pISSN 2234-3806

eISSN 2234-3814

eISSN 2234-3814

Validation and assessment of calibration regression

Method | Interpretation | Acceptance criteria | Other comments |
---|---|---|---|

Plot relationship between residuals and concentration [31, 32, 35] | The optimal regression model and weighting factor will result in randomly distributed variation around the concentration axis. | Not available | Quick and easy graphical visualization; reveals whether the assumptions on the errors and the model are correct. May not always be easy to interpret, especially when the data size is limited. |

Relative errors and sums of relative errors [30, 31] | The optimal regression model will result in the least absolute sum of relative errors and a narrow distribution band in a plot of residual error against concentration. | Acceptable deviation in relative error is 20% at the lower limit of quantitation and 15% for the rest of nominal concentrations [36]. | Less empirical approach to assessing the linearity of a calibration curve compared to graphical visualization. Acceptance criteria of 15%–20% could be considered excessively high [3]. |

ANOVA F-test to compare the variance of the signals at the lowest and highest calibrator concentration levels [14, 31] | Data are heteroscedastic if |
Acceptance limit based on statistical significance. | None |

Bartlett’s or Levene’s test to compare the variances of replicates at all concentration levels [32, 35] | Test for homogeneity of variances. Data are heteroscedastic if |
Acceptance limit based on statistical significance. | Bartlett’s test can be used to compare variances with unequal sample sizes. |

ANOVA partial F-test to compare the variances of linear and quadratic models [14] | If the quadratic calibration model significantly improves the captured variance of the data in comparison with the linear model ( |
Acceptance limit based on statistical significance. | None |

ANOVA-lack of fit test [14] | Lack of fit of the regression model is determined if |
Acceptance limit based on statistical significance. | Sensitive to the number of replicates and calibration levels. |

Ann Lab Med 2023;43:5~18

© Ann Lab Med