Quo Vadis Regulated NIR Analytical Procedures?

Dec 01, 2015
Volume 30, Issue 12, pg 10–16

Some parts of the United States Food and Drug Administration’s new draft guidance entitled “Development and Submission of Near Infrared (NIR) Analytical Procedures” are reviewed and critiqued here. What’s in it for spectroscopists?



Near infrared (NIR) spectrometry as an analytical technique has been used in the pharmaceutical industry for a variety of analytical tasks during the past 25-30 years. Historically, NIR applications may have been based on filter-based instruments and the chemometric modeling technique of choice was multiple linear regression. The skill and time was in choosing the wavelengths to include in the regression model. This generally resulted in applications that were slow to develop, but which were relatively robust. Increased computerization power linked to the easy-to-use application of more-powerful chemometric models in the NIR software resulted in faster development of NIR applications, but like many other analytical techniques, does a user know what is happening inside the instrument and within the software?

NIR is used within the pharmaceutical industry for the following purposes:

  •  identification of raw materials using spectral libraries that are built in-house after the instrument has been  qualified and the software validated for intended use
  •  quantitative analysis
  •  off-line in laboratories or near-line in warehouses or other production areas
  •  on-line or in-line production with process analytical technology (PAT) methods.

IR Versus NIR

Using classical infrared (IR) spectroscopy for sample identification, a sample would be taken and sent to the laboratory, then prepared for analysis. Historically, this involved preparing a Nujol mull, compressing the sample with potassium bromide to form a disk, or preparing a thin film. The sample is then presented to the instrument and a transmission IR spectrum is obtained. More recently, high-pressure attenuated total internal reflection (ATR) has become the sample preparation method of choice for IR identification (1) because it can significantly reduce variation in the IR spectrum caused by sample preparation (Nujol mulls and KBr disks are dependent on the user’s skill). ATR simplifies sample preparation, but it still requires an experienced analyst or spectroscopist to manually perform the comparison of the sample spectrum with the approved reference spectrum for the material. One important aspect of this approach is the need to interpret the sample results and the significance of any differences between the test and reference spectra. Additionally, pharmacopeia wording for identification by IR needs to be understood, because the wording typically implies that differences might be caused by the presence of an undetected polymorph in the test sample. Compare this approach to the use of a NIR system with an optic probe where a nonscientist may be working in a sampling booth in a warehouse. They insert a probe into a drum of material and within seconds the “computer says yes” and the identity is confirmed. All the science is now incorporated into the NIR library and with the spectroscopists who built and validated it for operational use.

NIR analysis is fast when compared with IR, but there is the nonscientist operator versus the trained and experienced scientist to consider as well. At the heart of the discussion is the contrast between an organization investing in training analytical scientists to perform IR analysis versus investing in the NIR application software, selecting a model, and determining how a library has been built.

NIR Deployment Strategy

NIR can be used in both qualitative and quantitative modes for identification and measurement of materials, respectively. Usually qualitative methods are deployed before quantitative ones because the former are easier to develop and validate.

FDA Draft NIR Guidance

In March 2015, the United States Food and Drug Administration (FDA) issued a draft guidance for industry entitled “Development and Submission of NIR Analytical Procedures” (2). At 21 pages, the draft guidance is quite detailed and in this column, we will review and comment on some the contents of the document. Before we begin we just need to consider terminology—often we talk about analytical methods, but the guidance and the United States Pharmacopeia (USP) refer to analytical procedure. Are they the same? No, is the short answer. An analytical procedure covers the whole process from sampling to calculating the reportable result, and the analytical method only refers to the instrumental analysis within the overall procedure.

Scope of the Draft Guidance

The draft guidance (2) covers the development, validation, submission, and maintenance of NIR analytical procedures. The document includes the following sections:

  • Modes of measurement that could be implemented
  • Development of NIR models
  • External validation of NIR analytical procedures
  • Implementing and maintaining NIR procedures
  • Information submitted in an application

Post-approval changes


Figure 1: Diagram of the scope of the draft FDA guidance on NIR analytical procedures.


The scope of the draft guidance is shown diagrammatically in Figure 1. As can be seen, the guidance divides into two main sections covering different ways of using NIR analytical procedures. There are off-line and at-line as well as process control and PAT procedures. The four terms are defined as follows:

  • Off-line: The sample is analyzed away from the process as in, for example, identity testing of raw material samples by NIR in the quality control (QC) laboratory or warehouse.
  • At-line: The sample is removed from, isolated from, and analyzed in close proximity to the process stream or reactor. An example is measurement of tablet assay or content uniformity by NIR, where the NIR analyzer is located next to the tablet press and samples are fed manually or automatically to the instrument.
  • On-line: The sample is diverted to a side stream off the main manufacturing process and may be returned to the process stream or reactor. An example is the measurement of cell density in an anaerobic fermentation process using a flow-through cell.
  • In-line: The sample is not removed from the process stream or reactor. An example is in-line monitoring of blend uniformity by NIR, where the NIR analyzer is interfaced with the blender through a window and takes continuous spectral measurements.

