ASTM work item WK92658 wants to standardise how cleaning processes are judged capable — one-sided limits, small data sets, non-normal residues and continued verification included. It is the statistical gap E3106 left open.
ASTM Puts Process Capability on the Cleaning Validation Agenda
A new ASTM work item, WK92658, sets out to give cleaning validation what it has long operated without: an agreed way to express how capable a cleaning process really is — a Cpu rather than a passed swab result.
What happened
On 29 October 2024, ASTM International registered work item WK92658, "New Guide for Evaluating the Risk of Cleaning Processes Using Process Capability for Pharmaceutical Products and Medical Devices." It sits under Subcommittee E55.13 on Process Evaluation and Control (Committee E55); technical contact Andrew Walsh, staff manager Travis Murdock.
The stated rationale is blunt for a standards body: "There are no standards or guidance on how to use process capability for evaluating cleaning processes." Scope reaches APIs, all dosage forms, OTC and veterinary products, biologics, clinical supplies, medical devices and cosmetics. Listed keywords are process capability, process capability index and Cpu — the one-sided upper capability index, the correct index for a residue limit with no lower limit.
The ASTM work item record shows WK92658 at ballot (E55.13 (25-01)), title and scope still marked as draft. A work item is a declared intention, not a published standard.
Why it matters
Cleaning validation has largely been argued from outcomes: a finite number of runs, a finite number of swabs, all below the limit, therefore the process is validated. Process capability changes the question. It asks how much margin separates the process mean from the acceptance limit, and how much of that margin survives sampling uncertainty. A process passing by a factor of two and one passing by a factor of five hundred look identical in a pass/fail table but carry different risk.
E3106 already frames the level of effort, formality and documentation as proportional to risk, and the wider ASTM cleaning programme discusses a target capability of at least 2.0. What is missing — and what WK92658 addresses — is the arithmetic: how capability should be computed, reported and interpreted for cleaning data.
What is particularly interesting
Cleaning data violate almost every comfortable assumption of textbook capability analysis, and the work item is written from that reality.
One-sided specifications. Residue has an upper specification limit and no lower. A two-sided Cpk would be a category error; Cpu is the relevant index, and it needs a lower confidence bound, because with small samples the point estimate flatters the process.
Small data sets. Qualification and continued verification runs generate tens of samples, not thousands. Methods needing large-n behaviour are useless to an engineer holding twelve swab results and a deadline.
Non-normal, left-censored data. Residue results are frequently below the detection limit. Substituting zero, the LOD, half the LOD, or LOD over the square root of two yields materially different means, standard deviations and capability values from one data set. Discarding non-detects inflates capability; substituting zero understates it.
Continued verification. Stage 3 is where capability belongs. A single qualification snapshot proves the process was capable once; monitoring shows it still is.
The work item builds on articles by ASTM cleaning team members: a capability primer (2021), an examination of total organic carbon data, treatments of partially censored data (May 2022) and completely censored data (October 2022), then work on sample size and run counts.
What could we learn
A passing result and a capable process are not the same claim, and reviewers increasingly know the difference. Non-detect handling should be a documented, pre-specified decision rather than whatever the spreadsheet did by default. And qualification run counts can be justified by data rather than habit — several of the underlying articles argue two runs may suffice.
Questions for our programme
- Which of our cleaning processes have a documented Cpu with a lower confidence bound, and which have only a pass/fail conclusion?
- How do we currently treat non-detect results, and is that rule written into the protocol before the data arrive?
- Can we justify our number of performance qualification runs statistically rather than by precedent?
- Does our capability data match the risk claim we make about it — per product, per equipment train, or per cleaning process?
- Are our analytical methods, particularly non-specific ones such as TOC, sensitive enough at the concentrations that matter for the compounds we handle?
- What would change in our continued verification plan if capability, not compliance, drove sampling frequency?
Wider context
WK92658 is one thread in a broader ASTM effort to make cleaning validation quantitative. E3470-25 on using FMECA and HACCP for risk identification, analysis and control on cleaning processes, placing a process capability score (ten divided by Cpu) on one axis of a Shirokizawa criticality matrix against a toxicity score from health-based exposure limits — capability becomes a decision input. In January 2026, AAPS PharmSciTech published work on specification limit normalisation for non-parametric capability analysis with AI-enabled monitoring, addressing the same constraints: multiple limits, small samples, left-censored non-normality.
Sources
- ASTM work item WK92658, E55.13, initiated 29 October 2024 — https://store.astm.org/workitem-wk92658 (ASTM page returned HTTP 403; facts confirmed from the work item record and the E55.13 listing)
- ASTM E3106-22; ASTM E3470-25 — https://store.astm.org/e3470-25.html
- Yang X. et al., Specification Limit Normalization for Nonparametric Process Capability Analysis, AAPS PharmSciTech, Jan 2026 — https://link.springer.com/article/10.1208/s12249-025-03267-5
- Walsh A. et al., Calculating Process Capability of Cleaning Processes With Partially Censored Data, Pharmaceutical Online, May 2022
Validation takeaway
Process capability turns a cleaning validation from a claim about results into a claim about margin. With one-sided specifications, small sample sizes and left-censored data, the index is only as credible as the pre-specified rule used to handle non-detects and the lower confidence bound reported alongside it.
Questions worth asking in our own programme
- Which of our cleaning processes have a documented Cpu with a lower confidence bound, and which have only a pass/fail conclusion?
- How do we currently treat non-detect results, and is that rule written into the protocol before the data arrive?
- Can we justify our number of cleaning performance qualification runs statistically rather than by precedent?
- Is our capability data assembled per product, per equipment train, or per cleaning process, and does that grouping match the risk claim we make?
- Are our analytical methods, particularly non-specific methods such as TOC, sensitive enough at the concentrations that matter for the compounds we handle?
- What would change in our continued verification plan if process capability, not compliance, drove sampling frequency?
