On this page
The Recurrence Problem CAPAs Are Meant to Solve
A corrective and preventive action that closes fast but does not fix the root cause is worse than no CAPA at all, because it tells management the problem is handled while the same temperature deviation keeps coming back. The classic pattern in a Kenyan distribution warehouse is a CAPA log that looks excellent on paper, almost everything closed inside the target window, while the monthly deviation count quietly climbs. The closure metric is green and the actual control is failing.
PPB Good Distribution Practice and the wider quality-system expectation are explicit that management review must be forward looking. It is not enough to count last month's excursions and confirm CAPAs were raised. The review needs to see which deviation streams are trending up and which CAPAs are weak enough that the problem is likely to repeat, so scarce engineering and QA time goes to the right place first. That is the gap the CAPA Recurrence Predictor fills.
The tool combines two signals your systems already hold: the trend in monthly deviation counts and the quality of the CAPAs raised against them. A flat trend with strong, verified CAPAs is low risk. A rising trend with CAPAs that closed quickly but never had a real effectiveness check is high risk, and it deserves escalation before the next excursion forces the issue. This guide walks one realistic example end to end, explains each output, and shows how to write it up so it survives an audit. It is a companion to your deviation and CAPA SOPs, not a replacement for them.
Worked Example: A Warehouse With a Rising Trend
Take a distributor running a single 2 to 8C warehouse zone and pull 12 months of monthly temperature-deviation counts from the deviation tracker. The series climbs through the year:
- Months 1 to 4: 2, 2, 3, 2 deviations per month. Stable, roughly two a month.
- Months 5 to 8: 3, 4, 4, 5 deviations per month. The count is drifting up.
- Months 9 to 12: 5, 5, 6, 6 deviations per month. The recent run sits at three times the early baseline.
Fitting a straight line to those 12 points gives a trend slope of about plus 0.4 deviations per month, a clear and sustained rise rather than random noise. Next the tool reads the CAPA closure-quality side. Of the CAPAs raised against this stream, most closed inside ten days, which looks efficient, but the effectiveness-check field is blank or marked "monitor only" on roughly 70 percent of them. Fast closure, weak verification: the corrective actions were administrative, not effective.
A rising slope multiplied by poor CAPA effectiveness is exactly the high-risk combination. The model translates it into a recurrence probability of about 70 percent for a repeat deviation next month, places this stream in the high risk band, and projects the next-period count near 6 to 7. Set against the other CAPA streams on site (a packing-line label deviation that is flat and well verified, and a documentation deviation already trending down after a strong CAPA), this warehouse temperature stream ranks first for escalation at management review. You did not wait for the seventh deviation to tell you the CAPAs were not working. The trend and the verification gap told you a month early.
What Data You Need and Why
Three inputs drive the recurrence score. The first is a series of monthly deviation counts for the stream you are assessing, one count per month, taken straight from your deviation or excursion tracker. This is what produces the trend slope, so it must be the genuine count, not a rounded or capped figure.
The second is a set of CAPA closure and effectiveness indicators for the CAPAs raised against those deviations: closure time, whether an effectiveness check was performed, and whether that check passed. A CAPA that closed in five days with a documented, passed effectiveness review is strong. One that closed in five days with the effectiveness field left blank is weak, and the model treats it that way. If your tracker does not separate closure from effectiveness, that is the first record gap to fix, because the whole forecast hinges on it.
The third is a sufficient time span. Aim for at least 12 months of monthly counts so the trend is real rather than a two-point guess. A short window cannot tell a rising trend from ordinary month-to-month variation, and it will either miss a building problem or cry wolf on noise. The minimum useful file is two columns, month and deviation count, with the CAPA-quality indicators alongside or in a linked column. Keep headers on row 1 and remove any summary or total rows. If your export is messy, clean it through the Data Logger QA Cleaner before uploading here.
