278
B. Igne et al.
12.1 Introduction
An analytical method is a collection of documents and procedures describing how
an analytical signal is collected and processed, how information is generated from
the signal, and how it is reported. It is not only the chemometrics model that takes
spectra as inputs to provide predictions or classifications, but also a description of
how the samples are measured (frequency, instrument configuration, and operation),
how the instrument performance is monitored (hardware calibration, frequency of
recalibration and system suitability), how the model was built, tested, validated (if
applicable), and will be maintained throughout its lifecycle, and what is done with
the output (reporting, link to informatics systems). In this chapter, the general procedure for method development and lifecycle management will be discussed through
the concept of analytical quality by design [1, 2] (AQbD) developed by the pharmaceutical industry, but applicable to all analytical fields of use of near-infrared
spectroscopy (NIRS).
12.2 General Procedure for Method Development,
Validation, and Lifecycle
Near-infrared spectra are usually too complex for directly establishing a relationship
between an absorbance at a particular wavelength and a parameter of interest. While
for clear liquids Beer’s law may be directly applicable, most samples will exhibit
diffuse reflectance or transmittance. The resulting spectra will differ in path lengths
and require the development of empirical models. These models can be supervised
or unsupervised depending on the intended use. The development of these models
must follow a rigorous methodology to ensure that the resulting analytical method
meets its intended purpose. The analytical quality by design framework presents an
approach to building, testing, validation, and maintaining a NIR method.
Practitioners of near-infrared spectroscopy will be very familiar with the process
described in Fig. 12.1 corresponding to a generic flow diagram of how a supervised
model is built, tested, and validated if required by the regulatory environment of
use. The model is critical, but only a part of an analytical method and the general
framework of method development will be discussed in the next section. However,
general principles need to be explored before focusing on the AQbD framework. The
following paragraphs present generalities about model development.
As indicated in Fig. 12.1, a calibration set, corresponding to relevant variability for
which the model will need to account during operational deployment (i.e., variability
in chemical and physical parameters expected to be encountered), is regressed against
the “true” (known or measured) quantity in the parameter of interest. The range of
sample variability included in the model will dictate how robust a model is. If the
model does not span an appropriate range (what the model is expected to encounter
during deployment), the model usefulness will be limited to the variability included
B. Igne et al.
12.1 Introduction
An analytical method is a collection of documents and procedures describing how
an analytical signal is collected and processed, how information is generated from
the signal, and how it is reported. It is not only the chemometrics model that takes
spectra as inputs to provide predictions or classifications, but also a description of
how the samples are measured (frequency, instrument configuration, and operation),
how the instrument performance is monitored (hardware calibration, frequency of
recalibration and system suitability), how the model was built, tested, validated (if
applicable), and will be maintained throughout its lifecycle, and what is done with
the output (reporting, link to informatics systems). In this chapter, the general procedure for method development and lifecycle management will be discussed through
the concept of analytical quality by design [1, 2] (AQbD) developed by the pharmaceutical industry, but applicable to all analytical fields of use of near-infrared
spectroscopy (NIRS).
12.2 General Procedure for Method Development,
Validation, and Lifecycle
Near-infrared spectra are usually too complex for directly establishing a relationship
between an absorbance at a particular wavelength and a parameter of interest. While
for clear liquids Beer’s law may be directly applicable, most samples will exhibit
diffuse reflectance or transmittance. The resulting spectra will differ in path lengths
and require the development of empirical models. These models can be supervised
or unsupervised depending on the intended use. The development of these models
must follow a rigorous methodology to ensure that the resulting analytical method
meets its intended purpose. The analytical quality by design framework presents an
approach to building, testing, validation, and maintaining a NIR method.
Practitioners of near-infrared spectroscopy will be very familiar with the process
described in Fig. 12.1 corresponding to a generic flow diagram of how a supervised
model is built, tested, and validated if required by the regulatory environment of
use. The model is critical, but only a part of an analytical method and the general
framework of method development will be discussed in the next section. However,
general principles need to be explored before focusing on the AQbD framework. The
following paragraphs present generalities about model development.
As indicated in Fig. 12.1, a calibration set, corresponding to relevant variability for
which the model will need to account during operational deployment (i.e., variability
in chemical and physical parameters expected to be encountered), is regressed against
the “true” (known or measured) quantity in the parameter of interest. The range of
sample variability included in the model will dictate how robust a model is. If the
model does not span an appropriate range (what the model is expected to encounter
during deployment), the model usefulness will be limited to the variability included