Are HPLC-UV Methods Fit for Purpose as True Arbiters of Quality for APIs?

• GlaxoSmithKline Biopharmaceuticals

Introduction

Glossary

An evaluation of an analytical method’s specificity should be performed as part of the validation process in accordance with ICH Q2 [1] and the approach used is dependent on the intended objective of the analytical procedure. In reality, certain methods may either be not specific (or not specific enough) for their intended objectives. In these cases, orthogonal approaches using two or more complementary analytical methodologies would be necessary to achieve the appropriate discriminatory power. For example, titrimetric and UV potency assays for API are non-specific and cannot detect the presence of related substances, e.g., process impurities or degradants, but have better precision (ca. 0.1-0.5% RSD) than the corresponding specific HPLC assay methods (>0.5% RSD) and therefore can trend data more effectively [2].

Are HPLC Methods Fit for Purpose?

Hofer et al. [3] modeled the ability of an HPLC assay to rapidly identify significant changes (≥0.5%) in the true mean of an API assay determination. They modeled the potential scenario where for the first 50 batches of a new API, the true mean potency was 99.5% and the standard deviation of the HPLC assay was 0.5%; thereafter the true mean potency dropped to 99.0%, with the same standard deviation. The modeling demonstrated that it is virtually impossible to detect the 0.5% change by trending the HPLC assay data and, more worryingly, if the assessor does indeed believe that a significant change in the process has occurred, it is very difficult to assess when this change took place. Tellingly the lack of this important information will hinder any future investigations into the root cause of that change. The authors advocated the use of a mass balance approach (100% - % total impurities), where the total impurity levels include related substances, solvents, water, non-volatile residues, residual metals, etc. They re-modeled the above simulation using a mass balance approach and confirmed unequivocally that it was relatively simple to detect both the change and, equally importantly, when this change occurred. An additional advantage of this approach is an understanding of changes that occur in the HPLC assay when reference standards are changed or there is a re-designation of the purity value of the existing reference standard. Finally, the authors identified those areas where the existing HPLC assay would still be required: (i) when utilizing API sourced from third-party suppliers, where detailed knowledge of synthesis and related substances may not be fully divulged (for example in a closed DMF) and therefore insufficient data are available to calculate the mass balance assay; (ii) where there is poor mass balance, i.e., where degradation to multiple compounds is seen; (iii) during the early development and scale-up activities, where there may be inadequate knowledge of the impurity fate profile; (iv) when monitoring a process that is insufficiently controlled, where degradation chemistry is not fully understood, where there is the possibility that new impurities may be generated, or where there are concerns of contamination or adulteration; and (v) where there are legally enforceable requirements arising from a pre-defined public standard, for example, to support a pharmacopoeial compendial method.

Intermediate precision is the most appropriate validation parameter for evaluation of process capability (CpK) and should be assessed when proposing any specification limits, or when assessing the capability of the method when the specifications are constrained, i.e., 98.0-102.0% for APIs. The variability associated with the analytical methodology is frequently greater than the variability associated with the manufacturing process, particularly for API manufacture. Tsang [4] showed that for any proposed assay specification operating at 3σ, i.e., process mean ±3σ, a specification of ±2% (4% range) is equivalent to a total variability of 0.67%; thus the method variability needs to be at least half this value, i.e., 0.34%. Methods showing high process capability (often termed 6σ) are those where the total method deviation is ≤ one-twelfth of the total allowable spread or tolerance [2]. From the perspective of standard API specifications (98.0-102.0%), titrimetric methods have process capability of >6σ, whereas most HPLC methods have process capability of only about 3σ.

As a general rule of thumb the standard deviation (σ) of the analytical method should be less than or equal to one-sixth of the proposed specification range, i.e., 6σ capability. Ermer [5] assessed the maximum permitted standard deviation (σ) for an API or drug product assay method and demonstrated the dependence that this has on both the proposed specification range, process capability, and the number of repetitions of the assay (see Table 1).