For the purposes of this column, we are restricting our discussion of the draft guidance to off-line or at-line analytical procedures because those are the main applications of NIR used in the pharmaceutical industry. Although NIR can be used quantitatively, we are focusing specifically on NIR used for the qualitative identification of materials, typically incoming raw materials such as active pharmaceutical ingredients (APIs) and excipients. In this situation, the spectrum of the material under test is compared to a composite spectrum of the same material generated over time. This requires that

  • Sufficient numbers of reference materials and hence spectra are used to generate the composite spectrum that will be used to ensure positive identification.
  • If one material is used from more than one supplier, then sufficient samples from each supplier need to be used to generate the reference spectra and allow for inter-supplier differences. From a spectroscopic perspective, it is better to have a single supplier of each material because the library build is simpler.
  • If an organization holds retained samples, the question is which samples can be used to build a library because the storage conditions for retention samples are different than those for reference standards. Therefore, when do retained samples become too old to be useful to generate the reference spectra? There needs to be a process in place for identifying and using these samples to generate reference spectra.


Exclusions from the Guidance

The guidance explicitly excludes any discussion of how to set up or qualify the instrument itself including any accessories that may be used such as an autosampler or a fiber-optic probe, but it does state the requirement to implement both instrument maintenance and the NIR procedure (for example, the calibration model). This aspect will come under the authority of USP <1058> on analytical instrument qualification (AIQ) (3) for the approach and the individual chapters in the USP and European Pharmacopoeia (EP) for the details. The draft guidance also does not discuss which mode of measurement is best, such as transmission or reflectance, because measurement mode is in the domain of the developer of the analytical procedure. 

Pharmacopeial Chapters on NIR

Currently, there are general chapters for NIR in both the USP (4) and EP (5).

USP <1119>, entitled “Near Infrared Spectrometry,” (4) covers a general introduction to the technique and then discusses the following aspects:

  • Measurement modes—transmission and reflectance
  • Factors that affect NIR spectra
  • Instrumentation: equipment, NIR reference spectra, qualification of the instrument
  • Method validation: validation parameters, on-going method evaluation and method transfer

Of note is the fact that USP <1119> is currently an informational general chapter rather than a mandatory one. When revised, <1119> will be split into two general chapters: one that will be mandatory (USP <856>) and one that is best practice that is informational (USP <1856>) (6). The new mandatory chapter will contain the instrument parameters and tolerances for qualification and the process will be contained in the updated general chapter <1058>.

EP 2.2.40, strangely enough also entitled “Near-Infrared Spectrometry” (5), has the following sections:

  • Apparatus
  • Measurement methods
  • Sample preparation and presentation
  • Factors affecting spectral response
  • Pretreatment of NIR spectra data
  • Control of instrument performance
  • Qualitative analysis (identification and characterization)
  • Limit analysis
  • Trend analysis
  • Quantitative analysis
  • On-going model evaluation
  • Transfer of databases

Although the chapter is only six pages long, it contains a lot of information that requires careful reading and interpretation.

To keep the discussion to a reasonable length we have excluded publications on NIR from the Therapeutic Goods Administration (TGA), World Health Organization (WHO), and Japanese Pharmacopoeia (JP). As Smith and Sellors (1) note for Fourier transform infrared (FT-IR) spectroscopy, training is essential to ensure compliance with pharmacopeial requirements and that differences between monographs and regulatory agency guidance are taken into account when developing analytical procedures. The same is also true for NIR methods.

Qualification, Validation, and Verification

Fitness for intended use is a mandatory requirement of any instrument and associated software (7). The first part of this process is the qualification of the instrument coupled with the integrated validation of the associated software and is shown as the bottom two layers in Figure 2. However, these steps are only part of the overall process for fitness of purpose when using NIR for identification. The second, and most important, part is building, validating, and maintaining the spectral libraries used for the identification of materials. This process is shown in the top section of Figure 2 with the selection of the calibration model to be used and the discrimination required (what is the number and structure of compounds to be included in the library and how close are they in structure?). From these two elements of information, the number of libraries that would be required can be deduced. Additional factors are the number of suppliers of each material coupled with batch consistency: only one supplier producing a consistent product is easier to handle than multiple suppliers of the same product where different manufacturing processes produce differences in NIR spectra. Validation of the library is discussed later in this column.

