Finding a reliable limit of detection in the NIR determination of residual moisture in a freeze-dried drug product.

The validation of an analytical method in the pharmaceutical industry follows strictly regulated guidelines. The introduction of multivariable calibration methods requires a revision of these recommendations, since some of them are contradictory regarding the limit of detection (LOD). This work compares the LOD values obtained using pseudounivariate and multivariate procedures in the PLS-NIR determination of residual moisture content (RMC) in a freeze-dried drug. As NIR has proved to be more precise than Karl-Fischer at low RMC values, LOD has been estimated by ordinary and by orthogonal least squares regression. The precision of the RMC determination in approx. 2000 industrial vials was used as an indirect evidence of the reliability of the LOD values obtained. The effect of reducing the number of calibration samples and increasing the RMC values have also been studied. No significant differences were observed using a number of calibration samples ≥ 20. Based on our findings, when the size of the calibration sample set is high and the range of RMC values is close to the limit, the LOD estimated with the ICH formula and using orthogonal regression should be recommended. If water content moves away, the ICH formula should be replaced by the LOD equation as a practical, reliable and simple procedure.