两种化学发光检测系统测量不确定度的合成与评估

Composing and evaluating the measurement uncertainty of two kinds of system

目的 使用不同方法对2种化学发光检测系统的测量不确定度进行合成和评估。方法 以临床样本批内CV（CVW%）、批间CV（CVB%）、偏倚不确定度（CVBias%）和校准不确定度（CVcal%）为分量，选择4种合成方法对美国Abbott公司的Architect i2000SR检测系统和美国Backman公司的DXI800检测系统的不确定度进行评估。第1种方法以CVW%、CVB%和CVBias%为分量合成U1%，第2种方法以CVB%、CVcal%为分量合成U2%，第3种方法以CVW%、CVB%和CVcal%为分量合成U3%，第4种方法以CVW%、CVB%、CVBias%和CVcal%为分量合成U4%。采用Pearson和Spearman相关分析、配对t检验及Mann-Whitney u检验进行统计学分析。结果 对于Architect i2000SR检测系统，U1%、U2%、U3%和U4%均相关（r＝0.727～0.988，均P〈0.05），U3%与U2%差别有统计学意义（t＝6.88，P〈0.05），U4%与U1%差别有统计学意义（t＝6.21，P〈0.05）；其检测项目的扩展不确定度与CVW%、CVB%、CVBias%和CVcal%均有良好的相关性。按同样方法评估DXI800系统，U1%、U2%、U3%和U4%均相关（r＝0.608～0.975，均P〈0.05），U3%与U2%差别无统计学意义（z＝－1.33，P〉0.05），U4%与U1%差别无统计学意义（z＝－1.04，P〉0.05）；其检测项目的扩展不确定度与CVW%、CVB%和CVBias%分量相关（rs＝0.653～0.912，均P〈0.05），而与CVcal%无相关性（rs＝0.548，P〉0.05）。结论 对于Architect i2000SR检测系统，宜使用第4种方法合成不确定度，对于DXI800检测系统使用第1种方法合成不确定度更为合适；根据不同分量对测量不确定度的贡献大小，可以通过降低不精密度和偏倚，为相关检测工作的质量改进提供科学依据。

Objective To compose and evaluate the measurement uncertainty of two kinds of chemiluminescence detection system using different methods. Methods The measurement uncertainty was composed by 4 different methods： （1） U1% was composed of within-run CV（CVW%）, between-run CV（CVB%） and bias（CVBias%）; （2） U2% was composed of CVB% and uncertainty of calibration（CVcal%）; （3） U3% was composed of CVW%, CVB% and CVcal%; （4） U4% was composed of CVW%, CVB%, CVBias% and CVcal%. The measurement uncertainty of Architect i2000SR system （Abbott, USA） and DXI800 system （Beckman, USA） was assessed. Pearson correlation analysis, Spearman correlation analysis, Paried t test and Mann-Whitney u test were performed to analyze the data.Results For Architect i2000SR system, U1%, U2%, U3% and U4% were significantly correlated （r=0.727-0.988, all P〈0.05）, U3% and U2% were significantly different （t=6.88, P〈0.05）, U4% and U1% were significantly different （t=6.21, P〈0.05）. For DXI800 system, U1%, U2%, U3% and U4% were also significantly correlated （r=0.608-0.975, all P〈0.05）, no significant difference was found between U3% and U2% （z=-1.33, P〉0.05）, or between U4% and U1% （z=-1.04, P〉0.05）; the expanded measurement uncertainty was correlated with CVW%, CVB%, CVBias%（rs=0.653-0.912, all P〈0.05）, but not with CVcal%（rs=0.548, P〉0.05）.Conclusions For Architect i2000SR system, the fourth method is more proper to compose the measurement uncertainty （U4%）. For DXI800 system, the first method is more appropriate （U1%）. According to the contribution of different components to the measurement uncertainty, the measurement quality could be improved by reducing the imprecision and bias.