Fauskanger Pernille Kjeilen, Sandberg Sverre, Johansen Jesper, Keller Thomas, Budd Jeffrey, Miller W Greg, Stavelin Anne, Delatour Vincent, Panteghini Mauro, Støve Bård
Norwegian Organization for Quality Improvement of Laboratory Examinations (Noklus), Haraldsplass Deaconess Hospital, Bergen, Norway.
Department of Mathematics, University of Bergen, Bergen, Norway.
Biom J. 2025 Feb;67(1):e70032. doi: 10.1002/bimj.70032.
Correct measurement results from in vitro diagnostic (IVD) medical devices (MD) are crucial for optimal patient care. The performance of IVD-MDs is often assessed through method comparison studies. Such studies can be compromised by the influence of various factors. The effect of these factors must be examined in every method comparison study, for example, nonselectivity differences between compared IVD-MDs are examined. Historically, selectivity or nonselectivity has been defined as a qualitative term. However, a quantification of nonselectivity differences between IVD-MDs is needed. This paper fills this need by introducing a novel measure for quantifying differences in nonselectivity (DINS) between a pair of IVD-MDs. Assuming one of the IVD-MDs involved in the comparison exhibits high selectivity for the analyte, it becomes feasible to quantify nonselectivity in the other IVD-MD by employing this DINS measure. Our approach leverages elements from univariate ordinary least squares regression and incorporates repeatability IVD-MD variances, resulting in a normalized measure. We also introduce a plug-in estimator for this measure, which is notably linked to the average relative increase in prediction interval widths attributable to DINS. This connection is exploited to establish a criterion for identifying excessive DINS utilizing a proof-of-hazard approach. Utilizing Monte Carlo simulations, we investigate how the estimator relates to population characteristics like DINS and heteroskedasticity. We find that DINS impacts the mean, variance, and 99th percentile of the estimator, while heteroskedasticity affects only the latter two, and to a considerably smaller extent compared to DINS. Importantly, the size of the study design modulates these effects. We also confirm, when using clinical data, that DINS between pairs of IVD-MDs influence the estimator correspondingly to those of simulated data. Thus, the proposed estimator serves as an effective metric for quantifying DINS between IVD-MDs and helping to determine the quality of a method comparison study.
体外诊断（IVD）医疗器械（MD）的正确测量结果对于实现最佳患者护理至关重要。IVD-MD的性能通常通过方法比较研究来评估。此类研究可能会受到各种因素的影响。在每项方法比较研究中都必须检查这些因素的影响，例如，检查所比较的IVD-MD之间的非选择性差异。从历史上看，选择性或非选择性一直被定义为一个定性术语。然而，需要对IVD-MD之间的非选择性差异进行量化。本文通过引入一种新颖的度量来满足这一需求，该度量用于量化一对IVD-MD之间的非选择性差异（DINS）。假设参与比较的IVD-MD之一对分析物具有高选择性，那么通过采用这种DINS度量来量化另一个IVD-MD中的非选择性就变得可行。我们的方法利用了单变量普通最小二乘回归的元素，并纳入了IVD-MD的重复性方差，从而得到一个标准化的度量。我们还为该度量引入了一个插件估计器，该估计器特别与归因于DINS的预测区间宽度的平均相对增加相关联。利用这种联系，通过风险证明方法建立了一个识别过度DINS的标准。利用蒙特卡罗模拟，我们研究了估计器与DINS和异方差等总体特征之间的关系。我们发现DINS会影响估计器的均值、方差和第99百分位数，而异方差仅影响后两者，并且与DINS相比影响程度要小得多。重要的是，研究设计的规模会调节这些影响。我们还通过临床数据证实，IVD-MD对之间的DINS对估计器的影响与模拟数据的影响相对应。因此，所提出的估计器是量化IVD-MD之间DINS并有助于确定方法比较研究质量的有效指标。
