Majority of the pharmaceutical plants have SOPs to define yield limits for each product but many do not have any kind of yield limits & they often have a figure which was copied from another product, or set years ago by someone who has left, or rounded to something that looks tidy. A limit that is not based on the process does two kinds of harm. If it is too loose, real losses pass without comment. If it is too tight, good batches trigger deviations and the team learns to treat them as routine.
The better approach is to let the process tell you what normal looks like.
Why defined limits matter
Good manufacturing practice expects yields to be calculated and compared with expectations. In the United States, 21 CFR 211.103 requires the actual yield and the percentage of theoretical yield to be determined at the end of each appropriate phase of manufacturing, processing, packaging or holding. A limit gives that comparison something to be measured against, and it tells operators/professionals when a result needs to be investigated. The same idea runs through the reconciliation of every stage.
Percentage Yield Calculator, which compares actual with theoretical yield and shows the loss.
Step 1: define the stage and the basis
Set a separate limit for every stage that has its own yield, such as granulation, compression, coating, encapsulation, blistering and packing. Their losses come from different causes, so one limit for all of them cannot be right. For each stage, write down how the yield is calculated, weight or count, dry or wet basis, and which quantities are included. If the basis changes, the data from before the change is not comparable. For help with the formula, see the Percentage Yield Calculator.
Step 2: collect enough comparable batches
Pull the final yield for each stage from recent batches of the same product, strength and process, made on the same equipment train and at the same scale. 20 to 30 batches is a common target, because the spread of the results becomes stable at that size. With fewer than that, the limits are provisional and should be reviewed as more batches are made.
Here are the compression yields of 24 consecutive batches of one product, as an example:
| Batch | Yield % | Batch | Yield % | Batch | Yield % | Batch | Yield % |
|---|---|---|---|---|---|---|---|
| 1 | 98.2 | 7 | 98.1 | 13 | 97.8 | 19 | 98.1 |
| 2 | 97.6 | 8 | 98.2 | 14 | 98.6 | 20 | 97.9 |
| 3 | 98.5 | 9 | 97.7 | 15 | 97.6 | 21 | 98.4 |
| 4 | 97.9 | 10 | 98.5 | 16 | 98.1 | 22 | 97.6 |
| 5 | 98.8 | 11 | 97.4 | 17 | 98.9 | 23 | 98.0 |
| 6 | 97.2 | 12 | 98.2 | 18 | 97.0 | 24 | 98.3 |
Step 3: handle the unusual batches honestly
Look at every batch that sits far from the others and find out why. If a batch had a known and documented cause, such as a press breakdown or a mix-up that was corrected, you can exclude it from the calculation, but record that you did and why. Never remove a batch only because it is inconvenient. Removing low results quietly makes the limits tighter than the process really is, and the next normal batch will fail them.
Step 4: look at the shape of the data
Plot the yields in batch order and as a histogram. You are looking for three things
- A trend up or down over time
- Clear step change after a change in equipment or material
- Lopsided spread.
If the process drifts or has changed, use only the recent stable period. If the data is strongly skewed, so that the mean and standard deviation describe it poorly, percentiles may describe it better than the mean and standard deviation.
Step 5: calculate the mean and the standard deviation
For stable, roughly bell-shaped data the usual method is simple
- Mean = sum of yields ÷ number of batches.
- Standard deviation (SD) = how far the batches typically sit from the mean. A spreadsheet function such as STDEV does the calculation.
For the 24 batches above:
| Statistic | Value |
|---|---|
| Number of batches | 24 |
| Mean yield | 98.03% |
| Standard deviation | 0.48 points |
| Mean − 2 SD (alert) | 97.06% |
| Mean − 3 SD (action) | 96.58% |
| Lowest batch | 97.0% |
| Highest batch | 98.9% |
Step 6: set alert and action limits
A common method is to use two levels. An alert limit at the mean minus 2 SD flags a result that is unusual and worth a look. An action limit at the mean minus 3 SD marks a result that is unlikely to come from the normal process and that must be investigated. In the example the alert limit is 97.06% and the action limit is 96.58%. A lower limit matters most for yield, because a low result is a loss. In this data set one batch, at 97.0%, sits just under the alert limit, which is what you expect from a stable process now and then, and it does not come near the action limit.
The upper side is different. A yield cannot exceed 100% of the theoretical, so the upper limit is 100.0% in practice, and a result above it should be treated as an error to explain. The mean plus 3 SD in the example is 99.47%, which is below 100, but it still plays no role, the upper limit stays at 100.0% because a yield cannot exceed it. See Yield Above 100% for the causes.
Step 7: round, cross-check and write it down
- Round to a sensible figure, for example to the nearest 0.5% or 1%, in the direction that stays honest about the data.
- Compare the result with the lowest batch seen. If the action limit is far below anything the process has produced, it may be too loose to catch a real problem.
- Check it against the validated process and against your product registration, if either states an expected yield range.
- Put the limit into the master batch record through change control, with the data and the reasoning attached.
Step 8: review the limits regularly
Limits are not permanent. Recalculate them at a set interval, for example each year, and after any change to equipment, tooling, material source or scale. Review the trend between times, even when every batch passes, as a slow fall in yield is often the first sign of wear or drift. The stages where tablets and material are lost are covered in Yield Loss Points Across Stages.
Common mistakes
- Setting limits from only five or six batches.
- Mixing different products, strengths, scales or equipment in one data set.
- Deleting poor batches without a documented reason.
- Copying the limit from another product because the process looks similar.
- Using a mean plus or minus 3 SD on data that is clearly skewed or drifting.
- Setting limits so tight that normal batches fail, or so loose that nothing is ever flagged.
- Never reviewing the limits after the first time.
Frequently asked questions
How many batches do I need to set a yield limit?
Twenty to thirty is a common target. Fewer than that can be used for provisional limits, as long as they are reviewed when more data is available.
Should I use 2 SD or 3 SD?
Many plants use both: 2 SD as an alert and 3 SD as an action limit. The right choice is the one described in your procedures and justified by your data.
Can I set a yield limit by copying a similar product?
It is a starting point at best. The limit should be confirmed against the real data of the product and the equipment it is made on.
What if my data is not bell-shaped?
Then the mean and standard deviation may describe it poorly. Look at the lower percentiles of the data, or ask a statistician to help, and document the method used.