Reading the Outputs
The tool returns five outputs. Read them top down:
- Risk score: a single composite that blends the trend direction with CAPA effectiveness. It exists to rank one CAPA stream against another, so use it for prioritisation, not as a standalone verdict.
- Risk band: the score bucketed into low, medium, or high. The band is what you carry into management review, because it maps directly to an escalation rule (for example, every high-band stream gets a named owner and a deadline that month).
- Recurrence probability: the estimated chance of a repeat deviation in the next period, the 70 percent figure in the worked example. Read it as a likelihood, not a promise; it is high enough to act on well before certainty.
- Trend slope: the rate of change in monthly counts. A positive slope means the problem is growing even if each individual month still looks acceptable. This is the early-warning number.
- Next-period projection: the expected deviation count next month if nothing changes. It turns the trend into a concrete target you can check against reality at the following review.
Never read these in isolation. A high score driven by trend with strong CAPAs points to an equipment or process problem the CAPAs have not yet reached; a high score driven by weak CAPAs points to your CAPA process itself. Pair the numbers with the operational context behind each deviation: door openings, loading peaks, defrost timing, maintenance state, and sensor placement. If the deviations themselves are clustered around a recurring event, the fix may live upstream. An excursion impact assessment tells you whether each of those deviations actually threatened product, which sharpens how hard you escalate. And if the deviations are nuisance alarms rather than real breaches, the Alarm Threshold Optimizer may cut the count at source.
Documenting It for Management Review and GDP Audits
A recurrence forecast is a management review input and a CAPA effectiveness verification, so capture it the way an inspector expects to see it:
- Attach the raw deviation-count export and the CAPA quality data used for the analysis, archived unchanged for traceability.
- Record the trend slope, recurrence probability, risk band, and next-period projection for each stream, with the run date and the analyst.
- State the management decision the band drove: which stream was escalated, who owns it, and the deadline. A high band with no recorded action is a finding waiting to happen.
- Use the forecast as your CAPA effectiveness verification step. A CAPA closed last quarter that sits in a still-rising, high-band stream has demonstrably not been effective, and that conclusion belongs in the record.
- Route any resulting change through change control with QA sign-off, and re-review at the next management review against the actual deviation count, comparing it to the projection.
That loop, project the count, act, then check the projection against what happened, is what makes the practice defensible under PPB GDP and credible as continuous improvement rather than box-ticking. If you are still building these records, our GMP readiness checklist covers the deviation, CAPA, and management review documentation an inspector will ask for.
Frequently Asked Questions
How is this different from just counting last month's deviations?
A count tells you what already happened. The recurrence predictor combines the trend across months with CAPA effectiveness to estimate what happens next, so management review can escalate a building problem before the next deviation rather than after it.
How many months of data do I need?
Aim for at least 12 months of monthly deviation counts. A short window cannot separate a real rising trend from ordinary month-to-month variation, so it either misses a building problem or raises a false alarm on noise.
Why does CAPA closure speed not lower the risk on its own?
Fast closure without a verified effectiveness check is administrative, not effective. The model weights effectiveness, so a stream where most CAPAs closed quickly but never passed an effectiveness review still scores high if the deviation trend is rising.
Can I use the recurrence probability as a CAPA effectiveness verification?
Yes. A CAPA that sits in a still-rising, high-band stream has demonstrably not worked, and recording that conclusion satisfies the effectiveness verification step your quality system requires.
Will PPB accept a forecast as a basis for escalation?
Inspectors accept evidence-based, forward-looking decisions. A forecast tied to documented monthly counts, CAPA effectiveness data, and a recorded management decision is defensible under Good Distribution Practice. An undocumented escalation, or a green closure metric with no trend view, is not.
Forecast Your Own Recurrence Risk
Upload 12 months of deviation counts and CAPA quality indicators and get a risk score, risk band, recurrence probability, trend slope, and next-period projection you can take straight into management review.
Open CAPA Recurrence Predictor