qsteadgroup@yahoo.com (619) 718-1853

Advanced Case Study

Recurring Low-Purity OOS With Multiple Causal Pathways

The same failure does not necessarily mean the same root cause. An advanced RCA case for experienced QA, QC, and CAPA professionals.

Based on real-world events. To protect confidentiality, the names of companies, products, people, and places, along with any proprietary or confidential details, have been changed or removed. Some details have been combined or simplified for teaching, and any data shown are illustrative. Any resemblance to a specific company, product, or person is coincidental.

This case is designed as an advanced RCA case because the investigation can reach a technically valid root cause, implement an apparently successful CAPA, and still encounter the same observable failure later for a completely different reason.

The central lesson is that the same failure phenotype does not necessarily mean the same root cause.

At a glance

ProductRecombinant protein therapeutic
TestValidated QC HPLC purity method
SpecificationPurity ≥ 95.0% (historically 97–99%)
Pathway AIntermittent buffer-valve failure on the purification skid → abnormal separation → low purity
Pathway BExcipient characteristic combined with extended intermediate hold → protein degradation → low purity
The hard partAttributing historical OOS results when the evidence to tell the pathways apart was never collected
Who it’s forExperienced investigators, CAPA owners, QA reviewers, and trainers

How to use this case

Read it in stages. At each Tool decision box, pause and ask what you would do next and why. The Advanced practice boxes go beyond the basic scenario: evidence grading, competing hypotheses, failure fingerprints, cognitive traps, and how to write conclusions that hold up to regulatory scrutiny.

1. The initial problem

A biotech company manufactures a recombinant protein therapeutic. Manufacturing uses a purification chromatography system (purification skid), while QC uses a validated HPLC method to measure final product purity.

The specification is purity ≥ 95.0%. Historically, batches typically produce approximately 97–99% purity.

Then sporadic OOS results begin appearing:

94.6% → 94.1% → 93.8% → 94.7%

Most batches continue to pass. There is no immediately obvious pattern.

Advanced practice · Before choosing a tool

  • Confirm the laboratory result first. Complete the Phase I laboratory investigation (FDA OOS guidance) to show the HPLC result is valid before investigating manufacturing.
  • Characterize the impurity, not just the number. “Purity < 95%” is a symptom. Is the extra impurity an aggregate, a fragment, a charge variant, or a process-related impurity? Different mechanisms leave different fingerprints. This one step becomes decisive later.
  • Plot all batches, not just the failures. A run chart of purity with the 97–99% historical band shows whether the passing batches are drifting too.

2. Start simple

The investigation does not immediately require an elaborate FTA or DOE. A reasonable approach is:

Problem definition → data review → brainstorming → 5 Whys

The investigators begin following one plausible causal chain:

  1. Why was QC purity low? Because excessive impurity remained in the product.
  2. Why did excessive impurity remain? The manufacturing purification step did not achieve normal separation.
  3. Why was separation abnormal? The buffer conditions delivered to the column were inconsistent.
  4. Why were the buffer conditions inconsistent? An automated valve controlling buffer delivery was intermittently malfunctioning.

This produces a very reasonable candidate root cause.

3. Don’t confuse finding a defect with proving causation

Engineering examines the valve and finds evidence of intermittent malfunction. That’s important, but:

Key principle

A defective valve is not automatically proof that the valve caused these OOS results.

The investigation therefore goes further. Engineering performs controlled diagnostic testing of the removed valve and demonstrates that it intermittently sticks. Additional testing shows:

Valve sticks
   ↓
Incorrect buffer mixture
   ↓
Abnormal conductivity/composition
   ↓
Altered chromatography conditions

Historical process data from affected batches are also consistent with these abnormalities. Development or engineering studies can then reproduce the abnormal buffer condition and demonstrate impaired purification corresponding to the observed impurity.

Now the causal argument becomes substantially stronger:

Valve malfunction
   → Incorrect buffer composition
   → Poor chromatographic separation
   → Increased impurity
   → QC HPLC purity < 95%

At this point, the valve is much more than something investigators happened to find broken. There is evidence supporting a causal mechanism.

Advanced practice · The evidence ladder

Each rung is a stronger claim than the one below it. Write down which rung your evidence actually reaches.

RungClaimEvidence needed
1Something is wrongThe valve is found defective
2It can cause the failureBench testing shows sticking produces the wrong buffer; scale-down studies reproduce the impurity
3It did cause these failuresBatch records for the affected batches show the valve’s process signature (conductivity or composition deviations) at the right time
4It caused all of themEvery OOS batch shows the signature, and no other mechanism fits
5It is the only causeCan’t be proven. It can only be supported by continued monitoring.

