Disease prevalence matters
As explained elsewhere on this website the proportion of correct and incorrect classifications of healthy subjects and patients is co-determined by disease prevalence in the study population. On that account the predictive value of a test differs in principle strongly in general practice, in occupational medicine and in an outpatient or clinical population. This is exemplified by the following examples, in which the sensitivity and specificity are the same, and which relate to two populations comprising 1800 subjects, say population A in a hospital setting and population B in a GP setting.
|Clinically||Test result A||Test result B|
|+ Predictive value||80%||18.4%|
|- Predictive value||89%||99.9%|
The table demonstrates what seems a common sense observation: if very few people in a population have the disease for which you are testing, then a very large proportion of subjects will be misdiagnosed as having the disease (low + predictive value), and very few will be incorrectly identified as not having the disease (high – predictive value). Conversely, if a large proportion of the population has the disease, a significant proportion of those who are sick will still be missed, and a significant proportion will be incorrectly identified as sick.
Using the software and associated databases we can apply this to the general population of men and women in the Netherlands and the UK, as well as to the Dutch hospital population.
|LLN Criterion||GOLD criterion
|+ Predictive value||65.20%||81.17%||75.88%|
|- Predictive value||99.86%||99.03%||96.90|
Not surprisingly, if the disease is more prevalent (in the hospital population by a factor 2 based on the LLN criterion) the predictive value of a positive test result is greater.