Prevalence - Limitations

Limitations

It can be said that a very small error applied over a very large number of individuals (that is, those who are *not affected* by the condition in the general population during their lifetime; for example, over 95%) produces a relevant, non-negligible number of subjects who are incorrectly classified as having the condition or any other condition which is the object of a survey study: these subjects are the so-called false positives; such reasoning applies to the 'false positive' but not the 'false negative' problem where we have an error applied over a relatively very small number of individuals to begin with (that is, those who are *affected* by the condition in the general population; for example, less than 5%). Hence, a very high percentage of subjects who seem to have a history of a disorder at interview are false positives for such a medical condition and apparently never suffered a fully clinical syndrome.

A different but related problem in evaluating the public health significance of psychiatric conditions has been highlighted by Robert Spitzer of Columbia University: fulfillment of diagnostic criteria and the resulting diagnosis do not necessarily imply need for treatment.

A well-known statistical problem arises when ascertaining rates for disorders and conditions with a relatively low population prevalence or base-rate. Even assuming that lay interview diagnoses are highly accurate in terms of sensitivity and specificity and their corresponding area under the ROC curve (that is, AUC, or area under the receiver operating characteristic curve), a condition with a relatively low prevalence or base-rate is bound to yield high false positive rates, which exceed false negative rates; in such a circumstance a limited positive predictive value, PPV, yields high false positive rates even in presence of a specificity which is very close to 100%.