The Reference Interval That Does Not Fit
A community hospital laboratory is verifying a manufacturer's reference interval for serum creatinine on a new analyzer. The manufacturer's interval is 0.60–1.10 mg/dL for adults, derived from a mostly White, mostly male reference population at a large academic center. The local patient population is 38% Black, 30% Hispanic, and 52% female, with a high prevalence of chronic kidney disease and diabetes. The lab's verification protocol requires 20 reference individuals, but the lab has recruited only 14 volunteers who meet the exclusion criteria (no known kidney disease, no diabetes, not on nephrotoxic drugs, not pregnant). Of the 14, three have creatinine values above 1.10 mg/dL: 1.18, 1.22, and 1.31 mg/dL, all in middle-aged men. The laboratory director asks whether the interval can be adopted as-is.
1. Evaluate whether the manufacturer's interval can be adopted. Address the sample size and the three outlying values.
Read the worked answer
Adoption as-is is not defensible on this evidence. Two problems:
Sample size. The lab's own protocol calls for 20 reference individuals; only 14 were recruited. Fourteen is below the protocol requirement and well below the 120 individuals that CLSI-style guidelines recommend for establishing an interval de novo. Verification with a small sample is a weaker claim than establishment, and the lab should not treat a 14-person check as equivalent to a validated interval.
The three outliers. With n=14, three values above the upper limit is 21% - far more than the ~5% expected if the interval were correct for this population. That is a signal, not noise. Possible explanations: (a) the manufacturer's population genuinely differs from the local one (the manufacturer's cohort was mostly White and male; the local population includes groups with different creatinine distributions and a high CKD burden); (b) the exclusion criteria were applied inconsistently; (c) the analyzer is biased at the upper end. The lab should not simply discard the three values as "outliers" - that would be circular.
The correct position is: the interval is not verified, and the lab should either expand recruitment toward the protocol's 20 (and ideally more) or, if that is infeasible, establish a local interval or adopt a statistically defensible alternative with director sign-off and a documented rationale.
2. The director proposes simply widening the upper limit to 1.35 mg/dL to accommodate the three men. What is wrong with this approach, and what should be done instead?
Read the worked answer
Widening the upper limit to fit three observed values is fitting the interval to the data rather than deriving it from a reference population. Problems:
- Circularity. The purpose of a reference interval is to describe the central 95% of a healthy population. Setting the limit so that all 14 observed values fall inside guarantees that no value will ever be flagged, which defeats the purpose.
- Statistical invalidity. With n=14, the 95% interval cannot be estimated reliably; the upper reference limit is essentially the maximum observed value, which is highly unstable. Moving it to 1.35 is not a principled estimate.
- Clinical consequence. Raising the upper limit for creatinine means more patients with early kidney disease will be reported as normal. In a population with high CKD prevalence, that is a patient-safety issue, not a convenience.
- Provenance. A locally invented interval must be documented, justified, and communicated; an ad hoc widening with no derivation is not defensible in an inspection.
What to do instead: complete recruitment to at least the protocol's 20 (more if feasible), apply exclusion criteria rigorously and document them, and analyze the distribution - including whether the local population differs enough to warrant a locally established interval. If a local interval is established, it must be derived by an accepted statistical method (e.g., nonparametric central 95%), documented with the reference population description, and approved by the laboratory director. If the local population is genuinely different, the interval should reflect that - and the report should say so.