Identifying relevant diagnostic studies in MEDLINE. The diagnostic value of the erythrocyte sedimentation rate (ESR) and dipstick as an example.

OBJECTIVE

We aimed to examine sensitivity and positive predictive value of MEDLINE searching for diagnostic studies, relevant for the primary health care setting.

METHOD

Results of MEDLINE searches were compared with a reference standard collection of studies on two subjects, the diagnostic value of ESR in discriminating between 'pathology' and 'no pathology', and the dipstick method in diagnosing urinary tract infections. The main outcome measures were sensitivity (proportion of the total number of reference standard diagnostic studies that could be identified by the search) and positive predictive value (proportion of the total number of publications retrieved by MEDLINE that were incorporated in the reference standard).

RESULTS

The combined MeSH and freetext search was more sensitive than MeSH term searching only, for both the ESR and the dipstick search. With this combined search sensitivities of 0.91 and 0.98 and predictive values of 0.10 and 0.68 were found for ESR and dipstick respectively. By restricting the search with keywords describing the primary health care setting the predictive values increased to 0.72 and 1.00 but sensitivity dropped to 0.10 and 0.07 (ESR and dipstick respectively).

CONCLUSION

Combining freetext and MeSH term searching, without restriction to the primary health care setting, is a valuable strategy in systematically searching for available evidence on the value of a diagnostic test in the scope of a specific disease. The predictive value seems to depend on the breadth of the disease area. MEDLINE should provide a term such as 'diagnostic evaluation study' to be used in the limit field Publication Type to specify diagnostic studies.