On August 1, 2026, the harmonized revision of USP <788> became official. Most coverage has focused on the new title, Subvisible Particulate Matter in Injections, and the explicit inclusion of IM and SC routes. The sampling language deserves more attention.
The revised chapter states that the number of test samples must be adequate for a statistically sound assessment. It allows single units to be tested within a statistically sound sampling plan. Products with less than 25 mL may be tested with a justified 1–5 mL volume.
That flexibility is welcome, especially for high-value biologics. It also places a question in front of every lab: is the volume you analyzed enough to support the conclusion you are drawing? That question applies to compendial testing, and it applies just as much to the orthogonal flow imaging data USP <1788> recommends alongside it.
What the revised chapter says about sampling
The default procedure is unchanged in its structure. The analyst removes four portions of about 5 mL each, disregards the first result, and averages the remaining three. What changed is the room around it:
- Single units of large- and small-volume parenterals may now be tested within a statistically sound sampling plan.
- Products without sufficient volume, for example less than 25 mL, may be tested with 1–5 mL when the smaller volume is justified.
- Particle limits are unchanged, and light obscuration and microscopic particle count remain the compendial methods.
For more information, refer to the USP <788> compliance guide.
Counting statistics: why volume matters at ≥25 µm
Particle counts follow Poisson statistics. The relative uncertainty of a count is roughly one divided by the square root of the number of particles counted. Count 4 particles and the uncertainty is about ±50%. Count 100 and it falls to about ±10%.
This matters most at ≥25 µm, where counts in a well-controlled product are low. Consider a product running at about 2 particles/mL at ≥25 µm, comfortably within limits:
| Volume imaged | Expected particles ≥25 µm | Counting uncertainty (±1 SD) |
|---|---|---|
| 0.5 mL | 1 | ±100% |
| 1 mL | 2 | ±71% |
| 5 mL | 10 | ±32% |
| 15 mL (three counted <788> portions) | 30 | ±18% |
| 20 mL (all four <788> portions) | 40 | ±16% |
Below about ten counted particles, a real change in particle levels can be indistinguishable from counting noise. For stability trending, lot-to-lot comparability, and container-closure studies, the volume you image sets the floor on what you can detect.
Imaging the same aliquots as your light obscuration test
USP <1788> expects light obscuration and flow imaging to report different counts for the same sample. Explaining those differences is the job of a bridging study, and the cleanest bridge is to image exactly the portions the LO test measured.
Volume is the obstacle. At flow rates of a fraction of a milliliter per minute, imaging 20 mL takes from over an hour to several hours. In practice, many labs image a sub-milliliter subsample and compare it with a 15 mL LO result. The two numbers describe different volumes, with very different statistical confidence.
The Raptor 1788 images at flow rates up to 40 mL/min, so a full 20 mL <788> sample set passes through the flow cell in about 30 seconds. Every portion your LO test counted can be imaged, counted, and classified. You can see which particles LO counted: silicone oil droplets, protein aggregates, or foreign particulate.
The sample you measure should be the sample you prepared
A long run gives a sample time to change. Silicone oil droplets are buoyant and rise toward the top of the container or syringe. Dense particles such as glass, metal, and stopper fragments settle toward the bottom. Over time, droplets can also coalesce into larger droplets, shifting the size distribution.
Continuous mixing reduces these effects, but mixing can introduce bubbles or shear stress on protein formulations. Shortening the run is the simpler fix. When a full sample set is measured in seconds rather than hours, the particle population you image is the one you prepared.
Speed without sacrificing image quality
High flow is only useful if every image stays sharp. The Raptor 1788 optical system is designed to capture clear, measurable particle images at full flow rate, so shape-based classification works at the same speed as counting.
What this means for your program
- Justify volume with statistics. When you use a smaller test volume, document the expected counts and counting uncertainty, not only the product constraint.
- Image what you counted. Orthogonal data is most useful when it covers the same portions as the compendial test.
- Shorten the time between preparation and measurement. Less time means less flotation, settling, and coalescence.
Flow imaging complements light obscuration; it does not replace it for USP <788> release testing. The revision makes the case for orthogonal data that matches the compendial test in volume and confidence.
See it run. Watch the Raptor 1788 image a full USP <788> sample set in real time. Request a demo or explore the Raptor 1788.