Figure 2: Qualification and validation of an NIR spectrometer followed by library build for identification of materials.


Linkage to ICH Q2(R1)

When reading the draft NIR guidance, an immediate problem that is seen is that it is based on the International Conference on Harmonization (ICH) Q2(R1) on the subject of validation of analytical methods (7). Unfortunately, this document is out of date and is scheduled for revision by ICH in 2017 following an update of the USP general chapters on the subject. Currently, there are three general chapters in the USP dealing with transfer of analytical procedures, validation of compendial procedures, and verification of compendial procedures (8-10). Martin and colleagues (11) in a stimulus to the revision process article in Pharmacopeial Forum have suggested that a life-cycle approach would be much better, covering development through to the end of life of an analytical procedure. The aim would be to retire the existing three chapters in the USP and replace them with a new mandatory chapter called USP <220> containing mandatory requirements for validation, verification, and transfer as well as an informational chapter <1220> containing best practice guidance. The overall aim is to move away from the tick-box approach that the pharmaceutical industry currently uses, based simply on the contents of ICH Q2(R1).

However, as we shall see in the next section, the draft FDA guidance notes that ICH Q2 was written before NIR spectroscopy became routine for identification of materials in good manufacturing practice (GMP) facilities.

Development of NIR Models for Identification
Before we go into discussion on the draft guidance, we need to consider one aspect of the GMP regulations under the Laboratory Controls sub-part. Section 211.160(b) (12) requires that anything done in the laboratory must be scientifically sound. Therefore, the underlying message before even reading the FDA guidance is that anybody working in this area needs to understand the science and specifically what the instrument and the software are doing. “Computer says yes” is not a valid rationale for working in this area and we discuss this topic further under the section on training later in this column.

The draft guidance (2) states that NIR analytical procedures typically combine the following steps:

  • elements of instrumentation (analyzer consisting of a NIR spectrophotometer, reflectance or transmission probe, spectral analysis software, and so forth),
  • acquisition parameters,
  • sample presentation (interface) and sampling,
  • composition of spectral data sets,
  • spectral pretreatment,
  • wavelength range or ranges, and
  • a chemometric model.

It further states that the procedures can therefore be considered more complicated than the types of analytical procedures for which ICH Q2(R1) was written.

Consider the NIR process outlined in Figure 2. After the qualification of the instrument there is the library build, which can take several months depending on the number of materials, the number of batches to build the library as well as the time to verify that the library works and can correctly identify materials. Therefore, an NIR analytical procedure for identification of material can be more complex and take longer to develop and validate than, say, a chromatographic procedure for determining the purity of an active ingredient.  

From a philosophical perspective, consider the following questions about quality by design of NIR libraries:

·      How do you get calibration with sufficient variation within the design space as the process can be adjusted within the design space without a problem? A decision needs to be made at the design space stage that you will be using NIR analysis, as mentioned in the draft guidance (2).

·      Considering the relationship between the design space of the manufacturing process for a material and the NIR calibration, if you change the process do you invalidate the calibration?

The key to confirming identity by NIR is knowing and understanding the science involved and how the software operates so that meaningful decisions can be made. At the core is knowing how the NIR calibration model works, as shown in Figure 3.

Figure 3: Use of NIR for identification of materials.


Model Selection

Model selection is the section in the draft guidance (2) that focuses on development of chemometric models. There are various models that can be used to identify compounds from simple comparison type models such as correlation coefficient, to powerful chemometric models such as principal component analysis (PCA), partial least squares (PLS), or principal component regression (PCR). The key requirement here is that the spectroscopist who develops the library needs to know how the selected calibration model works and how it discriminates between one compound and another even with batch to batch variation of a single compound. Part of the problem is that whatever model is used, the selected calibration model produces a number that determines if the spectrum matches or not.

Therefore, we need to consider a number of questions:

·      Is the generated number significant to identify the material?

·      When is the number significant enough to fail?

·      How do you defend the model to avoid the accusation that the data have been manipulated?

To help with the overall approach, a multidisciplinary team should be assembled to work with developing the methods and the libraries especially focused on the statistics elements of the chemometric models used for generating the spectral reference libraries. Often these people are not available.