Table 1. Largest permitted standard deviations (σ) for an assay method (showing the dependence on the proposed specification range and the number of repetitions of the assay; adapted from ermer, 2001 [5])

Thus, for example, for an API assay method using duplicate repetitions to generate a mean potency value, where there is a lower specification limit of 98.0% and with a limit of 0.5% for total impurities (that is, a lower basic specification limit of 99.5%), the analytical method standard deviation should be 0.17% (or less). Even doubling the assay replicates still necessitates an analytical standard deviation of 0.64% (or less). Dejaegher et al. [2] indicated that one way of decreasing method variability was to increase the sample/standard weights fivefold (from ca. 32mg to >160mg); this then aligns the sample sizes to those typically seen for titrimetric methods where the precision is significantly better (ca. 0.1-0.5% RSD). However, Skrdla et al. [6] were skeptical of this approach, indicating that analytical balances in their organization were typically calibrated to a precision of 10.00 ± 0.03mg, i.e., an error of only ±0.1%.

Therefore, the method validation data can impact on the analytical procedure, for example, the number of replicate determinations, size of sample/standards, or the calibration mode required [5].

Building on this initial work [5], Ermer et al. [7] used a total of 2915 assays (utilizing 44 different APIs, manufactured by several different large pharmaceutical companies and using 156 different stability studies) to establish a typical HPLC assay precision assessment. The cumulative API intermediate precision for HPLC assays was found to be 1.1% [8]. Hofer et al. [3] reported that the mean intermediate precision values for API HPLC assays were between 0.6 and 1.1%, with ranges of between 0.2 and 1.7%. This was aligned with Görög [9], who assessed the errors attributable to a drug substance HPLC assay method as being about 1%.

In the Ph. Eur., for potency assays of API, a maximum permitted HPLC system precision is defined, which is dependent on both the upper specification limit and the number of replicate injections. Using an analytical range of 2% (100.0-98.0%, i.e., theoretical mean – lower specification limit), gives an allowable precision of 0.73 and 0.85% RSD, respectively. Similarly, the FDA and Canadian guidelines recommend system precisions of not greater than 1% RSD. Kaminski et al. [10] recently assessed analytical instrument qualification (AIQ) criteria for HPLC equipment. They indicated that the allowable tolerance for precision of injection volume from the auto-injector was proposed to be <1.0% RSD. This is again supportive of typical errors being about 1%

Based on this significant analytical variability, and assuming an allowable API specification ranges of ±2.0% (for specifications in the range of 98.0-102.0%) or in reality -2.0% as the content of the API cannot be greater than 100.0%, several commentators [3,5,6,7,9] have expressed significant reservations about the utility of HPLC assay methods to monitor drug substance quality (to trend changes in API purity, to trend changes in API stability, release batches whose true potency is 98.0-102.0%, or reject batches whose true potency is <98.0% or meaningfully investigate OOS results, that could be attributable to method variability, not specification failures). Skrdla et al. [6] endorsed this view stating that, “assay results are simply not stability-indicating, to the degree required for most such studies to be meaningful (i.e., following ICH guidelines for the reporting of organic impurities), due to the large assay variability associated with them.” The impact of method variability on OOS results is also significantly constrained by FDA’s 2006 guidance, which requires that “all individual sample replicates, as well as the average, fall within the acceptance criteria” [11]. Hofer et al. [3] modeled the probability of finding a false OOS and found that this was very dependent on the method variability and the true mean of the API batch. They also observed that there was only a 1% chance of OOS results when the standard deviation was 0.6%, with a true mean of 99.4%, but this increased markedly (9-fold) when the standard deviation increased to 1%, with the same true mean. The possibility of seeing false OOS results also increases based on the number of tests performed on the same batch, for example, as is the case with routine stability testing. They also modeled this scenario. If the true batch mean is 99.6% and the method variability is modeled as being 0.6, 0.8, or 1.0% (RSD), then the probabilities of observing “false OOS” is relatively low, i.e., 0.4, 2.4, or 6.3%, respectively. Thus, for instance, if this batch is placed on stability with 5 time points (0, 3, 6, 12, and 24 months) and independent duplicate assays are performed at each time point, i.e., 10 assays in total, then the chances of obtaining a “false OOS” increase markedly, using the same true mean and the same method variabilities (0.6, 0.8, or 1.0% RSD), to 4, 22, and 49%, respectively. Therefore, for a stable drug substance placed on stability it is still very likely that “false OOS” results “will be observed within a surprisingly small number of tests.” The authors [3] commented that this will waste significant resources and may result in non-productive measures, as there is nothing wrong with the batch in question; it is a statistical artefact of the method. That is, the API process is under control, the batch is stable, and the batch continues to meet specification—but this is unfortunately not reflected by the data! This of course can be addressed by registering broader specifications that are based on process capability rather than narrower specifications based on regulatory expectations. However, it is a moot point as to whether these broader, more meaningful specification ranges would ever be accepted by regulatory reviewers.