Pathway A reached rung 3 for some batches. The trouble later in this case comes from quietly assuming rungs 4 and 5.

4. CAPA

The company replaces the valve and may also address the systemic reason it failed, for example preventive maintenance, inspection frequency, component-life limits, diagnostics, or process monitoring.

Subsequent batches pass. The effectiveness check might include:

Valve performance + buffer conductivity/composition + relevant purification parameters + absence of valve-associated purity failures.

Twenty or thirty consecutive batches might perform normally. The CAPA appears effective. And importantly: it may actually be effective.

Advanced practice · Design effectiveness checks around the mechanism

  • Measure the mechanism, not just the outcome. “No OOS for 30 batches” is an outcome. “Buffer conductivity within limits on every run, with no valve alarms” shows the specific cause stays controlled. That distinction is what saves you in Section 5.
  • Use leading indicators. Valve response time, conductivity variability, and purity trending within spec warn before an OOS occurs.
  • State the scope honestly. Write: “Effective for the valve-related mechanism under the conditions evaluated,” not “root cause eliminated.”
  • Record the conditions covered. Note which excipient lots, hold times, and campaigns the effectiveness batches included. It tells future investigators what the check did and didn’t test.

5. Months later: another OOS

Several months later: purity = 94.3%. Now there is a major reasoning trap. Someone might immediately conclude:

“The valve CAPA failed.”

Or even:

“Apparently the valve wasn’t the original root cause.”

Neither conclusion is justified yet. The correct question is:

The correct question

Did the same causal mechanism recur?

Engineering evaluates the purification skid:

  • The replacement valve is functioning correctly.
  • Buffer composition is correct.
  • Conductivity, flow, and pressure are normal.
  • The process signatures associated with the original valve malfunction are absent.

That changes the investigation dramatically.

6. The investigation broadens

Now investigators use more discriminating RCA techniques:

  • Kepner-Tregoe Is / Is-Not
  • Change Analysis
  • Batch stratification
  • Good-vs-bad comparison
  • Cause Mapping

Change Analysis identifies several things that changed over the relevant period. Two factors become particularly interesting:

  • A particular excipient characteristic / lot condition
  • Extended intermediate hold time

But there’s a complication. Some batches using that excipient pass. Some batches with extended hold times also pass. Therefore:

Excipient alone ≠ sufficient, and extended hold alone ≠ sufficient.

Then investigators examine the combination:

Excipient conditionHoldResult
NormalNormalPass
NormalExtendedPass
AffectedNormalPass
AffectedExtendedOOS

Now there is evidence of an interaction. Controlled experiments reproduce the phenomenon. The combination causes increased protein degradation and ultimately:

Excipient characteristic + extended hold
   → Protein degradation
   → Increased impurity
   → QC HPLC purity < 95%

Advanced practice · Make the interaction useful to manufacturing

  • Find the boundary. Extend the DOE to map how long a hold is safe for affected excipient lots. That turns a finding into a control limit.
  • Identify the excipient attribute. Characterize what differs in affected lots (for example, a trace impurity, pH, peroxide level, or metal content) so it can be tested on incoming lots.
  • Feed the control strategy. The attribute may be a critical material attribute and the hold time a critical process parameter (ICH Q8/Q11 thinking). Update specifications, supplier agreements, and batch record limits accordingly.

7. Now there are two demonstrated pathways

The causal structure looks more like this:

PATHWAY A

Intermittent valve failure
        ↓
Incorrect buffer composition
        ↓
Abnormal purification
        ↓
Increased impurity
        ↓
       OOS
   Purity <95%


PATHWAY B

Excipient characteristic ─────┐
                              ├── Interaction
Extended intermediate hold ───┘
                                     ↓
                              Protein degradation
                                     ↓
                              Increased impurity
                                     ↓
                                    OOS
                                Purity <95%

That’s why 5 Whys becomes limited in this scenario. It wasn’t necessarily wrong: it successfully followed Pathway A. The problem is that a linear 5 Whys chain doesn’t naturally reveal that Pathway B also exists.

