Six Sigma

There's a new term getting used by the cool kids: "67" What does this nonsense meme mean? More close to home, what does it mean when labs ask for Six Sigma metrics without defining any of the details?

Is Sigma getting “67”’d?

Sten Westgard, MS
December 2025

The strange internet thing of the year 2025 appears to be “six-seven.” Dictionary.com has named it the word (?) of the year for 2025. What does 67 mean? https://www.merriam-webster.com/slang/six-seven None of us oldsters know, and perhaps that is the point. A younger generation, however, is shouting out “6-7!” whenever they come across it in their math class.

Since it appears to mean almost anything, I’m going to steal it and associate it with a new phenomenon I’ve encountered several times this year in the laboratory world: asking for Six Sigma metrics without knowing anything about them.

In my work, I am sometimes involved in creating, mediating, or responding to RFPs from hospital groups seeking new instrumentation. One of the new trends I’ve seen in 2025 is the request for analytical Sigma metrics. “Show us your metrics,” orders the bid. But there is an appalling lack of specificity about those metrics.

1.  What goals are being used to calculate the metrics? CLIA 2025? EFLM 2025? Rilbaek? RCPA, Chinese goals? These goals are all different – and will produce different Sigma metrics.
2.  What constitutes the “Sigma metric” of a test? Is it the average of all control levels? The median? The highest number? The lowest number? Does it matter which control level is used for the Sigma metric?
3. What kind of control is acceptable for calculating analytical Sigma metrics? Independent 3rd party controls? Or can you use the manufacturer’s own optimized controls?
4. Over what time period must the Sigma metric be calculated? Is a within-run study of a single day acceptable? Should it be 5 days? 20 days? 3 to 6 months?
5. How is bias determined? Is it bias vs. the assayed/target/package insert mean? Bias vs. the peer group mean? Bias vs. a predicate device, like the previous model being replaced by this new model? Bias vs an EQA/PT group mean? Bias vs. a reference method or material?
6. Does the Sigma metric of a single lab suffice? Or should there be a consensus of multiple labs using the same instrument?
7. If a claim is made for 6 Sigma performance, is the evidence available for review? What data exists that can be made available that proves any particular metric?

Different answers to these questions will generate different Sigma metrics. Failure to specify these conditions leaves the door open to manipulation. Given the obvious pressure to achieve Six Sigma, manufacturers will be sorely tempted to pick and choose the specifics so that their methods come out best. Like any statistic, when the stakes are high, don’t be surprised is someone games the system, torturing the numbers until Six Sigma is achieved. I’ve already seen incredible claims being put forth, without any real evidence to back them up.

In other words, that 6 Sigma metric you see could actually be 3 Sigma, or 2 Sigma, or even 7 Sigma, under different circumstances.

We advise every lab that is seeking Six Sigma metrics to spell out their requirements/conditions. And we advise every vendor that submits Sigma metric data to be as transparent as possible, showing all the work, all the choices being made, all the references. Make it so transparent that the lab in question can double-check your work – they can gather all the inputs and run the calculations themselves and get the same answers. For labs that are suspicious of the numbers being given to them by vendors, I encourage them to reach out to other current customers of the method or instrument. Get real-world performance data from a lab, unmediated by the manufacturer. Again, that allows you to run the numbers honestly and independently, using the goals and conditions that are relevant to you.

If we don’t apply this rigor, the analytical Sigma metric will be just like every other abused statistic, one that’s impossible to trust, because the underlying data is missing. And you won’t know your 6’s from your 7’s.

A Sample Set of Sigma Specifications, Suitable for the USA

1. Use CLIA 2025 goals. Where those aren’t available, specify what goals are chosen instead.
2. Calculate an average of Sigma metrics across all levels measured. Or follow the critical decision levels detailed in the Westgard Sigma Verification of Performance (Sigma VP) program. Or show all Sigma metrics of all levels.
3. Use independent 3rd party controls. If those are not available, only then are manufacturer controls acceptable.
4. Estimates of imprecision should be measured over 3-6 months, ideally, but a minimum of 100 data points.
5. Ideally, bias should be measured against reference materials or methods. But realistically that happens for a tiny handful of methods. More practically, bias can be measured by comparing the observed mean vs the peer group mean. EQA/PT bias is also acceptable.
6. Ideally, Sigma metrics from multiple labs should be reported. A single lab is acceptable if the lab is fully identified. Internal manufacturer studies are acceptable when no other studies are available.
7. Make the data available upon request.