There is a joke about the fellow who comes into the plant and says, “I’m from the government; I’m here to help you.” In an effort to help pharma, the ICH (International Conference on Harmonization) represents the countries interested in unified standards: the FDA, EMA and other drug regulatory authorities; Departments of Health of Australia, Brazil, China, Taipei, India, Russia, Singapore and South Korea all have permanent representatives. Also included are Europe (EFPIA), Japan (JPMA), PhRMA and the International Federation of Pharmaceutical Manufacturers & Associations (IFPMA).
ICH developed more than 60 guidelines to eliminate duplication in the development and registration process, so that a single set of studies can be generated to demonstrate the quality, safety and efficacy of a new product. The ICH facilitates international electronic communication through the provision of Electronic Standards for the Transfer of Regulatory Information (ESTRI), giving the Electronic Common Technical Document (eCTD), allowing for the electronic submission of the Common Technical Document from applicant to regulator. The Medical Dictionary for Regulatory Activities Terminology is also from the ICH.
It holds biannual Geneva meetings of the Steering Committee, rotating between EU, Japan and United States. Their guidelines are not intended to be comprehensive, but are intended to be used in combination with regional requirements. Guidelines can be used as a means of reducing testing duplication.
These are good intentions of the involved parties. Unfortunately, that many chefs in the kitchen have multiple points of view on what is the “best” way to perform any test. The EMA alone has numerous independent agencies, often many within a single country. Guidelines take years to produce a draft for comment, with the final version taking more time to be approved.
This is not critical for “classical” analyses (e.g., Karl Fischer), but newer technologies are introduced and advance too rapidly for this arcane process to keep current. Technologies that rely on Chemometrics advance even more rapidly; NIR, Raman and THz are examples of “Guidance-lag” technologies.
Since PAT and QbD were introduced, NIR and Raman have become lynchpins of controlling many processes. The USP chose to retain NIR and Raman as “information-only” chapters, allowing reviewers to reject them for analyses, out of hand. Consequently, the chapters are not as robust as they could be, with no plans to update them.
The EMA and FDA have generated Guidances for NIR in the pharmaceutical industry. Unfortunately, they used the ICH Q 2(B) (validation of analytical methods) for their model. Since the Agency Guidances are based on the ICH Guidance, the methodology for validation is largely based on Beer’s law — the method of calculating peaks in HPLC. Beer’s law assumes several things: linearity within the range measured, non-reactivity between the matrix and analyte, known pathlength, etc.
In NIR, none of these parameters apply. The matrix absorbs as well as the analyte, in a solid sample and there are seldom clean peaks to measure; the peaks move, due to hydrogen bonding; and there is rarely a hint of the pathlength (even in transmission).
The most egregious error is assuming that NIR and HPLC are equivalent. In HPLC, the chemist can generate “recovery” standards of between 50 and 150 percent of label claim. “Samples” are merely API and excipients weighed into a volumetric flask, solvent added, and the API “recovered.” No actual doses were made within that range.
For NIR, intact dosage forms are required. It may be difficult to impossible to actually make tablets in ranges of 70 to 130 percent of label claim. The math (linear regression via R or R2) may be apropos to Beer’s law (with clear, isolated peaks), but NIR has overlapping, complex peaks. Physical state changes also cause shifts due to H-bonding, pathlength and particle size variation. Several levels of “synthetic” samples are made, seldom representing physical parameters of an actual product.
I would ask the agencies to consult analysts who have successfully generated models, based on production samples when they update their next set of guidances.