Advanced practice · The fault tree view

A fault tree shows the same structure with logic gates, which makes the “either pathway” and “both factors needed” relationships explicit:

TOP EVENT: QC HPLC purity < 95%
  └─ OR ─┬─ Pathway A: valve failure → wrong buffer → poor separation
         ├─ Pathway B: AND ─┬─ excipient characteristic present
         │                  └─ extended intermediate hold
         └─ Unknown / not yet identified pathway

Two practices worth adopting: always keep an “unknown pathway” branch on the tree, and keep the tree as a living document in the CAPA or knowledge-management system, updated when new evidence appears (ICH Q10 knowledge management).

8. The historical problem

Now QA asks: which mechanism caused the original OOS batches? And there may not be a complete answer. Suppose historical review finds:

Historical OOSValve evidenceExcipient conditionExtended holdAttribution
#1StrongNoNoValve strongly supported
#2PossibleYesYesAmbiguous
#3UnknownYesYesExcipient pathway plausible
#4StrongNoNoValve strongly supported
Later OOSNoYesYesExcipient pathway strongly supported

For #1 and #4, contemporaneous process data may strongly support the valve. But #2 and #3 are now questionable. Were they caused by the valve? The excipient/hold interaction? Both?

There may no longer be enough historical evidence to know. Samples may be unavailable. Relevant excipient characteristics may never have been measured. Process data may not have sufficient resolution.

The scientifically appropriate conclusion may therefore be:

Honest conclusion

Historical causal attribution remains indeterminate for certain events.

That’s not necessarily a failure of QA. The evidence required to distinguish mechanisms may simply never have been collected because nobody knew the second mechanism existed.

Advanced practice · Handling indeterminate attribution

  • Use defined attribution categories so conclusions are consistent: demonstrated, strongly supported, plausible, indeterminate. Record the evidence behind each.
  • Look for a fingerprint before giving up. If retained samples of #2 and #3 exist, characterize their impurity profile (for example, SEC, CE-SDS, or LC-MS). A separation failure and a degradation pathway often leave different impurity species. Matching each batch’s profile to the Pathway A and Pathway B signatures may resolve the ambiguity.
  • Weigh evidence explicitly. For each batch, ask how much more likely the observed data are under Pathway A than Pathway B. Evidence that fits both equally (such as “purity was low”) is worthless for attribution, however compelling it feels.
  • Consider “both.” A marginal valve fault and a partial degradation effect could each contribute. The pathways aren’t necessarily exclusive.
  • Document it and move on. Addend the original investigation records with the updated attribution and rationale, confirm the product-impact assessment still holds for released batches (21 CFR 211.192 extends investigations to other potentially affected batches), and make sure the CAPA covers both pathways, so the unresolved history doesn’t leave a gap in current control.

9. What each RCA tool contributes

This scenario is particularly useful educationally because no single RCA tool solves the whole investigation.

ToolRole in this case
BrainstormingQuickly develops initial hypotheses
5 WhysFollows the initial valve causal chain
Change AnalysisIdentifies what changed before failures
KT Is / Is-NotIdentifies meaningful distinctions between passing and failing batches
StratificationLooks for relationships among skid, lot, hold time, materials, dates, etc.
Paired comparisonCompares closely matched good and bad batches
Cause MappingRepresents multiple simultaneous or independent causal pathways
FTAModels combinations and alternative pathways capable of producing the top event
DOE / controlled testingTests the interaction between excipient condition and hold time
CAPA effectivenessDetermines whether the targeted causal mechanism remains controlled

Advanced toolkit: when to choose what

For experienced investigators, these additional tools and practices strengthen a multi-pathway investigation. Each has a best moment and a real cost.

Tool / practiceBest stageWhy choose itProsCons
Impurity profiling (“failure fingerprint”)Day one, and again at each recurrenceTells mechanisms apart by what the impurity is, not just how muchCan separate pathways and resolve historical attribution from retained samplesNeeds orthogonal methods and retained samples; costs time and money
Analysis of Competing Hypotheses (ACH)After the first hypotheses formScores every piece of evidence against every hypothesis, focusing on what disprovesCounters anchoring and confirmation bias; auditable reasoningTakes discipline; only as good as the hypothesis list
Evidence ladder / causal criteriaBefore closing any conclusionMakes explicit whether evidence shows “can cause,” “did cause,” or “only cause”Prevents overclaiming in the CAPA and to regulatorsRequires judgment; can feel slow under release pressure
Multivariate data analysis (e.g., PCA/PLS on batch data)When many process variables are recordedFinds patterns and combinations humans miss in large datasetsHandles many variables at once; can reveal interactions to testCorrelation only; needs clean data and statistical skill
Classification tree or logistic regressionStratification stageShows which factor combinations predict OOSQuickly surfaces candidate interactionsFew OOS events means overfitting risk; hypotheses, not proof
Scale-down model studiesConfirming a mechanismReproduces failure conditions safelyAllows “turn it on and off” testingModel must be qualified as representative of full scale
Fault tree as a living documentAfter the first pathway is confirmedKeeps OR/AND structure and an “unknown” branch visibleMakes future triage fast; supports knowledge managementNeeds ownership and upkeep
Pre-mortem / red team reviewBefore CAPA closureAsks “if this fails again, what would the cause be?”Surfaces untested pathways cheaplyCan generate many speculative branches; needs prioritizing
SPC on purity and mechanism indicatorsOngoing, after CAPADetects drift before an OOSLeading indicator; catches new pathways earlyNeeds defined limits and someone reviewing the charts

