The variation of diagnostic accuracy across different healthcare settings

Natasja Vijfschagt, Huibert Burger, Michiel De Boer, Marjolein Berger, Ann Van Den Bruel, Mariska Leeflang, Thomas Fanshawe, Gea Holtman

Keywords: Sensitivity, Specificity, Non-referred settings, Referred settings

Background:

Tests developed in specialist care may have different accuracy when applied to primary care. Sensitivity and specificity may vary due to selection that occurs before referral to specialist care, but extensive studies evaluating this claim are lacking. We reanalyzed meta-analyses that included studies of diagnostic test accuracy in primary and specialist care to investigate variation between settings.

Research questions:

What is the variation in diagnostic accuracy between non-referred and referred settings?

Method:

We systematically searched EBSCOhost MEDLINE for meta-analyses that included at least 10 original studies of the same diagnostic index test. Meta-analyses with at least three studies done in primary care and at least three studies done in specialist care were included in the re-analysis. We excluded studies without a clear setting definition. Firstly, we distinguished non-referred (primary care, community, other) and referred settings (specialist care, other), and secondly, we restricted to studies performed in countries in which primary care has a gatekeeping function. We constructed bivariate random-effects models with setting as a covariate.

Results:

We included nine meta-analyses evaluating 12 index tests; signs and symptoms (n=7), questionnaire (n=1), rapid tests (n=2), fecal test (n=1), and ultrasound (n=1). In 8 of 12 index tests sensitivity was higher in non-referred versus referred setting, with a range of 0.05 (abdominal pain for colorectal cancer) to 0.21 (rapid influenza diagnostic tests for Influenza infection), and in 10 of 12 index tests specificity was lower in non-referred setting, with a range of 0.01 (4 index tests) to 0.31 (paroxysmal cough in pertussis). Analyses limited to gatekeeping health care systems showed similar results.

Conclusions:

In a majority of the index tests there is a difference between health care settings, in which for most tests the sensitivity was higher and specificity was lower in a non-referred setting. More research is needed into explaining the variation in diagnostic accuracy between settings.

Points for discussion:

Did you expect that the sensitivity would be higher and the specificity would be lower in non-referred settings?

Which mechanisms can explain the variation?

How can we make these results applicable in general practice?

#40