Reference ranges are defined statistically, not clinically
Expansion
How reference ranges are made: the central 95 per cent of results from a healthy reference population, so by construction 5 per cent of healthy people fall outside it, 2.5 per cent at each end. “Abnormal” therefore means unusual, not diseased.
The multiple testing problem: the chance that all of n independent tests are normal is 0.95 to the power n.
| Tests ordered | Chance of at least one abnormal |
|---|---|
| 1 | 5 per cent |
| 10 | 40 per cent |
| 20 | 64 per cent |
| 40 | 87 per cent |
This is the arithmetic behind the advice not to order panels of tests without a question.
Other reasons a result may not mean what it appears to
- Population specificity: ranges vary with age, sex, ethnicity, pregnancy and altitude, and the laboratory’s range may not match the patient
- Analytical variation: assay imprecision, interference from biotin, haemolysis, lipaemia, paraproteins and heterophile antibodies
- Biological variation: diurnal (cortisol, testosterone, iron), postural (renin, aldosterone), postprandial (glucose, triglycerides), menstrual cycle
- Pre-analytical error: wrong tube, tourniquet time, delayed transport, wrong patient
The clinically useful questions
- Is the result plausible given the patient? If not, repeat it before acting
- Is it different from the previous value by more than the assay and biological variation?
- Would the management change whatever the answer?
Incidental findings are the imaging equivalent, and follow the same logic: the more you look, the more you find, and most of what you find in an asymptomatic person did not need finding.