Hofer et al. [3] indicated that the “HPLC assay is more a test of a laboratory’s ability to achieve high precision than of drug substance quality.” Bunnell [12] agreed, stating that although the API HPLC assay gives “potency results within specification, the exact value will not be indicative of quality.” Bunnell [12] also observed that it was practically impossible to meaningfully differentiate between HPLC assays that differ by ≤1%.

Hofer et al. [3] compared the data from the classical external standard HPLC assays versus the mass balance HPLC assay approach, generated on eight API batches. They found that mean assay data were similar (99.85% versus 99.75%), but the precision of the former data (pooled σ 0.55, range 0.31-0.80) was about 6 to 8-fold higher than the corresponding mass balance HPLC assay approach (pooled σ 0.09, range 0.04-0.20). Skrdla et al. [6] proposed the complete elimination of the classical external standard HPLC percent assays from routine use within stability studies, replacing with the more precise mass balance HPLC assay approach, which provides “much better (earlier and more sensitive) detection of low-level degradation products.” The authors claim that the mass balance HPLC assay approach is much better aligned with the current ICH reporting practices (<0.05%) for impurities and degradation products and that its implementation can lead to better trending and significantly less OOS reporting. They indicated that this might necessitate a different approach to the validation of the mass balance HPLC assay, i.e., the use of several orthogonal methods and/or detection approaches might be required as part of a risk mitigation strategy if the standard HPLC assay method were removed from common practice.

Finally, method variability has a deleterious effect on the predicted shelf life of the API or drug product [13]. This is because the “difference between the point estimate of shelf life and its lower confidence limit depends on the width of the confidence interval, which is positively related to the amount of error.” He indicated that for relatively wide intervals, the shelf life determination is often rendered “practically meaningless,” or at best extremely conservative. Magari [13] ran several simulations relating to shelf life prediction and the intrinsic variability encountered and concluded that a 1-year shelf life prediction is only accurate to ±1 month (i.e., ±8.33%). He indicated that utilizing an analytical method that is accurate with a high degree of precision would considerably reduce the shelf life error.

In conclusion, without some relaxation of the current API specification limits (typically, 98.0-102.0%) there seems little doubt that the use of the standard HPLC assay to monitor API quality (to trend changes in API purity, to trend changes in API on stability, to release batches whose true potency is 98.0-102.0%, or to reject batches whose true potency is <98.0% or to meaningfully investigate OOS results) must be approached with severe reservations. Tsang et al. [4], based on a retrospective analysis of the assay data for four different APIs from QC laboratories, as well as R&D, indicated that the default 4% specification range, i.e., 98.0-102.0% did not allow for any meaningful variation in the registered process. In fact, the authors indicated that the assay data would dictate that a 5% specification range, i.e., 97.5-102.5% was more appropriate. They concluded that the quality of the API can be more accurately assessed when HPLC potency data are evaluated holistically, with impurity data and other supporting data. Indeed, this is the original concept of a pharmacopoeial specification (at least in Europe). The Ph. Eur. [14], in discussing specificity of assays indicates that, “For the elaboration of monographs on chemical active substances, the approach generally preferred by the Commission is to provide control of impurities (process- related impurities and degradation products) via a well-designed Tests section, with stability-indicating methods, rather than by the inclusion of an assay that is specific for the active moiety. It is therefore the full set of requirements of a monograph that is designed to ensure that the product is of suitable quality throughout its period of use.”