In generating the model, using PCA, for example, principal component factors are a fundamental part of the model. The factors represent decreasing levels of orthogonal data variance, so the first factor, for example, will typically resemble something such as the average spectrum, while other factors may contain less information, with lower number factors including components that may relate more to instrument noise or even a representation of the optical signature of the instrument. Building models that include more principal component factors generally result in a better fit of the spectra included within the model. However, the draft guidance (2) cautions about over-fitting factors as this reduces the robustness of the model (for example, the model is very good at predicting the properties of the spectra included within it, but poor at modelling spectra not included in the data set). The instrument application software used typically includes statistical tools to help determine the optimum number of factors, but the need to include a multidisciplinary team needs to be emphasized. For example, if the standard error of prediction (SEP) of the calibration spectra was over relied upon by an inexperienced analyst, the model might be over-fitted (to reduce the SEP).

NIR spectra represent a complex mix of information. Generally, NIR spectra are associated with overlap and combination bands of mid-IR absorptions; therefore, although some regions can be attributed to particular features (such as water), they are typically difficult to interpret. Additionally, the NIR spectra of solid materials are also associated with the scattering properties of the material and can be dependent on particle size. The draft guidance (2) advises that electronic spectra should be pretreated to reduce variation to particle size and techniques such as derivative spectroscopy are often used. Scattering properties and particle-size dependency can be both a curse and a blessing. The process used to produce the material is changed, a new calibration model may be required (the curse), but this might also detect changes in the process used by a raw material supplier (the blessing). It’s all a matter of perspective, but it has implications for the way libraries are set up and how different suppliers of material may need to be managed in the library. In an extreme instance, a new library may be required when a new material supplier is used.

Data Integrity Issues with NIR

The hottest topic in regulated laboratories currently is data integrity, or rather the lack of it. It is vitally important to have clear rules of how to investigate and manage failures to verify the identity of a material using NIR. Is the failure a result of the sample, an outlier not being within the calibration set of the material, or because of poor quality of the sample? It is important, as with data integrity at the top of the regulatory menu, that this is dealt with in a documented and transparent way otherwise an organization can be seen as testing into compliance or worse falsifying data.

Currently the regulatory focus is on chromatography data systems regarding data integrity, but it can easily move to NIR to assess how the calibration model was built, used, and maintained. The scope for potential manipulation because of either stupidity or intent is huge. Moreover, how are complete data defined for an NIR analysis to comply with 21 CFR 211.194(a) (12)?


Where to Deploy NIR for Identification?
Although it appears to be a large amount of work to establish and validate the library, nonetheless it is possible to save money with NIR identification methods. However, to do so requires relatively large batches of raw materials to get the consistency of identification. As such, it is not suitable for pilot-scale work in early development as there are not enough samples to provide sufficient numbers nor batches for adequate throughput.

Moreover, there is also the impact of different suppliers who provide the same raw material or active ingredient. Differences in the production conditions can produce the same chemical substance with different NIR characteristics, which will complicate building the library and the discrimination of the target material with closely related compounds.

External Validation of NIR Analytical Procedures
Please remember that NIR is an alternate analytical technique not an alternative technique under a pharmacopeia, especially the USP. This is not a play on words. If one looks at any drug or excipient monograph in any pharmacopeia there is a list of analytical techniques for purity, identification, and so on. Of the test methods listed, not one will be NIR; you may find IR but not NIR. In the case of a dispute with an analysis, the pharmacopeial test takes precedence and is legally enforceable. Therefore, a second analytical method should be used to cross validate the NIR method and this is typically IR.

The FDA draft guidance recommends that the developed library is validated for specificity, which is defined in ICH Q2(R1) as the “ability to assess unequivocally the analyte in the presence of components which may be expected to be present” (2,7). This generalized statement needs to be interpreted for identity testing using NIR. Typically, if a company purchases a lot of lactose as an excipient, what is expected to be present? Well, and there is no simpler way to express this: lactose, lactose, and only lactose. However, the specificity test needs to take samples of lactose and be able to distinguish them from other compounds to answer the following question: Is NIR appropriate for the identity testing of this analyte?

Skills to Develop, Use and Maintain NIR Libraries
Do you know what is happening inside the NIR system, which comprises both the instrument and the application software? Of the two components, it is the application software that is the critical one. In many cases, the developed library is used by a nonscientist in a sampling booth in the goods-inwards section of the warehouse. The training element here is focused on the sampling, presentation of the material to a probe or instrument, and reporting the results. The ease of the operation is totally dependent on the spectroscopist who developed the spectra library or libraries being used.

Does the spectroscopist understand what is happening when a particular model is selected? What about over- or under-fitting the model? Application software is easy to operate, but is the decision based on the application of scientific knowledge or simply a lucky dip selection?

