Clinical and Laboratory Standards Institute guideline EP06—Evaluation of Linearity of Quantitative Measurement Procedures is intended to provide both manufacturers and users of quantitative measurement procedures with an economical and user-friendly method of validating and verifying the linearity interval. This guideline also can be used to determine the extent to which a quantitative measurement procedure meets medical requirements or the manufacturer's linearity interval claims.

This guideline replaces the previous edition of the approved guideline, EP06-A, published in 2003. 

The first edition, EP06-P, published in October 1986, relied on fitting a straight line to measurements of five equally spaced samples, four replicates each, judging linearity by a goodness-of-fit test based on comparing dispersion around the regression line with the repeatability (ie, within-run imprecision) exhibited in the experiment. Unfortunately, this statistical test puts measurement procedures with excellent repeatability at risk of inappropriately failing. Conversely, it might fail to identify nonlinearity in measurement procedures with very poor repeatability. 

To rectify this shortcoming, the second edition, EP06-P2, published in December 2001, and the first approved guideline, EP06-A, published in April 2003, adopted a different and computationally more complex statistical test for linearity. EP06-A called for fitting not only first-order but also second- and third-order polynomials (ie, linear, quadratic, and cubic models) to the data, judging the measurement procedure to be linear if, by internal statistical criteria, the first-order fit is best. In effect, EP06-A asked whether the trajectory of experimental results had a shape more closely resembling a straight line rather than a parabolic or sigmoidal curve. Unfortunately, this method placed no restriction on the trajectory's orientation. EP06-A, unlike major publications cited therein, was not sufficiently clear that, with suitable allowance for random error, the trajectory should be aligned with the origin. (Intuitively, for example, a measurement procedure exhibiting little or no decrease in measured results under progressive dilutions, such as so-called "analog" procedures for free thyroxine, is not considered linear even when the trajectory of results approximates a straight-line segment.) 

This edition of EP06 builds on the previous editions, introducing several important refinements, including: 

• The discussion of dilution schemes, designed to minimize errors in preparing the test panels, has been extended. There is no longer any suggestion that samples need to be equally spaced. This guideline encourages judicious interpolation of additional mixtures to improve coverage of concentration gaps between calibrators, as well as concentrations important for decision-making or monitoring. 

• Like EP06-A, this edition emphasizes that suitable visualizations of the study data are important, and many examples are provided. 

• Consistent with other CLSI method evaluation guidelines, this guideline calls for judging results in terms of the clinical acceptability of deviations (ie, deviations from linearity at each of the sample concentrations), as opposed to a global pass-or-fail assessment based solely on internal statistical criteria. This point of view makes this guideline's approach more relevant to clinical practice and more informative as to the location, magnitude, and significance of any deviations from linearity. 

• Chapter 3 is devoted to validating linearity (intended for manufacturers and developers), and Chapter 4 covers verifying (ie, spot-checking) linearity (intended for end-user laboratories). 

• Two study designs are discussed: one study design includes a high sample (whose concentration is known to exceed the procedure's analytical measuring interval) and a measurand-free sample. The other study design includes high and low samples with known concentrations or a known concentration ratio. These designs serve different purposes, have different limitations, and use somewhat different data analyses. 

• Computationally, this edition's approach is simpler than that of EP06-A, insofar as fitting second- and third-order polynomials is no longer included for validating or verifying linearity (although developers might find such analysis informative). Conversely, weighted first-order regression analysis is recommended under appropriate circumstances to limit the risk of failure due to chance. Advice is provided on determining adequate sample-specific weights in the absence of a precision profile. 

• The importance of stating a performance claim is emphasized.

This guideline provides recommendations for designing, analyzing, and interpreting linearity studies for quantitative measurement procedures. This guideline is intended for manufacturers and developers seeking to validate the linearity of a measurement procedure throughout a stated concentration interval, especially the interval that includes the measurement procedure's lower limit of quantitation (LLoQ) and upper limit of quantitation (ULoQ). It is also intended for laboratorians who verify the linearity of a measurement procedure and for regulatory agencies responsible for overseeing in vitro diagnostic (IVD) manufacturers or end-user laboratories. 

This guideline does not include information on linearity issues encountered during the measurement procedure development phase, such as efficiently identifying the widest possible interval for a linearity claim or selecting calibration points, although the experimental design and data analysis principles described herein can be of value during that phase. 

Before the laboratory begins formal linearity verification studies, the measurement procedure's intended analytical measuring interval claim should already have been determined based on the results of linearity, precision, and other studies that have been evaluated using a clinically informed error budget for imprecision, bias, etc.