Interesting study published recently in the October 2025 issue of Philippine Journal of Pathology:
- Mark Angelo Ang, Karen Cybelle Sotalbo, Quantifying Total Allowable Error Violations in Serum-Sodium Quality Control: A Computer Simulation Experiment of Two- to Six-Sigma Processes, October 2025, Philippine Journal of Pathology 10(2), DOI: 10.21141/PJP.2025.08
The key paragraph for me:
"From a risk-management perspective, any test with ≥5-sigma capability for sodium essentially meets a "six-sigma" quality target after accounting for biological variation (≈0.5%).2,4,5 Such performance permits less intensive QC scheduling, potentially reducing reagent waste, technologist time and instrument down-time without compromising patient safety. Conversely, a 2-sigma process is categorically inadequate. Even under idealized random-error conditions, it would deliver approximately 1,600 out-of-specification results per year at the simulated workload. A 3-sigma process, though markedly better, still produces about 1 error every four days, underscoring the need for
either tighter imprecision goals or supplementary QC rules (e.g., Westgard multirules, moving averages) if this level of performance cannot be improved."
While it's a bit unusual to create a simulation of data points on QC chart based on a theoretical level of Sigma performance, the paper does give some interesting"feels." For example, 5 Sigma feels the same as 6 Sigma for the consumers of laboratory testing. That's partly a function of the volume of testing that occurs - if clinicians millions and millions of test results, they could see the difference between 5 and 6. Note, however, that doesn't mean 6 Sigma is no longer attractive. At 5 Sigma, you're using 3 Westgard Rules for QC. But at 6 Sigma, that drops down to just one rule.
Also useful to see that 2 Sigma is where problems start blowing up. 3 Sigma is acceptable (not great but you can struggle through it). But 2 Sigma is where things start blowing up, and on a regular basis.
There are a lot of conditions specific to this simulation, such as the use of the CLIA goal for sodium 4 mmol/L as the quality goal, and an assumption that bias is zero. But these "feels" are not confined to just the sodium method. Other methods with similar Sigma metrics will demonstrate the same outputs.