In examining a diagnostic test, we make the assumption that the characteristics of the test – its sensitivity and specificity (or likelihood ratios, the way I prefer to think) – will stay constant across different populations, although the positive and negative predictive values will change * . This is sort of true, and sort of false.
Take an imagined example: using leucocyte urine dipsticks to diagnose UTIs. If the original study is undertaken in the acute assessment department of a paediatric hospital, then the sticks will be used to differentiate between children with similar alternative diagnoses and conditions (e.g. viral infections, UTIs and avoiding school) . This is the disease spectrum. It might be that in the referred population of the assessment unit, 10% of febrile children have UTIs (that’s the prevalence).
Now the sensitivity and specificity that his study describes are likely to be true when used in a paediatric ED, or when seeing kids in primary care. Why? Because although the prevalence of UTI may vary (maybe 1% in primary care and 5% in ED’s), the other competing diagnoses are similar. The prevalence of the condition has changed, but the spectrum of illnesses seen hasn’t.
Now imagine using the sticks in the paediatric haematology-oncology ward, or outpatients. The proportion of kids with UTI may be similar to that in the ED, but here the spectrum is not the same. The competing diagnoses (e.g. neutropenic sepsis, chemotherapy-induced cystitis) will be very different. In this case, the spectrum change alters the expected sensitivity and specificity of the dipsticks.
So – take away sensitivity, specificity (or LRs) from diagnostic studies when the spectrum is similar regardless of the prevalence in your patients. If the spectrum differs – beware.
* Just a reminder: if the population has a lower prevalence of disease, then the positive predictive value will be lower (ie a positive test will be wrong more often) but the negative predictive value will be higher (ie a negative test will be right more often). The reverse is true if the prevalence of the disease is higher.