Determining Number of Process Performance Qualification Batches Using Statistical Tools
Prior Discussion Paper: “Topic 1 – Stage 2 Process Validation: Determining and Justifying the Number of Process Performance Qualification Batches (Version 2)”
Authors: Christopher Breen (Eli Lilly), Dilip Ayyala Somayajula (Grand River Aseptic Manufacturing), Maneesha Altekar (AstraZeneca), Pritesh Patel (Pritesh Patel Consulting, LLC), Richard A. Lewis (GlaxoSmithKline)
1 Introduction
This is the second discussion paper written by the ISPE Process Validation (PV) Team on the topic of determining and justifying the number of initial process qualification batches (e.g., FDA Stage 2 [10], EudraLex Annex 15 process validation [13], EMA Guideline on process validation [14] [15], etc.) needed to demonstrate a high degree of assurance in the product manufacturing process and control strategy, and thereby support commercialization of the product. Most regulatory authorities agree on implementation of the lifecycle approach to PV. The science- and risk-based approach to determining the number of batches for the PV exercise is preferable, as opposed to defaulting to three batches. Differences of opinion and questions remain about the best way to translate the risk-assessment results into the actual number of batches, including acceptable statistical methodology options and the required level of statistical confidence, if applicable.
The Team’s original discussion paper, “Topic 1 – Stage 2 Process Validation: Determining and Justifying the Number of Process Performance Qualification Batches (Version 2),” by Bryder, et al., was issued in August 2012 and updated in 2014 [3]. It featured three approaches to assigning an appropriate number of PV batches to risk-assessment results in an attempt to address the expectation outlined in the FDA PV Guidance of 2011 [10], which stated that a high degree of assurance in the process must be obtained prior to release of commercial batches. This process assurance evaluation must also consider the result of process development (e.g., FDA Stage 1 [10], EudraLex Annex 15 planning for qualification and validation [13]) together with the results of the process-validation exercise (e.g., FDA Stage 2 [10], EudraLex Annex 15 process validation [13]). The three approaches presented in the first paper were useful in provoking thought and moving pharmaceutical industry understanding forward through various discussions, including discussions at scientific forums. As a result, several additional approaches have been developed, four of which are discussed in this paper.
2 Scope
The scope of this paper is limited to determining the number of initial PV, or Process Performance Qualification (PPQ) batches, using statistical tools identified in this paper. This paper supplements the “Topic 1 – Stage 2 Process Validation” Discussion Paper [3] which covers additional approaches in determination of number of PPQ batches.
This paper presents various statistical tools that can be used in determination of the number of PPQ batches. This paper will not attempt to explain basic statistics, but will present statistical tools with case studies or examples to demonstrate the use of the tool to arrive at the number of PV batches. Each statistical tool presented in this paper is intertwined with risk-based approaches, making it a hybrid approach. This paper assumes that a quality risk-management program is established, as described in ICH Q9, Quality Risk Management [6]. Alternatively, the risk-assessment methodology presented in the “Topic 1 – Stage 2 Process Validation” Discussion Paper may be used [3].
Strategies presented in this paper may be modified as needed by manufacturers to suit their established systems, processes, procedures, and/or policies. Suggestions and recommendations presented in this paper are intended to furnish a wide range of tools in determining the number of PV batches. They are not intended to represent an industry consensus, nor is it the paper’s intention to present one as more preferable than the others, as each approach may have its own limitations.
The statistical approaches, in combination with the risk ratings presented in this paper, can be applied to both Active Pharmaceutical Ingredients (APIs) and drug products. However, the risk assessment should consider the differences in the subject (i.e., API versus drug product).
3 Background
The “Topic 1 – Stage 2 Process Validation” Discussion Paper [3] explains FDA’s revised guidance on process validation and provides a risk-based framework for determining the number of batches for process validation. It discusses three risk-based approaches and outlines the pros and cons of each. Summaries of the three approaches are presented below:
Approach #1: Based on Rationales and Experience
This approach seems quite reasonable, but does not use statistics in the determination of the number of batches (see Table 1). It suggests using rationales based on historical information and process knowledge.
Approach #2: Target Process Confidence and Target Process Capability
This approach uses statistics, but it is immediately applicable only to normally distributed data and those processes that are in a state of statistical control. Although transformations from non-normal to normal distribution are sometimes possible, identifying a suitable transformation method can be a matter of trial and error. In addition, the user needs to draw conclusions consciously so as to apply them to the original process that generated non-normal data. This approach recommends implementing improvements to the process at the end of the PPQ batches if an expected process capability index (Cpk) range is not met.
Approach #3: Expected Coverage
This is a nonparametric approach based on order statistics—the minimum and maximum observed results are referred to as x(1) and x(n), respectively—and the properties of intervals based on these order statistics. In particular, the expected probability of a future observation being within the range defined by [x(1), x(n)] is (n – 1)/(n + 1). Users of this approach should select an appropriate level of coverage that is most suitable to their processes.
It is apparent that there is no single approach that fits all scenarios or processes, and that additional approaches may be useful.
Read more by downloading Determining the Number of Process Performance Qualification Batches Using Statistical Tools Discussion Paper.