Reference Range vs Personal Range

A population reference interval is the central 95% of results measured in a defined reference group, usually the 2.5th to 97.5th percentiles; a personal range is the distribution of your own repeated results around your own homeostatic set point. They answer different questions. The first asks whether you resemble the reference cohort. The second asks whether you resemble yourself last quarter.

How it works

A lab establishes a reference interval by recruiting reference individuals, applying exclusion criteria (pregnancy, known disease, certain medications), measuring the analyte on one specific instrument and reagent lot, and taking percentiles. CLSI guidance calls for at least 120 qualifying subjects per partition (per sex, per age band), which is why many published intervals are partitioned coarsely or not at all ¹. The interval is a property of that population on that method, not a biological constant. Intervals for the same analyte differ across regions and instruments, and locally derived intervals often diverge from the ones printed on imported test kits ².

Two consequences fall out of the definition. First, 5% of healthy reference subjects are outside the interval by construction. Run a 14-analyte panel of independent tests and the probability that every result lands inside is roughly 0.95^14, about 49%. A single flag on a broad panel carries very little information. Second, the interval only discriminates well when between-person variation is large relative to within-person variation. The index of individuality (within-subject CV divided by between-subject CV) captures this. For analytes with high individuality, such as serum creatinine, calcium, and many hormones, your personal set point occupies a narrow band inside a wide population interval, and you can move 40% from your own baseline while still reading “normal” ³.

A personal reference interval is built the other way: repeated measurements of you under standardized conditions, from which you estimate your own mean and your own within-person CV. Statistical treatments differ in how they handle analytical variation, non-normality, and drift, and in how many samples are needed before the estimate is stable ⁴.

In your own data

Results arrive as HL7 v2 OBX segments, FHIR Observation resources, or a CSV export. Keep these columns at minimum: LOINC code, value, unit, referenceRange.low, referenceRange.high, performing lab identifier, instrument or method when available, collection timestamp, and fasting status. Without the lab and method, a multi-year series is not comparable to itself.

What to check:

• Unit drift. Glucose in mg/dL versus mmol/L, ferritin in ng/mL versus µg/L (identical), creatinine in mg/dL versus µmol/L. Normalize on ingest, keep the original string.
• Reference interval changes mid-series. If referenceRange.high for ALT shifts from 40 to 33 U/L between draws, the lab changed method or adopted new criteria. Do not merge those segments without noting the break.
• Qualitative assays. This is what a “negative reference range” means: serologies, drug screens, and some autoantibody panels report Negative or Not Detected as the reference value, with no numeric interval. In FHIR these appear as valueCodeableConcept, not valueQuantity. Parsers that assume a numeric value silently drop them.
• Preanalytical conditions. Posture, tourniquet time, hydration, and time of day move several analytes more than the changes people go looking for. Standardize: same lab, morning, fasted, seated five minutes before the draw.

For deciding whether a new result differs from your baseline, compute the reference change value: RCV = 2^(1/2) × Z × sqrt(CV_A² + CV_I²), where CV_A is analytical imprecision (ask the lab, or take it from their method validation) and CV_I is your within-person CV. At Z = 1.96 this gives the two-sided 95% threshold. Anything smaller than the RCV is noise, whatever side of the population interval it falls on.

The common mistake is estimating CV_I from three draws taken across two labs. Use five or more from one lab, drop any sample taken during acute illness, and recompute as the series grows.

Limitations

Personal intervals require that your baseline was measured while you were in a stable state, which is unverifiable in retrospect. If a slow process was already underway, your personal range encodes it as normal. Critical appraisals of personalized intervals point to unsettled questions about sample size, handling of analytical drift, and whether the resulting intervals outperform population ones for any specific clinical decision ⁵. General practitioners presented with personal intervals raise a related concern: narrower personal bands can generate more flags and more follow-up testing without more diagnostic yield ⁶.

Neither kind of interval is a diagnosis. Reference intervals are not clinical decision limits, and for analytes where a fixed threshold exists, that threshold governs regardless of where your personal band sits ³. Take persistent out-of-range results, or large within-person shifts, to a clinician who can see the rest of the picture.

Woolf Software builds longitudinal molecular profiles of individuals: whole-genome sequencing, RNA sequencing, proteomics, blood biomarkers, and continuous glucose data, integrated into one model of you. Build your profile.