Trends Identification and Analysis
During the production and quality control of active pharmaceutical ingredients and medicinal products large amounts of data are generated which can be used to draw conclusions on the quality of the corresponding product. These conclusions serve as basis for decision- making for the further approach in the production process or for or against the release of the product. Therefore, it is crucial to be able to reliably assess the quality of the data and to reliably identify trends which indicate a quality defect of the product or deviations in the production process.
The ECA Foundation's Analytical Quality Control Group (AQCG) has targeted this issue with the "Laboratory Data Management Guidance - Out of Expectation (OOE) and Qut of Trend (OOT) Results" and has formulated standards according to which trends can be identified and analysed. This document describes statistical tools for the detection of trends and deviations which might possibly be hidden by noise and are consequently not obvious immediately.
The first version of this ECA Guidance was published in December 2015. The revised version V 1.1 was published on the ECA website in November 2016 and is available in the members' area on the ECA Academy website* as well as on the website of the Analytical Quality Control Group (AQCG)*.
Scope and Application of the Guidance
This guidance applies to all chemical/physicochemical-based laboratory testing resulting in continuous, usually normally distributed data (for example, determination of content or impurities) or discrete data (for example, identity tests or cosmetic defects, discolorations etc.). The statistical methods described in this document are applicable to two standard situations which occur regularly in the area of quality control:
- A trend is not expected since the analytical process or the production process is validated and should deliver consistent results of consistent quality.
- A trend is expected, for example, in a stability study but the aim is to identify trends which might be hidden underneath.
The Guidance complements the AQCG's operating procedure "STANDARD OPERATING PROCEDURE Laboratory Data Management; Out of Specification (OOS) Results" from the year 2013 and should be used in connection with it.
The following chapters are the core of the OOE/OOT-Guidance:
1. The Control Charting Concept
Whether a process is stable and capable can be assessed by means of appropriate decision criteria. In this chapter two of these criteria are being described:
- WECO Rules: They cover four situations, each with a special distribution of data points which indicate a trend ("WECO" derives from "Statistical Quality Control Handbook" of the Western Electric Company from the year 1956).
- Nelson Rules: They describe eight special data distributions indicating a trend (Lloyd S. Nelson, "Technical Aids", Journal of Quality Technology 16(4), 238-239, October 1984).
Heidelberg, Germany21/22 September 2022
HPLC Data Integrity
The four WECO Rules are recommended. In most cases they are sufficient for the detection of a trend and, therefore, an instable process.
2. Detecting and Managing OOE Results
First, the following definitions are given:
- OOS: a test result that does not comply with the pre-determined acceptance criteria laid down in the marketing authorisation dossier.
- OOT: a result that is outside the expected variability/uncertainty of measurement of the analytical procedure or does not follow the expected trend (for instance of a stability study); it can be calculated or deduced from a data base which is comprehensive enough.
- OOE: a result that is outside the expected variability/uncertainty of measurement of the analytical procedure but cannot be deduced statistically.
OOE results can occur in the following two scenarios:
- Unexpected Variation in Replicate Determinations or Test Series. If a deviation does not lie within the usual range of variation Δ of ≤ 2.0% (for instance 2.2%) this OOE result is a deviation that requires an investigation into the cause.
- Unexpected Results in a Single Test or a Small Set of Tests Due to the small database, it cannot be decided whether the result or the results are within the expected variability. In this case validation data should be used. The combined standard uncertainty of measurement as performance indicator can be used as basis for decision-making (OOE - yes or no). The Guidance describes a procedure for the specification of such standard uncertainties of measurement.
3. Trend Analysis for Statistical Process Control
Control of continuous data
This Guidance recommends using Shewhart control charts for this purpose. Such a control chart consists in a centre line representing the mean (average) of the values collected during a reference period and in an upper and a lower control limit (upper control limit: UCL, lower control limit: LCL), providing a range of what is still acceptable for a result. If a root cause analysis has to be carried out the CuSum chart (cumulative sum) is the appropriate method for the analysis of historical data. Here, the difference of successive values is plotted against time (see chapter 5).
Control of discrete data
Discrete data can be controlled by means of p-charts or np-charts. Such as the control charts for continuous data these charts also consist in a central line and an upper and a lower control limit. In the case of an inspection of incoming materials for defects this method determines for each sampling the fraction of sample elements having a quality defect.
4. Trend Analysis for Stability Testing
The purpose of trend analysis for stability data is to detect if:
- a batch is out-of-trend with respect to historical batches (indicates an out-of-trend situation),
- one or more noticeable observations within a batch indicate an out-of-trend situation.
In the ECA Guidance two statistical procedures are outlined: the linear regression model and the random coefficients regression model. The linear regression model is simpler as concerns mathematics and provides information on the stability values with respect to time. It is suitable for the prediction of stability data. With the more complex random coefficients regression model multivariate parameters are taken into account and plotted multidimensionally. In both cases the slopes of the stability graphs (content with respect to time) are compared between batch under study and historical batch and the difference determined. If this difference exceeds a pre-defined limit the batch concerned is out-of-trend.
5. Trend Analysis for Root Cause Analysis
Sudden and/or large changes in an analytical measurement sequence can be detected easily by means of the Shewhart chart, but alternative methods have to be used for the detection of small but persistent deviations. Concerning this, the ECA Guidance recommends the post mortem CuSum analysis as a simple but effective method. This technique adds up the successive differences from a bench mark and plots them over the time axis. If there is no deviation there is a straight line parallel to the X axis. Are there positive or negative slopes on sections of this lines this indicates a corresponding deviation at a certain point of time. Using this technique, it is very easy to identify when (not why) an anomaly occurred. The reason for this deviation has to be determined by means of a subsequent root cause analysis.
27 September 2022
Identification and Management of OOT and OOS Results - Live Online Training
All things considered, with its description of the different techniques for the detection of OOE and OOT situations and the practical examples in the appendices, this ECA Guidance offers an excellent tool for the persons responsible in the area of quality control.
As already mentioned above, this Guidance should be used in close connection with the ECA-SOP on Out of Specification (OOS) Results.
About the Author
Dr Gerhard Becker joined CONCEPT HEIDELBERG in 2002 and organises and conducts courses and conferences on behalf of the ECA Academy in analytical and compliance topics.