This is a ready-to-use worksheet for the ICH Q1E shelf-life calculation: fitting the regression, testing whether batches can be pooled, computing the one-sided 95 percent confidence limit, and checking the result against the extrapolation cap. Replace every <<FILL: ...>> placeholder with your own data. A fully worked specimen with real numbers follows the blank template so you can check your own arithmetic against it. This worksheet documents standard practice; have a qualified statistician review the approach for your specific data, and confirm the current wording of ICH Q1E before you rely on any numeric threshold quoted here.
Worksheet header
| Field | Entry |
|---|---|
| Worksheet number | <<FILL: WS-ID>> |
| Product / attribute evaluated | <<FILL>> |
| Storage condition | <<FILL: e.g. 25 C / 60% RH>> |
| Specification limit(s) used | <<FILL: e.g. lower limit 90.0% label claim>> |
| Batches included | <<FILL: batch numbers, n>> |
| Prepared by / date | <<FILL>> |
| Reviewed by (statistician / QA) / date | <<FILL>> |
1. Assemble the data
| Batch | Time (months) | Result (% of label, or attribute units) |
|---|---|---|
<<FILL>> | <<FILL>> | <<FILL>> |
Confirm before proceeding: the attribute changes over time (a regression approach is only meaningful for a trending attribute; an attribute that shows no trend and stays within a tight range across all time points may instead be evaluated by pooled mean and variability against the specification, per Q1E; state which path applies: <<FILL>>).
2. Fit the regression for each batch individually
| Batch | Model form | Intercept | Slope | Residual SD (s) | R-squared |
|---|---|---|---|---|---|
<<FILL>> | Linear / other justified transform | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
3. Test batch poolability
Test whether the batches can be combined into a single regression line. Q1E describes testing the equality of slopes first, then, if slopes can be pooled, the equality of intercepts, commonly using a significance level around 0.25, chosen deliberately higher than the usual 0.05 so the test is more willing to flag real differences and less willing to let genuinely different batches be pooled.
| Test | Statistic | p-value (or F-value vs. critical) | Significance level used | Pass (poolable)? |
|---|---|---|---|---|
| Equality of slopes | <<FILL>> | <<FILL>> | <<FILL: e.g. 0.25>> | <<FILL>> |
| Equality of intercepts (only if slopes pooled) | <<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
If slopes are not poolable: do not force a single line. Evaluate the batch with the least favorable slope (fastest degradation toward the limit) on its own, and let that batch’s confidence limit set the claim, since a pooled line would understate the risk from the worst batch. Record: <<FILL: which batch governs and why>>.
If poolable: proceed to the combined regression in section 4.
4. Combined (pooled) regression, if applicable
| Field | Value |
|---|---|
| Pooled intercept | <<FILL>> |
| Pooled slope | <<FILL>> |
| Pooled residual SD (s) | <<FILL>> |
| Total data points (N) across all pooled batches | <<FILL>> |
| Sum of squared deviations of time values (Sxx) | <<FILL>> |
5. Compute the one-sided 95 percent confidence limit
The predicted value at time t is intercept + slope multiplied by t. The one-sided 95 percent confidence limit on the regression line at time t is approximately:
predicted value +/- t(df, 0.95) x s x square root of (1/N + (t minus mean of t)^2 / Sxx)
using the lower bound if the attribute is declining toward a lower specification limit, or the upper bound if it is rising toward an upper limit. Use a validated statistical tool or a qualified statistician for the exact calculation; this worksheet records the inputs and the result, not a substitute derivation.
| Time (months) | Predicted value | One-sided 95% confidence limit | Distance from specification limit |
|---|---|---|---|
<<FILL>> | <<FILL>> | <<FILL>> | <<FILL>> |
Shelf life from the confidence limit: the time at which the confidence limit line, not the predicted mean line, intersects the specification limit. Solve for t or read it from a plotted curve. Result: <<FILL: months>>.
6. Apply the extrapolation check
| Field | Value |
|---|---|
| Actual long-term data coverage (longest time point with real data, all batches) | <<FILL: months>> |
| Proposed shelf life from section 5 | <<FILL: months>> |
| Cap 1: no more than approximately double the long-term coverage | <<FILL: months>> |
| Cap 2: no more than approximately 12 months beyond the long-term coverage | <<FILL: months>> |
| Binding cap (the lower of the two) | <<FILL: months>> |
| Accelerated data show significant change within the extrapolation window? | Yes / No |
| If Yes: extrapolation is not supported by accelerated data; intermediate condition data (if available) governs instead | <<FILL>> |
| Final supported shelf life (statistical result, capped, and consistent with accelerated/intermediate findings) | <<FILL: months>> |
| Continuation commitment required? | Yes, ongoing testing continues per the stability protocol to confirm the extrapolation |
7. Acceptance criteria
- The regression model form is justified for the attribute’s actual behavior (a linear fit is not forced onto a curved trend without a stated transform).
