Blend uniformity & stratified samplingBlend Uniformity Analysis (BUA) and Stratified In-Process Sampling

Blend uniformity analysis (BUA) is the potency test performed on the powder or granule blend itself, before it is compressed into tablets, filled into capsules, or otherwise divided into dosage units. Samples are drawn from defined locations in the blender, bin, or intermediate bulk container and assayed for the active pharmaceutical ingredient. The result answers one question: is the active evenly distributed through the mass, or has it concentrated in pockets?

Stratified in-process dosage-unit sampling is the complementary check performed after division — during or immediately following compression, encapsulation, or filling. Units are pulled at defined time points across the run (beginning, middle, end, and often at every discharge or hopper refill) and assayed individually. Where BUA characterizes the blend, stratified sampling characterizes what the blend actually became once it moved through the process — accounting for segregation that happens in transfer, in the feed frame, or during discharge, none of which BUA alone can see.

Neither test is optional evidence layered on top of a correct dispense record. A weigh-and-dispense step that applies the right potency-factor-corrected charge proves nothing about the finished batch if the blend segregates afterward. BUA and stratified sampling are the controls that close that gap, and 21 CFR 211.110(a) requires written procedures describing the in-process controls used to monitor and validate performance of manufacturing processes that may cause variability in the finished product's characteristics.

FDA issued a draft guidance in 1999 titled Powder Blends and Finished Dosage Units — Stratified In-Process Dosage Unit Sampling and Assessment, proposing specific bin-sampling locations, minimum numbers of samples, and %RSD acceptance thresholds for blend uniformity. Industry and FDA's own reviewers increasingly recognized that thief-based bin sampling was itself a major source of error — the sampling device disturbed the powder bed in ways that biased results independent of the blend's true uniformity. FDA withdrew the draft in 2003 rather than finalize a method whose own measurement error could not be separated from genuine non-uniformity.

No replacement guidance was finalized. Current expectations are risk-based and are inferred from 21 CFR 211.110, from ASTM E2709/E2810 (which FDA reviewers routinely cite as an acceptable statistical basis), and from the general pharmaceutical development framework in ICH Q8(R2). The practical consequence: firms must build their own justified sampling plan, document the rationale for sample size, locations, and acceptance criteria, and be able to defend that the plan is scientifically sound and that the sampling technique itself does not introduce bias.

A sample thief is a probe inserted into the powder bed to withdraw a small core sample from a specific depth and location. Two mechanisms make thief results unreliable indicators of true blend uniformity:

• Insertion disturbance — pushing the thief into the bed compacts, shears, or displaces powder ahead of and around the sampling cavity, changing the local composition before the sample is drawn.
• Particle-size-selective withdrawal — thief cavities preferentially admit fine particles over coarse ones (or vice versa depending on cavity geometry and powder flow properties), so the sample composition does not match the true local composition, especially when the active and excipients have different particle-size distributions.
• Segregation induced by the thief itself during withdrawal, as the sample travels up through the cavity past powder at other depths.

The net effect is that a failing BUA result is frequently a thief artefact rather than evidence of a genuinely non-uniform blend, and a passing result can mask real non-uniformity if the thief happens to average across a segregated pocket. This is precisely why stratified dosage-unit sampling — which assays what a patient would actually receive, with no intervening sampling device — is now weighted more heavily than bin thief sampling in a defensible programme.

USP General Chapter <905> Uniformity of Dosage Units is the compendial finished-product test — it requires either the Content Uniformity method (individual assay of a small number of units, statistically evaluated) or the Weight Variation method, depending on the product. BUA and stratified in-process sampling are not substitutes for <905>; they are the in-process evidence that predicts and supports a passing <905> result at release, and they let a deviation be caught during the run rather than discovered at final release testing.

The link is direct: ASTM E2709 is explicitly framed as demonstrating capability to comply with a content uniformity test, and ASTM E2810 does the same for the stratified case. A stratified sampling plan built to E2709/E2810 methodology, applied at appropriate points across the compression or encapsulation run, gives statistical assurance that the batch would pass <905> even before the formal release samples are pulled.

