Pp / PpkProcess Performance Index

Pp and Ppk answer the question 'over the last reporting period — typically rolling 12 months or 25 batches — has this process actually delivered product within specification with adequate margin?'. They differ from Cp/Cpk in one variable: the standard-deviation estimate. Cp/Cpk uses σ_within (the within-subgroup standard deviation, estimated from R-bar / d2 or s-bar / c4), which captures only common-cause variation inside a single batch / shift / subgroup. Pp/Ppk uses σ_overall (the sample standard deviation across all measurements in the window), which captures common-cause AND between-batch variation — operator change, supplier-lot change, season, equipment drift. σ_overall is almost always larger than σ_within; therefore Pp ≤ Cp and Ppk ≤ Cpk for the same process. The gap is itself diagnostic.

• Pp = (USL − LSL) / (6 × σ_overall)
• Ppu = (USL − mean) / (3 × σ_overall)
• Ppl = (mean − LSL) / (3 × σ_overall)
• Ppk = min(Ppu, Ppl)
• σ_overall = √( Σ(xi − x̄)² / (n − 1) ) — pooled across every observation in the window.

Ppm (Process Performance Mean) — Ppk corrected for off-target operation, analogous to Cpm. Less commonly reported but valuable when target is the explicit aim (e.g. dose accuracy at 100% of label claim).

• Ppk ≥ 1.67 — process performing comfortably above 6σ; commonly required for CQAs in commercial pharma.
• Ppk 1.33-1.67 — adequate but watch trends; FDA's commonly-cited capability target sits in this band.
• Ppk 1.00-1.33 — acceptable but variable; investigate sources of between-batch variation.
• Ppk < 1.00 — process is producing out-of-spec material at a rate of >2,700 ppm; an OOS event is statistically likely soon.
• Ppk negative — process mean is outside the spec; immediate intervention required.

If Cpk is materially better than Ppk (say Cpk = 1.5 but Ppk = 1.0), the process is capable within a batch but drifting between batches. Diagnostically this points to assignable causes outside the subgroup definition: incoming-material lot variation, environmental drift (humidity, season), operator skill differences across shifts, calibration drift, equipment wear. The Cpk-Ppk gap is one of the first signals examined in a Stage-3 CPV review and the most useful single number for prioritising continuous-improvement effort. If Cpk ≈ Ppk the process is homogeneous over time — closing it tighter requires reducing common-cause variation (DoE, equipment upgrade, supplier consolidation).

A single Pp/Ppk number per quarter is necessary but insufficient — it tells you the historical capability of the window, but not whether the process is changing inside the window. The standard pairing is a rolling Pp/Ppk plus a control chart of subgroup means with Western Electric / Nelson out-of-trend rules: a point beyond 3σ, two of three beyond 2σ, four of five beyond 1σ, eight in a row on one side of the centerline, six trending, fourteen alternating, etc. The chart signals shifts and trends; the Pp/Ppk number quantifies the resulting margin to spec.

Pp/Ppk is statistically unstable below ~25 data points and unreliable below ~10. Stage-3 CPV programmes typically wait until 25 commercial batches have been made before reporting Pp/Ppk with confidence; before that they report individual-value charts and lot-by-lot pass/fail. The reporting window itself is a deliberate choice — too short and the index is noisy; too long and it averages over distinct process states (pre/post a change-control, pre/post a supplier change). A 12-month rolling window with a CUSUM or change-point analysis is a common middle ground.

1. Reporting Cpk in the APR but calling it Ppk — same letter swap, fundamentally different number.
2. Using a long window that spans a process change, washing out the very improvement the change was supposed to deliver.
3. Calculating Ppk on a non-normally-distributed attribute without transformation — Box-Cox or distribution-fit first, then index.
4. Ignoring one-sided spec attributes (e.g. assay ≥ 95%) — report Ppk against the one-sided limit, not a fake LSL.
5. Using Ppk to argue 'capability proven' without a parallel SPC chart — index is high because the average is on-target but the chart shows a clear up-drift.
6. Mid-window outlier (a lab error, a re-tested sample, a real OOS) included or excluded inconsistently — the rule must be pre-defined.

• Per-CQA rolling Pp/Ppk and Cp/Cpk shown side by side on the Stage-3 dashboard — the gap is highlighted automatically.
• Configurable window (rolling 12 months / 25 batches / since-last-change-control) — change-control IDs are visible on the chart so the analyst can see the process state.
• Western Electric and Nelson out-of-trend rules running in real time; a violation opens a Stage-3 investigation with all the lot context attached.
• Distribution-fit check (Anderson-Darling, Shapiro-Wilk) flagged before reporting Ppk — non-normal attributes are transformed or analysed by percentile.
• One-sided spec handling — Ppk against the relevant limit, not a synthetic LSL.
• APR / PQR consumes the rolling Pp/Ppk per CQA — the annual review and the live dashboard agree by construction.