Methods for Identifying Out-of-Trend Results in Ongoing Stability Data

The authors discuss three methods for identification of out-of-trend (OOT) results and further compare the z-score method and the tolerance interval in OOT analysis for stability studies.
Jun 01, 2013

It is important to distinguish between out-of- specification (OOS) and out-of-trend (OOT) results in stability studies. The authors discuss three methods for identification of OOT results—the regression-control-chart method, the by-time-point method, and the slope-control-chart method—and further compare the z-score method and the tolerance interval in OOT analysis. The results highlight the need for issuing a regulatory confirmed guideline for identification of OOT results for ongoing stability data.

The two terms out-of-trend (OOT) and out-of-specification (OOS) results are in many cases confused by pharmaceutical companies and regulatory agencies. OOT results are defined as a stability result that does not follow the expected trend, either in comparison with other stability batches or with respect to previous results collected during a stability study (1). OOT results are not necessarily OOS, but they do not look like a typical data point. Although OOT results are a serious problem, the scientific literature and regulatory guidelines do not fully address this issue.

According to FDA's Guidance for Industry: Investigating Out-Of-Specification (OOS) Test Results for Pharmaceutical Production (2), OOT results should be limited and scientifically justified. The guideline, however, does not define the process for identification of OOT results in stability data. The CMC Statistics and Stability Expert Teams of the Pharmaceutical Research and Manufacturers of America made an attempt to address this problem by suggesting several statistical methods for the identification of OOT results (3). The proposed statistical methods were redesigned and analyzed for the purposes of this study.

The aim of this study was to make a statistical confirmation of the statistical methods, which will prove their functionality in identification of OOT results in ongoing stability data within a batch or data among batches. In addition, a comparison was made between the z-score method and the tolerance interval (TI) in terms of defining the limits for identification of the present OOT result.

Materials and methods

For the purpose of this study, data from ongoing stability studies of a final drug product with a shelf life of 36 months were used. The ongoing studies were conducted on 10 batches of Product X. Product X is manufactured in a tablet dosage form and consisted of one active substance with defined strength of 10 mg and packaged in a primary aluminium–polyvinyl chloride (Al–PVC) blister and a secondary package. The ongoing studies were conducted for 36 months in stability chambers at a constant temperature of 25 °C ± 2 °C and relative humidity of 60% ± 5% in accordance with the ICH guideline Q1A(R2) (4).

The reported data are single data results for the assay attribute, calculated as a percentage of the declared active substance concentration. The assay attribute was analyzed in accordance to the validated internal method of the manufacturer at the time points of 0, 3, 6, 9, 12, 18, 24, and 36 months in all of the tested batches.

The first nine batches were used as historical data for the purposes of the by-time-point method and the slope- control-chart-method in addition to which the tenth batch was compared and analyzed. The historical data were used to define the limits for identification of present OOT results in the tenth batch; the regression-control-chart-method analysis was conducted only on the tenth batch.

In addition, simulated data also were implemented. The simulated data were comprised of eight test time points for each of the 10 simulated batches. Unlike the experiment, in the simulation, the 10 batches were tested using the regression-control-chart method. In the by-time-point method and the slope-control-chart method, however, the historical data of the real batches were used to individually analyze the 10 randomly generated batches.