The plan must be risk-based and documented before the batch runs, not improvised afterward. Factors that increase the number of locations and time points required:

1. Define blend sampling locations to capture top, middle, bottom, and near-discharge positions across the blender or IBC — enough to detect radial and axial segregation.
2. Define stratified dosage-unit time points to capture beginning, multiple mid-run intervals, end of run, and every hopper refill or bin change.
3. Set the number of units per time point per ASTM E2810's statistical model, not an arbitrary round number.
4. Pre-specify the acceptance model (mean, %RSD, and any individual-unit limits) before the batch runs.
5. Document the rationale in the validation protocol or product-specific control strategy, tied to the risk factors above.

Percent relative standard deviation (%RSD = standard deviation ÷ mean × 100) is the standard summary statistic for blend and stratified uniformity data. There is no single FDA-mandated %RSD number; industry commonly targets %RSD ≤ 5.0% for blend samples on established solid-dose products, tightening for low-dose or high-potency actives, with individual results also checked against a %RSD-independent range (e.g., 90.0–110.0% of target) to catch a single extreme outlier that a %RSD calculation alone might not flag.

The acceptance criteria must be set in the validation protocol before data is generated and must be defensible against the specific risk profile of the product — a blanket 5% target applied without justification to a genuinely difficult low-dose formulation is itself a finding, because it implies the criterion was chosen for convenience rather than derived from a risk assessment.

• Report both %RSD and the individual result range — %RSD alone can mask a bimodal distribution where high and low results average to an acceptable spread.
• State the number of results the %RSD is computed from and the sampling locations/time points they came from, so a reviewer can trace back to the plan.
• Distinguish a %RSD calculated across locations within one time point (spatial uniformity) from one calculated across time points (temporal/run uniformity) — they answer different questions and both matter.

Segregation is rarely introduced during blending itself if blend time and speed are adequately validated; it is far more commonly introduced downstream, during transfer and discharge, through well-understood powder-mechanics mechanisms:

• Percolation segregation — fine particles migrate downward through the interstitial spaces between coarse particles as the bulk is vibrated or moved, concentrating fines at the bottom of a bin.
• Trajectory (sifting) segregation — as powder is poured or free-falls from a discharge chute, particles of different size or density follow different flight paths, landing in different zones of the receiving vessel.
• Fluidization segregation — fine, low-density particles remain airborne longer than coarse particles when powder is dropped through air, settling last and forming a fines-rich top layer.
• First-in/last-out and rathole flow — non-uniform bin discharge patterns mean the material dispensed first is not compositionally identical to the material dispensed last, even from a blend that was uniform when blending finished.

Because these mechanisms act after blending and before compression, stratified dosage-unit sampling that brackets the entire discharge sequence — not just a single bin-thief sample taken once — is the control most directly aimed at catching them.

Process Analytical Technology, as framed in FDA's 2004 PAT guidance, offers a fundamentally different control strategy: continuous, in-line or at-line near-infrared (NIR) or Raman spectroscopic monitoring of the blend as it mixes, generating a real-time uniformity endpoint instead of relying on discrete, invasive thief samples pulled at fixed intervals. A properly validated NIR blend-monitoring method eliminates thief-sampling bias entirely because no physical sample is withdrawn, and it can detect the moment a blend reaches the statistically stable uniformity endpoint rather than assuming a fixed blend time is always sufficient.

Implementing PAT/NIR requires its own validation burden — chemometric model development, calibration against reference HPLC assay, and demonstration that the spectroscopic endpoint correlates with true content uniformity across the design space (ICH Q8(R2)) — so it is not a shortcut, but for high-value, high-volume, or historically difficult-to-blend products it removes the single largest source of measurement error in the classic programme.

1. No documented rationale for sample locations, sample size, or acceptance criteria — the plan exists but cannot be traced to a risk assessment or statistical basis.
2. Sampling technique not standardized or not verified — different operators use different thief insertion depths or techniques without documented equivalence.
3. %RSD acceptance criteria set loosely or inconsistently across products with no justification for the difference.
4. Reduced sampling adopted without a documented history of consistent full-plan results to support the reduction.
5. Stratified dosage-unit sampling time points that do not bracket the full run — missing the beginning, a hopper refill, or the end, exactly where segregation is most likely.
6. Investigation into an OOS/OOT blend or stratified result that concludes 'sampling error' without objective evidence, rather than treating it as a genuine deviation per 21 CFR 211.192.
7. No trending of %RSD batch over batch — a slow upward drift that predicts a future failure goes unnoticed because each batch is assessed only against its own pass/fail limit.