The catch-22 problem is that the library should incorporate the expected variation of the materials, but we don’t know the underlying population of sample spectra. For identification of materials with low intrinsic variability, simpler algorithms such as those based on correlation coefficient may be used. These have the advantage that they are conceptually simpler to understand and explain as seen in Figure 3. A correlation coefficient is calculated between the sample spectrum and the average library spectrum for each material. The same approach may also be used with IR because the use of diamond ATR and similar techniques can significantly reduce variation resulting from sample preparation. However, determining a pass-fail cutoff point that can be justified in an audit can be difficult (because the sample population is not known). If a limit is set too low, the NIR process will incorrectly identify materials, and if it’s set too high there may be a high incidence of test failures to manage.

Two of the more common approaches are to apply action and warning limits, so set a potentially high limit, accepting that some samples may fail this, but having a clear, well defined, and robust process to investigate the failure. If the library includes all materials that identification will be performed on, then the limit may be set through the validation procedure.

For materials that are more variable in nature, then simple comparison algorithms may be less effective and more robust chemometric models are required in which the intrinsic variability of the material is incorporated into the chemometric model and the identification process fits the model to the sample spectrum. The draft guidance warns against the inclusion of outliers within the calibration set, including the wording, “samples with high leverage or high residuals” (2). Here, the term residuals refers to the difference between the sample spectrum and the spectrum the chemometric model has fitted. From a data integrity perspective, one uncertainty relates to the extent to which electronic data such as residuals needs to be examined by a reviewer, bearing in mind that chemometric models typically incorporate information about the goodness of fit for the model. Fundamentally, the laboratory needs to be able to defend decisions about what spectra are incorporated into the library, bearing in mind the underlying sample population is not known.


We have reviewed some aspects contained in the draft FDA guidance for industry on the development and submission of NIR analytical procedures. This column installment does not cover all the topics within the document, and readers are encouraged to read the whole guidance. The focus in this column is on the use of NIR for identity testing and specifically on aspects of building and maintaining spectral libraries.


  1. P. Smith and J. Sellors, Pharm. Technol. Eur. 23(9), 85–89 (2011).
  2. US Food and Drug Administration, “Draft Guidance for Industry, Development and Submission of Near Infrared Analytical Procedures” (FDA, Rockville, Maryland, March 2015).
  3. General Chapter <1058> “Analytical Instrument Qualification” in United States Pharmacopeia 37-National Formulary 32 (United States Pharmacopeial Convention, Rockville, Maryland, 2014).
  4. General Chapter <1119> “Near-Infrared Spectroscopy” in United States Pharmacopeia 37-National Formulary 32 (United States Pharmacopeial Convention, Rockville, Maryland).
  5. General Text 2.2.40 “Near-Infrared Spectroscopy,” European Pharmacopoeia (European Directorate for the Quality of Medicines, Strasbourg, France). 
  6. C. Burgess and J.P. Hammond, Spectrosc. Eur. 27(1), 26–29 (2015)
  7. International Conference on Harmonization, ICH Q2(R1) Validation of Analytical Procedures: Text and Methodology (ICH, Geneva, Switzerland, 2005).
  8. General Chapter <1224> “Transfer of Analytical Procedures” in United States Pharmacopeia 37-National Formulary 32 (United States Pharmacopeial Convention, Rockville, Maryland, 2014).
  9. General Chapter <1225> “Validation of Compendial Methods” in United States Pharmacopeia 37-National Formulary 32 (United States Pharmacopeial Convention, Rockville, Maryland, 2014).
  10. General Chapter <1226> “Verification of Compendial Procedures” in United States Pharmacopeia 37-National Formulary 32 (United States Pharmacopeial Convention, Rockville, Maryland, 2014).
  11. G.P. Martin et al., “Lifecycle Management of Analytical Procedures: Method Development, Procedure Performance Qualification, and Procedure Performance Verification,” Pharmacopeial Forum 39(5), September-October, 2012.
  12. Code of Federal Regulations (CFR), Part 211, Current Good Manufacturing Practice for Finished Pharmaceutical Goods, 21 CFR 211 (U.S. Government Printing Office, Washington, D.C., 2008), clauses 211.160(b) and 211.194(a).


Paul Smith is Global Strategic Compliance Program Manager at Agilent Technologies. After initially specializing in spectroscopy and application of chemometrics to spectroscopic data, Paul developed his compliance expertise in a variety of quality and management roles within the 17 years he spent in the pharmaceutical industry. Paul worked as an independent consultant and university lecturer before moving into laboratory compliance consultancy and productivity roles.


R.D. McDowall is the Principal of McDowall Consulting and the director of R.D. McDowall Limited, as well as the editor of the “Questions of Quality” column for LCGC Europe, Spectroscopy’s sister magazine. Direct correspondence to: [email protected]


lorem ipsum