart, including 8 true alerts (true positive rate 88.89%) and 1 false alert (false positive rate 11.11%); specifically, K⁺, Na⁺ and Cl^- generated 2, 4, and 3 alerts, respectively. The EWMA model achieved a 100% (8/8) detection rate for systematic errors and provided early warnings before IQC out-of-control for 7 times, while the traditional IQC only had a 12.5% (1/8) detection rate for systematic errors and a 100% (11/11) detection rate for random errors.

The EWMA-based machine learning model, utilizing patient data, is suitable for laboratories with limited sample sizes and complements the limitations of conventional QC methods. This model can make up for the limitations of conventional daily quality control, and its combination with IQC enables the visualization of full-time quality control, as well as the early warning of minor variations in analytical performance and various systematic errors for the determinations of serum K⁺, Na⁺ and Cl^- assays.