where μ j and σ 2 j are the mean and variance of the j-th test. Note that C b depends in part on the “bias” if the interest is to estimate the difference between the means of the two tests, i.e. μ 1 − μ 2.C b is also called “bias correction factor”. 9 The CCC can therefore be conceived as the product of a measure of coherence (i.e. the Pearson correlation coefficient) and a measure of distortion. That is, the CCC quantifies not only how closely the observations are on the regression line (by ρ), but also how close this regression line is to the line of 45° of perfect agreement (above C b). This section explains the concepts relevant to this guide. Table 6C shows that the percentage of agreement between the new test and the non-reference standard is still 88.2% ((39+6)/51) for unconditional subjects (reference standard − column) and another 98.2% ((4+660)/676) for unconditional subjects (reference standard column). However, the overall percentage that combines subjects with and without conditions is 97.5% ((39 + 6 + 4 + 660) / 727), higher than the original 95.9%. What shows a more dramatic difference, the positive agreement percentage is much lower at 76.8% (43/(43+13)) compared to 90.9%, and the negative approval percentage is slightly higher at 99.2% (666/(666+5)) compared to 97.2%.

This document provides guidance on the submission of applications for advance notice (510(k)) and pre-market authorisation (PMA) for diagnostic equipment (tests). This guide deals with communicating the results of different types of studies evaluating diagnostic devices with two possible endpoints (positive or negative) in SMEs and 510(k)s. The guide is intended for both statisticians and non-statisticians. Estimates of sensitivity and specificity (and other estimates of diagnostic performance) may be subject to bias. Distorted estimates are consistently too high or too low. On average, distorted sensitivity and specificity estimates do not match the actual sensitivity and specificity. Often, the existence, size (size) and direction of distortion cannot be determined. Bias leads to inaccurate estimates. One could also calculate the proportion of New Test+ subjects who are not a reference standard+ and who receive a different number.

Therefore, when calculating the positive and negative percentage match, the FDA recommends explicitly indicating the calculation performed. When choosing an appropriate statistical approach, the theoretical aspects of the data would first be taken into account. .