Using Smoothing Splines to Quantify Differences in Nonselectivity between In Vitro Diagnostic Medical Devices in Method Comparison Studies and to Assess Commutability of External Quality Assessment Materials

In vitro diagnostic medical device (IVD-MD) comparison studies are important, and there is extensive literature on how such studies should be performed. However, most evaluation approaches assume a linear relationship between the IVD-MD results and do not consider differences in nonselectivity (DINS) among the IVD-MDs. The ordinary least-squares (OLS) approach to estimate DINS can be distorted by nonlinear relationships, leading to overestimation of DINS. This report develops a new model based on smoothing splines to estimate DINS. This model is evaluated against its traditional OLS counterpart. We evaluated both the OLS and smoothing spline DINS models using both clinical sample data and Monte Carlo simulations. The simulations modeled both linear and a set of nonlinear relationships between IVD-MDs. For the external quality assessment material (EQAM) commutability assessment application, Monte Carlo simulations were used to compare the performance of the smoothing spline prediction interval (PI) against the traditional Deming regression PI. The OLS model overestimated DINS under nonlinear relationships between IVD-MDs. In contrast, the smoothing spline DINS estimator did not. The smoothing spline PI achieved empirical coverage probability that was close to the considered PI confidence level (95%) under both linear and nonlinear relationships, while the Deming regression PI performed poorly under nonlinear relationships. The smoothing spline framework provides an accurate approach for quantifying DINS and for assessing EQAM commutability under nonlinear relationships. The smoothing spline DINS estimator is useful for method comparison studies and can enable commutability assessment for IVD-MD pairs previously deemed unsuitable by the linear model framework.