Spectrum bias
- Bias
A diagnostic test looks better than it is when it is tried only on very sick and clearly healthy people, not on patients like those seen in clinic.
What it is
Spectrum bias arises when a diagnostic test is evaluated in patients who do not represent those in whom it will be used. If the sick are advanced cases and the non-sick are healthy volunteers, the test tells them apart easily and its sensitivity and specificity come out inflated. In practice, the sick have mild or early disease and the non-sick have similar symptoms or other diseases, and the test performs worse.
How it arises
- Clear cases (often severe) and symptom-free healthy people are recruited separately.
- The differences between the two groups are extreme and the test detects them easily.
- In clinic, borderline cases appear that the test classifies less well.
Worked example
- Early study of a rapid blood test: 100 inpatients with severe infection and 100 healthy donors.
- Positives: 95 of 100 sick (sensitivity 95%); negatives: 98 of 100 healthy (specificity 98%).
- Clinic study: 500 consecutive patients with compatible symptoms; 200 have the infection (many mild) and 300 do not.
- Positives: 150 of 200 sick → sensitivity = 150 / 200 = 75%.
- Negatives: 255 of 300 non-sick → specificity = 255 / 300 = 85%.
- Reading: in real patients the test loses 20 points of sensitivity and 13 of specificity; the first study had spectrum bias.
Markers
- Sensitivity
- Specificity
How to prevent or control it
- When reading a diagnostic accuracy study, check whether patients were consecutive and had the real clinical suspicion.
- Prefer studies done in a setting like your own (primary care versus referral hospital).
Common pitfalls
- Using in primary care the sensitivity obtained in severely ill inpatients.
- Confusing it with the effect of prevalence: spectrum bias changes sensitivity and specificity; prevalence changes predictive values.
Test yourself
Think of the answer, then tap the question to see it.
What happens to sensitivity and specificity if a test is evaluated in severely ill patients and healthy volunteers?
They come out inflated, because the differences between the two groups are extreme and easy to detect.
How is spectrum bias avoided?
By evaluating the test in a consecutive series of patients with the real clinical suspicion, in the setting where it will be used.
How does it differ from the effect of prevalence?
Spectrum bias alters sensitivity and specificity; a change in prevalence mainly alters the predictive values.
Connections
Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.
Sources: Spectrum bias | Catalog of Bias (Catalogue of Bias Collaboration, 2019); Verification bias | Catalog of Bias (Catalogue of Bias Collaboration, 2019).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.