- Poolability was tested, not assumed; if batches were not poolable, the least favorable batch governs.
- The shelf life is set by the one-sided 95 percent confidence limit, never by the mean.
- Any extrapolation stays within both caps and is consistent with the accelerated and, where applicable, intermediate data.
- A continuation commitment is documented whenever extrapolation is used.
8. References
ICH Q1E, Evaluation of Stability Data. ICH Q1A(R2), Stability Testing of New Drug Substances and Products. 21 CFR 211.166 and 211.180(e) (stability testing and trend evaluation).
Confirm the current version and clause numbers before issue.
9. Revision history
| Version | Date | Author | Summary of change |
|---|---|---|---|
<<FILL: 1.0>> | <<FILL: date>> | <<FILL>> | Initial worksheet. |
10. Approvals
| Role | Name | Signature | Date |
|---|---|---|---|
| Prepared by | <<FILL>> | ||
| Statistician / independent reviewer | <<FILL>> | ||
| Quality Assurance | <<FILL>> |
Filled specimen
Three batches of an immediate-release tablet, assay in percent of label claim, lower specification limit 90.0 percent, long-term data through 18 months.
Section 1, data.
| Batch | Months | Assay |
|---|---|---|
| A | 0, 6, 12, 18 | 100.2, 99.1, 98.0, 97.0 |
| B | 0, 6, 12, 18 | 99.8, 98.6, 97.1, 95.9 |
| C | 0, 6, 12, 18 | 100.5, 99.3, 98.2, 97.1 |
Section 2, per-batch regression (linear). Batch A: intercept 100.15, slope -0.175%/month. Batch B: intercept 99.75, slope -0.208%/month. Batch C: intercept 100.45, slope -0.185%/month. Batches A and C degrade at close to the same rate; batch B degrades somewhat faster.
Section 3, poolability. Equality-of-slopes test at a 0.25 significance level: the three slopes are close enough not to reject poolability (batch B’s steeper slope is within the noise the test tolerates at this significance level). Result: poolable. Equality of intercepts likewise passes. Proceed to the pooled line.
Section 4, pooled regression. Pooled intercept 100.1, pooled slope -0.189%/month, pooled residual SD (s) 0.35%, N = 12 points, Sxx computed from the four time values repeated across three batches.
Section 5, confidence limit. Predicted assay at 18 months: 100.1 - (0.189 x 18) = 96.7%. Using a t-multiplier appropriate for the degrees of freedom and a standard error reflecting the spread around the line, the one-sided 95 percent lower confidence limit at 18 months works out to approximately 95.3%, comfortably above the 90.0% limit. Projecting the same confidence-limit line forward, it crosses 90.0% at approximately 50 months, well past the actual long-term data collected.
Section 6, extrapolation check. Long-term coverage on file: 18 months. Cap 1 (double coverage): 36 months. Cap 2 (12 months beyond coverage): 30 months. Binding cap: 30 months (the lower of the two). The regression math would support a shelf life out to roughly 50 months, but the extrapolation cap holds the claim to no more than 30 months regardless of what the line predicts, and in practice a sponsor would commonly propose 24 months here as a round, defensible claim comfortably inside the 30-month cap, with a written commitment to continue testing toward 36 months. Accelerated data at 40 C / 75% RH for this example showed no significant change through 6 months, supporting the use of extrapolation at all.
Conclusion (specimen): the statistically supported shelf life claim is 24 months, set by the confidence-limit method and held inside the Q1E extrapolation cap, with ongoing testing committed through 36 months to confirm the projection as real data accumulates.
Common inspection findings this worksheet prevents
- A shelf life read off the mean line instead of the confidence limit, overstating what the data actually support.
- Batches pooled without a poolability test, hiding a faster-degrading batch inside an average that looks fine.
- Extrapolation stretched past both Q1E caps because “the regression line still passes” at the proposed time point.
- No continuation commitment documented when extrapolation was used to support the initial claim.
- The wrong bound used (upper instead of lower, or vice versa) for an attribute moving toward its specification limit.
How to adapt this worksheet
- Replace the input data with your own batch results and let a validated statistical tool or your statistician compute the exact regression and confidence-limit values; the specimen above uses rounded illustrative arithmetic to show the logic, not to replace a validated calculation.
- If your attribute does not trend meaningfully over time and stays inside a tight range, use the alternative pooled mean and variability evaluation Q1E describes instead of forcing a regression.
- For biologics and other attributes with high batch-to-batch variability, weight the poolability and confidence-limit interpretation conservatively; see the biologics stability discussion of why real-time data is trusted more heavily than extrapolation for these products.
- Confirm the current wording and any numeric detail of ICH Q1E before issue; this worksheet describes the method in original wording and does not reproduce guideline text.