10. Advanced guidance

Cognitive traps in multi-pathway investigations

TrapHow it shows up hereCountermeasure
AnchoringEvery new OOS gets read as “the valve again”Run the signature check before assuming the known cause
Satisfaction of searchInvestigators stop once one real defect is foundAsk: does this explain every OOS batch, and every passing batch?
Confirmation biasOnly evidence that fits the valve story is gatheredSeek disconfirming evidence; use competing hypotheses
Outcome-based judgment“No OOS for 30 batches, so the root cause is gone”State what the effectiveness check covered and what it didn’t
Hindsight bias“The first team should have found the excipient issue”Judge the first investigation by what was knowable then

New investigation or reopen the old CAPA?

Open a new investigation for the recurrence, linked to the original. Only conclude the first CAPA was ineffective if the same mechanism recurred. Here it didn’t: the valve CAPA remained effective for its mechanism, and a second CAPA addresses Pathway B. Update the original record with a cross-reference and any revised historical attribution.

A triage checklist for the next low-purity OOS

  1. Confirm the laboratory result (Phase I).
  2. Characterize the impurity profile and compare it with known pathway fingerprints.
  3. Check Pathway A signatures: valve diagnostics, buffer conductivity and composition, flow, pressure.
  4. Check Pathway B conditions: excipient lot attribute and intermediate hold time.
  5. If neither pathway’s signature is present, treat it as a possible new pathway and open a full investigation.

Regulatory anchors

  • FDA OOS guidance: Phase I laboratory investigation before manufacturing investigation; no invalidation without an assignable cause.
  • 21 CFR 211.192: investigations must be thorough, documented, and extended to other batches that may be affected.
  • ICH Q10: CAPA effectiveness and knowledge management across the product lifecycle.
  • ICH Q9(R1): depth of investigation and control proportionate to risk.
  • ICH Q8 / Q11: material attributes and process parameters that affect quality belong in the control strategy.

The deepest lessons from this case

  • Finding something wrong does not prove it caused the event.
  • Demonstrating that something can cause the failure does not automatically prove it caused every historical failure.
  • Correcting a cause does not prove it was the only cause.
  • Recurrence of the same symptom does not automatically mean the previous CAPA failed.
  • A successful effectiveness check demonstrates control over the conditions evaluated; it cannot guarantee that another unknown causal pathway will never produce the same outcome.

And perhaps the most important:

The core principle

A good RCA is not about finding a convincing story. It is about building and testing a causal model that adequately explains the evidence.

That’s why this case works so well as an advanced capstone for root cause analysis training. The investigators aren’t incompetent, the first RCA isn’t necessarily wrong, and the first CAPA isn’t necessarily ineffective. Their understanding of the causal system becomes more complete as new evidence appears.

Questions for experienced professionals

  1. At which rung of the evidence ladder was the valve after Section 3? What would it have taken to reach the next rung for every OOS batch?
  2. How would you rewrite the original CAPA closure statement so it stays accurate after Pathway B is discovered?
  3. Which data, collected routinely from the start, would have made historical batches #2 and #3 attributable today?
  4. Design the smallest experiment that would prove the excipient × hold interaction. How many runs, and why replicates?
  5. Your site head asks, “Did the first CAPA fail?” Answer in two sentences an inspector would accept.
  6. What other products or processes share the excipient or the hold step, and how would you scope the read-across?

Related reading: The Oven That Only Failed Sometimes, The Standard That Failed the Samples, and Advanced Root Cause Analysis.

Kepner-Tregoe is a proprietary method of its owner, and Cause Mapping is associated with its developer; both are described here for educational purposes only. QStead Group is not affiliated with them. This case study is for educational purposes only.

Facing an investigation with more than one possible cause? Get in touch.

← Back to Case Studies