The Pre‐Classified PBRTQC Model Can Reduce the False Positive Rate of K⁺

ABSTRACT Background Patient‐based real‐time quality control (PBRTQC) has garnered increasing attention, yet false positive alerts are common in practical applications. In patients undergoing dialysis, serum potassium (K + ) levels exhibit large fluctuations before and after dialysis, often leading to false positive quality control alerts in routine PBRTQC applications. We aimed to reduce false positive alerts in PBRTQC applications by distinguishing between the test results of dialysis and non‐dialysis patients and constructing separate PBRTQC models. Methods We collected K + test results from 362,077 patients at our center from September 2023 to September 2024. The data were divided into dialysis, physical examination, and non‐dialysis groups, with data from September 2023 to February 2024 comprising the training set. We constructed PBRTQC models for dialysis patients ( n = 3217), those undergoing physical examination ( n = 7339), and non‐dialysis patients ( n = 153,565) using four statistical methods: moving median, moving average, weighted moving average, and exponentially weighted moving average. We validated the three models using data from the dialysis group (validation set 1) from March to September 2024 and the non‐dialysis group (validation set 2) from March to April 2024. By comparing false positive rates, the average number of patient results affected prior to error detection or median number of patient results affected prior to error detection, and the average probability of error detection in the three models, we evaluated whether the pre‐classified PBRTQC model can reduce the false positive rate of K + . Results Statistical analysis revealed significant differences among the dialysis, physical examination, and non‐dialysis groups ( p < 0.001). Based on the minimum sum of the false positive rate, false negative rate, and average number of patient results affected prior to error detection, the models for the dialysis and non‐dialysis groups used the exponentially weighted moving average; the MM method was used in the physical examination group. Validation set 1 showed false positive rates of 69.257% for the physical examination group, 1.143% for the dialysis group, and 35.675% for the non‐dialysis group. According to the total allowable error (TEA), the median number of patient results affected prior to error detection in the dialysis group (1/2TEA, positive: 307.30, negative: 795.20) was higher than that in the physical examination group (1/2TEA, positive: 10.57, negative: 4.67) and non‐dialysis group (1/2TEA, positive: 24.57, negative: 29.57). The average probability of error detection in the dialysis group (1/2TEA, positive: 2.83%, negative: 0.67%) was lower than that in the physical examination group (1/2TEA, positive: 41.47%, negative: 45.11%) and non‐dialysis group (1/2TEA, positive: 16.00%, negative: 18.00%). In validation sets 2 and 3, the false positive rate for the non‐dialysis group and physical examination group was 1.906% and 2.83%, respectively. This indicates that pre‐classifying dialysis specimens can significantly reduce the occurrence of false positives. Additionally, K + results in the non‐dialysis group exhibited notable seasonal variations. Conclusions Establishing PBRTQC models through pre‐classification of dialysis patients can significantly lower the false positive rate of K + , enhancing the accuracy of real‐time monitoring for laboratory testing systems.