[Study on the Selection of Parameters for Evaluating Drug NIR Universal Quantitative Models]

Area of Science:

• Analytical Chemistry
• Chemometrics
• Pharmaceutical Analysis

Background:

• Near-infrared (NIR) spectroscopy is crucial for quantitative analysis of pharmaceuticals.
• Optimizing NIR models requires robust evaluation parameters.
• ICH guidelines provide a framework for analytical method validation.

Purpose of the Study:

• To determine the optimal combination of evaluation parameters for drug NIR universal quantitative models.
• To establish value ranges for these parameters during model optimization.
• To provide evidence for evaluating and improving NIR quantitative models.

Main Methods:

• Collected and analyzed 13 common evaluation parameters from 92 drug NIR universal quantitative models.
• Studied correlations between parameters to identify optimal combinations.
• Utilized metrics such as RMSECV/RMSEP, ARD, RPD, R², and RMSEP/RMSECV ratio.

Main Results:

• Root Mean Square Error of Cross-Validation (RMSECV)/Root Mean Square Error of Prediction (RMSEP) typically within 3%.
• Ratio of Prediction to Deviation (RPD) generally above 2.
• Determination coefficient (R²) for linearity mostly between 80% and 100%.
• Ratio of RMSEP to RMSECV for robustness usually within 1.5.

Conclusions:

• Defined an optimal set of evaluation parameters and their value ranges for drug NIR quantitative models.
• Highlighted the importance of RMSECV/RMSEP, ARD, and RPD for model accuracy.
• Emphasized the need for standardized precision evaluation and further research into specific parameters for NIR models.