Shelf-Life Estimation from Stability Data (ICH Q1E)

About the Shelf-Life Estimation from Stability Data (ICH Q1E)

ICH Q1E sets out how a shelf life or retest period is read from stability data: fit each batch’s results against time, decide by analysis of covariance whether the batches can be combined, and take the earliest time at which the 95 % confidence limit of the mean meets the acceptance criterion. Choose your results file or paste the table — one result per row with the batch, the time in months and the result, or one row per batch with a column per time point — set the acceptance criteria and how the attribute changes, and the page works through the steps of Q1E’s Appendix B: the regression of each batch, the poolability tests of slopes and then intercepts at the 0.25 level, the most reduced model, the confidence-limit crossing for every batch, and the extrapolation that Q1E’s decision tree allows for your storage condition and data.

The report names the model it used, the batch or the rule that sets the shelf life, and whether the data support the shelf life you propose, with a chart of the results, the fitted lines and the limits. It applies the method you choose; your QA approves the result. The free preview runs the whole evaluation on your own data and shows the method, the limits that apply and the verdict; the computed figures — slopes, p-values, crossing times and the supported shelf life — the chart’s fitted lines, the PDF report and the workbook with the working need an Ultimate pass.

Common uses

• Proposing a shelf life or retest period for a registration dossier from the primary stability batches.
• Checking that commitment or annual batches still support the shelf life granted to the product, with the same method as the original evaluation.
• Re-evaluating after a change of site, pack or process with the new batches’ data.
• Teaching the Q1E evaluation step by step on real-looking data.

How the shelf life is worked out

• Each batch: result = a + b × time by least squares (zero order, which Q1E notes holds for assay and degradation products in general); choose the log scale for first-order change.
• Poolability (ICH Q1E B.2.2.1): slopes are tested first, F = [(SSE common slope − SSE separate lines) / (k − 1)] / MSE separate lines; if they do not differ, intercepts are tested with the common slope. Each test uses the 0.25 level, because a typical stability study has little power. The most reduced model is kept: separate lines (each batch’s own slope and intercept with the pooled mean square error of all batches), a common slope with separate intercepts, or one line.
• The limit: the lower one-sided 95 % confidence limit of the mean for a falling attribute, the upper for a rising one, two-sided 95 % limits when it can move either way (Q1E 2.6): ŷ(t) ∓ t × s × √(1/n + (t − t̄)² / Sxx).
• The shelf life: the earliest time the limit meets the acceptance criterion, the shortest over the batches when they are not combined, no longer than the extrapolation rule allows, rounded down to whole months (or multiples of 3 or 6, as you choose).

How far beyond the data: the rules of Q1E Appendix A

X is the period covered by long-term data (the page takes the last time point every batch has reached, or the value you enter); Y is the longest shelf life that may be proposed.

• Room temperature, no significant change at the accelerated condition: little or no change and variability, Y up to 2X but not more than X + 12 months; change or variability with a statistical analysis and relevant supporting data, the same.
• Room temperature, significant change at accelerated, none at intermediate: with the analysis and supporting data, Y up to 1.5X but not more than X + 6 months.
• Significant change at the intermediate condition: no extrapolation.
• Refrigerator: up to 1.5X but not more than X + 6 months when there is no significant change at accelerated; no extrapolation when there is.
• Freezer, or below −20 °C: based on the long-term data, no extrapolation.

Where Q1E ties extrapolation to relevant supporting data and you answer that there are none, the page applies no extrapolation. The page performs the statistical analysis, so it applies the limits Q1E gives for long-term data that are amenable to one. Where they are not (for example a change a straight line does not describe), Q1E allows less, again only with relevant supporting data: up to 1.5X but not more than X + 6 months at room temperature without significant change at the accelerated condition, and up to X + 3 months after significant change at the accelerated condition alone, or for a refrigerated product whose data change or vary. Significant change is defined in ICH Q1A(R2). A shelf life granted by extrapolation is verified with further long-term data as they come in.

Preparing the table

• One row per result: Batch (or Lot), Condition, Time (months), Result; a column named Attribute (or Test) holds several tests, or give each test its own column (“Assay (%)”, “Impurity A (%)”) and the unit in brackets is read.
• One row per batch: Batch, then one column per time point (0, 3, 6, 9, 12 or Initial, 3M, 6M, T9).
• One row per time point: Months, then one column per batch.
• Times may be written 0, Initial, T3, 3M, 6 months or M9; weeks, days and years are converted to months. Results may use a decimal point or comma.
• Cells that are not numbers (“< 0.05”, “ND”) are left out and listed in the report: enter the value your procedure uses for them if they should count.

What the report holds

The verdict and the model, a table per batch (results, slope, intercept, R², residual SD), the poolability tests with their F and p values, the three models compared, the estimate of every line used, every batch on its own line as Q1E B.2.1 describes, the extrapolation rule with X and Y, the chart, the method, the limits that apply, the cells left out and lines for preparation and QA review. The PDF carries the chart as vectors; the workbook has a sheet per table and the chart’s data, with numbers as numbers.

Speed and the device it needs

Measured on a laptop: 20,000 results (1,000 batches) are read and evaluated in under half a second with about 25 MB of memory; a usual study of three to six batches takes a few milliseconds, so any recent phone or computer is enough. The chart draws up to 12 batches (the batch that sets the estimate always among them); the tables list every batch.

Limitations

• A statistical aid for the Q1E evaluation of one quantitative attribute at a time: the shelf life you propose should not exceed the one supported by any single attribute, and qualitative and microbiological attributes are not amenable to this analysis (Q1E 2.1).
• It applies the method you choose; your QA approves the result. The extrapolation rule depends on your answers about significant change and supporting data, which the page cannot check.
• Linear regression on the time scale or the log scale only; a non-linear change pattern needs another model, and the residuals deserve a look before the estimate is used.
• One-factor (batch) studies: for several strengths or packs, evaluate each combination or use a multi-factor analysis as Q1E B.3 describes.
• Reduced designs (bracketing and matrixing, ICH Q1D) are evaluated like full designs, as Q1D 2.5 says; the missing time points simply are not in the table.

Privacy

On this device; its files download once

The stability results are read in your browser and never uploaded or stored; only your choices on the page stay until you reload it.

How to check that nothing is uploaded