Several authors ([3,5,6]) have proposed the complete elimination of the existing HPLC external standard assay and replacement with the more precise mass balance HPLC assay approach, which provides significantly better detection of changes in API quality.

References

1. ICH Q2 (R1). 2005. Validation of analytical procedures: Text and methodology. http://www. ich.org/fileadmin/Public_Web_Site/ICH_Products/Guidelines/Quality/Q2_R1/Step4/ Q2_R1__Guideline.pdf. Accessed on 28^th March 2014.
2. B. Dejaegher; M. Jimidar; and M. De Smet, et al. Improving method capability of a drug substance HPLC assay. Journal of Pharmaceutical and Biomedical Analysis 2006; 42: 155-170.
3. J.D. Hofer; B.A. Olsen; and E.C. Rickard. Is HPLC assay for drug substance a useful quality control attribute? Journal of Pharmaceutical and Biomedical Analysis 2007; 44: 906-913.
4. P.K.S. Tsang; J.S.A. Larew; and L.A. Larew, et al. Statistical approaches to determine analytical variability and specifications: I application of experimental design and variance component analysis. Journal of Pharmaceutical and Biomedical Analysis 1998; 16; 1125-1141.
5. J. Ermer. Validation in pharmaceutical analysis. Part I: an integrated approach. Journal of Pharmaceutical and Biomedical Analysis 2001; 24: 755-767.
6. P.J. Skrdla; T. Wang; and V. Antonucci, et al. Use of a quality-by-design approach to justify the removal of the HPLC weight % assay from routine API stability testing protocols. Journal of Pharmaceutical and Biomedical Analysis 2009; 50: 794-796.
7. J. Ermer; P. Arth; and P. De Raeve, et al. Precision from drug stability studies. Investigation of reliable repeatability and intermediate precision of HPLC assay procedures. Journal of Pharmaceutical and Biomedical Analysis 2005a; 38: 653-663.
8. J. Ermer; P. Arth; and P. De Raeve, et al. Validation in pharmaceutical analysis. Part II: central importance of precision to establish acceptance criteria and for verifying and improving the quality of analytical data. Journal of Pharmaceutical and Biomedical Analysis 2005b; 37: 859-870.
9. S. Görög. The sacred cow: the questionable role of assay methods in characterising the quality of bulk pharmaceuticals. Journal of Pharmaceutical and Biomedical Analysis 2005; 36: 931-937.
10. L. Kaminski; M. Degenhardt; and J. Ermer, et al. Efficient and economic HPLC performance qualification. Journal of Pharmaceutical and Biomedical Analysis 2010; 51: 557-564.
11. FDA. 2006. Guidance for industry: investigating Out-of-Specification (OOS) results for pharmaceutical production. US Department of Health and Human Services, Food and Drug Administration, CDER.
12. R.D. Bunnell. Using computer simulated results of a bulk drug assay to determine acceptance criteria for method validation. Pharmaceutical Research 1997; 14: 156-163.
13. R.T. Magari. Uncertainty of measurement and error in stability studies. Journal of Pharmaceutical and Biomedical Analysis 1997; 45: 171-175.
14. Ph. Eur (8^th Edition). 2014. Introduction, General principles, Specificity of Assays. http://www. legemiddelverket.no/Godkjenning_og_regelverk/NLS/Generelle-bestemmelser/Sider/IIIntroduction-( Ph.-Eur.-7th-Ed.).aspx. Accessed on 26th March 2014.

Author Biography

David Elder, Ph.D., studied chemistry at Newcastle upon Tyne (BSc, MSc), before moving to Edinburgh to study for a Ph.D. in Crystallography. Dr. Elder has 36 years’ experience at a variety of pharmaceutical companies (Sterling, Syntex, and GSK). He is currently a director within the Scinovo group in GSK R&D focused on externalization. He has seven patents to his name.