CAPA 里涉及检测复核的处理路径

结论：CAPA 里检测出现三次，每次目的不同

一个完整的 CAPA 流程中，检测数据通常要出现三次：确认问题时、分析原因时、验证有效性时。 三次的目的不同，需要的数据也不同。

混淆这三者是常见问题。比如用确认问题的那份数据去充当有效性验证，逻辑上不成立——那是问题发生前的数据，证明不了措施有效。

三个阶段的数据要求

问题确认阶段

这一步要回答两件事：问题是不是真的，以及影响了多大范围。

复现是关键。 收到一个投诉或发现一个不合格，先要确认它能不能复现。不能复现的，可能是偶发、可能是使用问题、也可能是条件没找对。直接进入原因分析，容易在错误的方向上花很多时间。

范围界定同样依赖数据：需要抽查一定数量的同批次、相邻批次产品，判断问题是个别的还是批量的。抽查数量要有依据，抽三台都合格不足以证明批量没问题。

原因分析阶段

这一步的数据要能区分变量。做法是设计对比：

怀疑是某个物料批次的问题，就用不同批次物料做出来的产品做对比；怀疑是工艺参数，就在不同参数下做样品对比；怀疑是设备，就用不同设备生产的产品对比。

没有对比的数据说明不了因果。 只测有问题的那批，发现某个指标偏低，这只能说明现象，不能说明原因——也许所有批次这个指标都偏低，只是这批碰巧出了问题。

分析阶段还有一个容易忽略的点：要区分「相关」和「因果」。两个现象同时出现，未必一个导致另一个，可能有共同的第三个原因。

有效性验证阶段

这是最容易走过场的一步。常见的不到位表现：

用一次合格结果就宣告有效。 措施实施后测了一批，合格，就关闭 CAPA。但如果原来的问题是偶发的，一批合格证明不了什么。

样本量不足。 原问题的发生率如果是较低的比例，验证的样本量要足够大才能说明问题确实消除了。

验证条件与问题发生条件不同。 原来的问题在某个特定条件下出现，验证时却在常规条件下做，这样的验证覆盖不到。

只验证了直接指标。 措施可能解决了原问题但引入了新问题，验证时应当一并关注相关指标有没有变化。

有效性验证的样本量和周期应当在制定措施时就确定，而不是做完之后凑一份数据。

与检测机构的配合

如果自身检测能力不足，CAPA 的某些环节可能需要外部支持。委托时要说清楚的是：这次检测的目的是哪一阶段。

目的不同，方案完全不同。问题确认阶段需要的是覆盖面，多测几个样品看分布；原因分析阶段需要的是对比设计，分组要清楚；有效性验证需要的是与历史数据可比，方法条件必须保持一致。

只说「帮我测某个项目」而不说目的，方案很可能不合用。

记录要留什么

CAPA 的检测记录除了数据本身，还应当记录：这次检测对应 CAPA 的哪个阶段、样品的来源与选取依据、与哪些历史数据做对比、结论如何支撑该阶段的判断。

这些上下文信息不记录，几个月后回看时数据就失去意义了。 审核时被问到某份数据为什么支持这个结论，也答不上来。

走过场的典型表现

把几种常见的形式主义列出来，自查时可以对照：原因分析写「操作人员疏忽」而无数据支撑；纠正措施写「加强培训」而无有效性数据；有效性验证用的是措施实施前的数据；范围界定只看了投诉涉及的那一台；预防措施与原因分析没有对应关系。

这些写法在文件上是完整的，实质上是空的。 审核员通常一眼能看出来，因为它们缺少的恰恰是数据。

预防措施的数据支撑

CAPA 里的「预防」部分常常比「纠正」写得弱。纠正针对已发生的问题，容易写具体；预防针对尚未发生的，容易写成口号。

让预防措施落地的办法是：把这次问题的原因抽象成一类风险，然后排查产品线上其他地方有没有同类风险。排查需要数据——不是问一遍有没有，而是实际去测、去查记录。

比如这次问题的原因是某个尺寸公差控制不住，那么预防措施就该是排查其他关键尺寸的过程能力，而排查结果应当是一组数据。

关闭之后的跟踪

CAPA 关闭不等于结束。建议对关键的 CAPA 设置关闭后的跟踪期，在一定时间或一定产量之后再核查一次问题有没有复发。

有些措施在短期内有效，时间一长就松懈了，尤其是依赖人员执行的措施。跟踪期的数据能发现这种回弹。

时间节点的管理

CAPA 各阶段应当有明确的时限，否则容易拖成悬案。实务上比较合理的做法是按阶段设定：问题确认与范围界定要快，因为它关系到是否需要采取紧急措施；原因分析可以给足时间，因为仓促下结论往往找错方向；措施实施按措施本身的复杂度定；有效性验证的周期由数据积累速度决定，不能压缩。

