1239. Reporting Bias in Imaging Diagnostic Accuracy Research: Update on Recent Evidence
Authors * Denotes Presenting Author
  1. Robert Frank *; University of Ottawa
  2. Lindsay Cherpak; University of Ottawa
  3. Lee Treanor; University of Ottawa
  4. Anahita Dehmoobad Sharifabadi ; McMaster University
  5. Daniel Korevaar ; University of Amsterdam AMC
  6. Patrick Bossuyt; University of Amsterdam AMC
  7. Matthew McInnes; Ottawa Hospital Research Institute; University of Ottawa
Reporting bias occurs when research publication (publication bias), time-to-publication (time-lag bias) or citation rate (citation bias) is influenced by the magnitude or direction of findings. Here we summarize findings from several recent studies performed by our research group, evaluating reporting bias in imaging diagnostic test accuracy (DTA) research.

Materials and Methods:
The association of reported diagnostic accuracy (Youden index) and conclusion positivity with publication, time-to-publication, and citation rate, were determined for 405 abstracts of DTA primary research from the Radiological Society of North America meeting, 781 published primary DTA studies and 1016 published primary DTA studies, respectively.

Publication Bias: Multivariable logistic regression revealed an odds ratio of 3.6 [95%CI 1.9-6.7, p<0.001] for full text publication in favor of conference abstracts with positive (vs. negative/neutral) conclusions. Diagnostic accuracy showed no association with publication. Time-Lag Bias: Multivariable cox regression showed that positive conclusions were associated with shorter time from study completion to publication, with a hazard ratio (HR) of 1.32 [95%CI 1.03-1.69, p=0.030]; median time-to-publication was 7 months less for studies with positive vs. negative conclusions. Higher Youden index (YI) was also associated with shorter time to publication (rho=-0.11, p=0.009), with HR of 1.07 [95%CI 1.01-1.13; p=0.021] per unit increase in logit-transformed YI, independent of conclusion positivity; median time-to-publication was 1 month less for studies with YI above (vs. below) the median. Citation Bias: Negative binomial regression showed that positive conclusions were associated with higher citation rate (regression coefficient [r] = 0.19 [95%CI 0.03–0.35; p= 0.031]); mean citations per month was 0.54 for studies with positive conclusions vs. 0.34 for negative conclusions. A positive association between YI and citation rate (r = 0.35, p=0.011) was present, but not as strong when adjusting for conclusion positivity (r = 0.22, p=0.12).

Reporting bias exists in the imaging DTA literature. Studies with positive conclusions are published more often, published faster, and cited more often than those with negative or neutral conclusions. Studies with higher accuracy estimates are not more likely to be published; however, they are published more quickly and cited more often. The magnitude of effect is stronger for conclusions than for accuracy. Reporting bias in imaging DTA research can result in misinterpretation of imaging tests and adverse patient outcomes. Preferential and expedited publication of positive studies may lead to overestimation of imaging test accuracy. More frequent citation of positive findings may lead to biased inclusion in systematic reviews due to the recommended practice of searching cited references from included studies, and preferential citation in guidelines may bias the perspectives of test users. Therefore, clinicians and reviewers should consider reporting bias when interpreting and synthesizing literature.