Misleading authors' inferences in obstetric diagnostic test literature

OBJECTIVE: Our goal was to determine the validity of authors' inferences about the value of the cervicovaginal fetal fibronectin test in the prediction of preterm birth and the utility of uterine artery Doppler waveform analysis in the prediction of preeclampsia. STUDY DESIGN: We evaluated all 35 diagnostic test studies (14 on fetal fibronectin and 21 on uterine artery Doppler) included in 2 meta-analyses. The information on authors' conclusions regarding the value of a positive or negative test result was independently abstracted from each article by 2 reviewers, and it was classified as definitely useful, moderately useful, slightly useful, or not at all useful, For the "gold" standard, likelihood ratios of >10 and <0.1 were regarded as definitely useful, 5 to 10 and 0.1 to 0.2 were regarded as moderately useful, 2 to 5 and 0.2 to 0.5 were regarded as slightly useful, and 1 to 2 and 0.5 to 1 were regarded as not at all useful. The agreement between the authors and the reference standard was computed by simple percentage agreement and weighted kappa statistic. RESULTS: Among articles assessing the diagnostic value of fetal fibronectin the simple agreement between the authors and the "gold" standard was 26% (7/26) with a kappa of 0.05 (P = .83), and authors overestimated the value of the test result in 66% (17/26) of instances. Similarly, among articles assessing uterine artery Doppler the simple agreement between the authors and the "gold" standard was 31% (13/42) with a ic of 0.28 (P = .31), and authors overestimated the value of the test result in 48% (20/42) of instances. CONCLUSION: Authors claimed more positive conclusions than could be supported by their data. When studies are reported in a misleading manner, the chance of misinterpretation on the part of the clinical reader is increased. The use of explicit criteria that are based on likelihood ratios may reduce the risk of erroneous inferences.