有效性验证这一段最容易被压缩，因为前面几段拖延了，到了最后为了按时关闭就草草验证。这是 CAPA 流于形式的常见路径。

多个 CAPA 之间的关联

如果一段时间内出现多个 CAPA，值得把它们放在一起看：是不是指向同一个系统性问题。

单个看都是孤立事件，合起来看可能发现共同的根源——比如都与某条产线有关、都发生在某个班次、都涉及某个供应商。这种跨 CAPA 的分析需要有人定期做，否则每个 CAPA 各自关闭，系统性问题一直存在。

我们的做法

接到 CAPA 相关的检测委托，我们会先问这次检测服务于哪个阶段。这个问题决定了抽样方式、对比设计和报告的写法。

对于有效性验证类的委托，我们会特别注意方法条件与历史数据保持一致——如果条件不同，数据没有可比性，验证就不成立。这一点需要委托方提供历史报告，我们才能对齐。

如果你正在处理一个 CAPA，需要数据支撑某个环节，可以把问题描述和已有数据发过来一起设计方案，或者直接联系：132 4819 8029。能力范围见服务介绍，流程见检测流程，更多内容见知识库。

English version

Conclusion. Test data normally appear three times in a complete corrective and preventive action: confirming the problem, analysing the cause, and verifying effectiveness. The three have different purposes and need different data. Confusing them is common. Using the data that confirmed the problem as effectiveness evidence does not work logically, because those data predate the action and cannot show it worked.

Confirming the problem. Two questions: is the problem real, and how far does it extend. Reproduction is central. On receiving a complaint or finding a nonconformity, first establish whether it can be reproduced. Something that cannot be reproduced may be sporadic, may be a usage issue, or may need different conditions to appear, and moving straight to cause analysis risks spending considerable effort in the wrong direction. Bounding the extent also needs data: sample enough units from the same and adjacent batches to judge whether the problem is isolated or systematic, and base the sample size on something, since three conforming units does not demonstrate a batch is sound.

Analysing the cause. Data here must separate variables, which means designing comparisons. Suspect a material batch, and compare product made from different batches. Suspect a process parameter, and compare samples made at different settings. Suspect equipment, and compare output from different machines. Data without comparison cannot establish causation: measuring only the problem batch and finding one indicator low shows a phenomenon, not a cause, since every batch might sit low and this one simply failed for another reason. Also distinguish correlation from causation; two phenomena appearing together may share a third cause.

Verifying effectiveness. This is the step most often handled superficially. Typical shortfalls: declaring success on one conforming batch, which proves little if the original problem was intermittent; an insufficient sample size relative to the original occurrence rate; verifying under conditions different from those in which the problem appeared; and verifying only the direct indicator, when the action may have solved one problem and introduced another. Sample size and duration for effectiveness verification should be decided when the action is planned, not assembled afterwards.

Working with a laboratory. Where internal capability is insufficient, external support may be needed, and the commission must state which stage it serves. The design differs completely. Confirmation needs breadth, measuring more units to see the distribution. Cause analysis needs comparative design with clear grouping. Effectiveness verification needs comparability with historical data, so method conditions must be held constant. Asking simply for a test of a particular item, without stating the purpose, will often produce something unsuitable.

What to record. Beyond the data, record which CAPA stage the testing served, where the samples came from and how they were selected, which historical data they are compared against, and how the conclusion supports the judgement at that stage. Without this context the data lose meaning within months, and questions in audit about why a dataset supports a conclusion cannot be answered.

Signs of a hollow CAPA. Cause analysis attributing the problem to operator oversight with no supporting data. Corrective action consisting of additional training with no effectiveness data. Effectiveness verification using data from before the action. Scope bounded by looking only at the unit complained about. Preventive action with no traceable relationship to the cause analysis. These are complete on paper and empty in substance, and auditors usually recognise them immediately, because what is missing is precisely the data.

How we handle it. For CAPA-related commissions we ask which stage the testing serves, because that determines sampling, comparative design and how the report is written. For effectiveness verification we pay particular attention to keeping method conditions consistent with the historical data, which requires the client to supply the earlier report so we can align.

Send us the problem description and existing data and we will design the approach. Phone or WeChat: +86 132 4819 8029.