Method Correlation Studies: A Practical How-To

Whenever a lab changes instruments, switches reagent manufacturers, or brings a new method online for a test it already performs, it needs to prove the new method agrees with the old one closely enough that patient results won't shift in a clinically meaningful way. That's a method correlation study — and doing it in a way that actually holds up to review comes down to a handful of design decisions made before the first sample is run.

When you need one

• Replacing or adding an analyzer for an existing test
• Switching reagent manufacturers or formulations
• Moving a test between two lab locations that will report interchangeably
• Introducing a modified version of an existing method

You generally do not need a full correlation study for routine reagent lot changes within the same manufacturer and method (that's covered by lot-to-lot QC verification instead) — but the line can blur, so when in doubt, document the decision either way.

Designing the study

Sample size. CLSI guidance (commonly referenced as EP09) recommends a minimum of 40 patient samples spanning the clinically relevant range, with more required if the assay has a wide reportable range or the acceptance criteria are tight. Fewer than 20–30 samples rarely produces a statistically defensible conclusion.

Sample selection. Samples should span the full reportable range, including values near clinical decision points — correlation at the extremes matters less than agreement right where a clinician would make a different call based on the result.

Timing. Run both methods on the same sample within a stability-appropriate window (often same-day) to avoid analyte degradation confounding the comparison.

Running the comparison

For quantitative assays, the standard approach is a Deming or Passing-Bablok regression (both account for measurement error in both methods, unlike simple linear regression) plus a Bland-Altman difference plot to visualize bias across the range rather than relying on a correlation coefficient alone.

A high correlation coefficient (r) is not the same as agreement — two methods can correlate strongly while one consistently reads 15% higher than the other. That's exactly what a Bland-Altman plot exposes and a raw r-value hides.

For qualitative assays, the comparison is simpler: percent positive agreement, percent negative agreement, and overall percent agreement, with every discordant result individually reviewed rather than folded into an aggregate percentage.

Setting acceptance criteria — before you look at the data

Define, in writing, what counts as acceptable agreement before running the comparison:

• Maximum allowable bias (often derived from CLIA proficiency testing acceptance limits or a total allowable error budget for the analyte)
• Regression slope and intercept ranges considered acceptable
• Maximum number of discordant qualitative results tolerated, and the review process for each one

Setting these after seeing the data — even with good intentions — invites the appearance of moving the goalposts, and it's one of the more common findings when correlation studies get reviewed closely.

Investigating discordant results

Every sample that falls outside acceptance criteria needs individual review: was it a specimen integrity issue, an interference, or a genuine method disagreement near a clinical cutoff? A discordant result that's explained and documented is fine. A discordant result that's silently excluded from the summary statistics is not.

The sign-off packet

• Study design and rationale (why this correlation study was needed)
• Pre-defined acceptance criteria
• Full data set (paired results, not just summary statistics)
• Regression and Bland-Altman analysis
• Discordant result review
• Director's dated approval before the new method goes live for patient reporting

Where this connects to validation, broadly

A method correlation study is one piece of a larger validation record — alongside precision, accuracy against a reference method, and reportable range verification. Keeping it as a standalone spreadsheet disconnected from the rest of the assay's validation history makes it hard to answer the question every reviewer eventually asks: "Show me everything that supports this method being clinically reportable, in one place."