280
B. Igne et al.
Once the calibration algorithm and testing approach are chosen, an iterative
process of sample selection, variable selection, and spectral preprocessing is initiated, with the goal of obtaining the most suitable model for the intended purpose.
Once that model has been developed, it should be tested, and if necessary validated,
against independent data to confirm its performance.
The workflow displayed in Fig. 12.1 should however only be applied as a result
of a careful consideration of the method purpose and expected performance and
numerous activities should take place before a single model is built. The frequency
at which spectra are collected should ensure that the information generated by the
method meets the intended goals of the analytical method and should be driven
by the measured process variability and the analytical instrument capability. The
most accurate model may not prove useful if not associated with a method that
measures the right information. For instance, an instrument known to be affected by
environmental conditions that cannot be controlled effectively (e.g., temperature) but
referenced only at the beginning and end of a week-long campaign may not allow the
method to perform appropriately. Another consideration is related to the error of the
reference method. If the desired error of the NIRS method is significantly lower than
the error of the reference method, the model may not be able to meet expectations.
Finally, if variability in raw materials are expected beyond what can be included
in the calibration set (i.e., year-to-year variability of natural products), resulting in
a significant change in the spectra, the model will lack robustness and need to be
updated. If collecting new samples for reference analysis and model update is not
feasible, the validity of the method will be limited.
For these reasons, following the analytical quality by design framework can be of
significant help to set the parameters of the method and ensure that the expectations
are set before undergoing a long and costly method development activity.
12.3 Analytical Quality by Design (AQbD)
12.3.1 Introduction to AQbD
The concept of analytical quality by design arises from the quality by design (QbD)
concept documented by the International Council for Harmonisation of Technical
Requirements for Pharmaceuticals for Human Use (ICH) [6]. Quality by design is
defined as “a systematic approach to development that begins with predefined objectives and emphasizes […] understanding and […] control, based on sound science
and quality risk management”. A process developed using the QbD framework is
well understood and delivers quality product within a design space identified during
method development. For analytical methods, AQbD allows for a “well understood,
fit for purpose, and robust method that consistently delivers the intended performance
throughout its lifecycle” [2]. Figure 12.2 presents the components of AQbD. Each
item will be discussed below.
B. Igne et al.
Once the calibration algorithm and testing approach are chosen, an iterative
process of sample selection, variable selection, and spectral preprocessing is initiated, with the goal of obtaining the most suitable model for the intended purpose.
Once that model has been developed, it should be tested, and if necessary validated,
against independent data to confirm its performance.
The workflow displayed in Fig. 12.1 should however only be applied as a result
of a careful consideration of the method purpose and expected performance and
numerous activities should take place before a single model is built. The frequency
at which spectra are collected should ensure that the information generated by the
method meets the intended goals of the analytical method and should be driven
by the measured process variability and the analytical instrument capability. The
most accurate model may not prove useful if not associated with a method that
measures the right information. For instance, an instrument known to be affected by
environmental conditions that cannot be controlled effectively (e.g., temperature) but
referenced only at the beginning and end of a week-long campaign may not allow the
method to perform appropriately. Another consideration is related to the error of the
reference method. If the desired error of the NIRS method is significantly lower than
the error of the reference method, the model may not be able to meet expectations.
Finally, if variability in raw materials are expected beyond what can be included
in the calibration set (i.e., year-to-year variability of natural products), resulting in
a significant change in the spectra, the model will lack robustness and need to be
updated. If collecting new samples for reference analysis and model update is not
feasible, the validity of the method will be limited.
For these reasons, following the analytical quality by design framework can be of
significant help to set the parameters of the method and ensure that the expectations
are set before undergoing a long and costly method development activity.
12.3 Analytical Quality by Design (AQbD)
12.3.1 Introduction to AQbD
The concept of analytical quality by design arises from the quality by design (QbD)
concept documented by the International Council for Harmonisation of Technical
Requirements for Pharmaceuticals for Human Use (ICH) [6]. Quality by design is
defined as “a systematic approach to development that begins with predefined objectives and emphasizes […] understanding and […] control, based on sound science
and quality risk management”. A process developed using the QbD framework is
well understood and delivers quality product within a design space identified during
method development. For analytical methods, AQbD allows for a “well understood,
fit for purpose, and robust method that consistently delivers the intended performance
throughout its lifecycle” [2]. Figure 12.2 presents the components of AQbD. Each
item will be discussed below.