Reporting bias in imaging research: association of diagnostic accuracy estimates in radiology conference abstracts with full-text publication




Poster session 1


Sunday 16 September 2018 - 12:30 to 14:00

All authors in correct order:

Cherpak L1, Korevaar D2, McGrath T3, Dang W1, Walker D1, Dehmoobad Sharifabadi A3, Salameh JP3, McInnes M1
1 Department of Radiology, University of Ottawa, Canada
2 Department of Clinical Epidemiology, Biostatistics and Bioinformatics, Academic Medical Center, The Netherlands
3 University of Ottawa, Canada
Presenting author and contact person

Presenting author:

Lindsay Cherpak

Contact person:

Abstract text
The impact of reporting bias in therapeutic interventional studies is well documented. However, there is less evidence for diagnostic accuracy studies. The associations between accuracy estimates and lack of publication in imaging research are not well understood. If higher accuracy estimates in imaging studies are associated with a higher probability of or shorter time to publication, this could lead to overestimates of diagnostic accuracy in systematic reviews and have negative downstream consequences for patient care.

To evaluate the associations between accuracy estimates and full-text publication at five years after abstract submission.

We evaluated diagnostic accuracy abstracts presented at the Radiological Society of North America meeting in 2011-2012. We used specificity (Sp) and sensitivity (Sn) from the abstracts to calculate Youden's index (YI = Sn + Sp -1). To identify full-text publications within five years after abstract submission we carried out both database searches and follow-up with authors. We used T tests to assess the associations between accuracy estimates (logit-transformed) and full-text publication.

We included 405 abstracts (256 from 2012 and 149 from 2011). The median YI was 0.79 (interquartile range 0.63 to 0.90). We identified full-text publications for 287/405 (71%) abstracts. The mean YI for abstracts with full-text at five years (0.76, 95% confidence interval (CI) 0.74 to 0.78) was significantly higher than for those without (0.72, 95% CI 0.68 to 0.76); P = 0.04.

Study abstracts that report higher diagnostic accuracy estimates were more likely to be reported in a full-text publication five years after initial submission. Higher probability of full-text publication for higher accuracy estimates could lead to overestimates of diagnostic accuracy by physicians, systematic reviews and guidelines.

Patient or healthcare consumer involvement:
There were no patients or healthcare consumers involved in this project.


Relevance to patients and consumers: 

Reporting bias in diagnostic accuracy research can lead to overestimates of diagnostic accuracy by physicians, systematic reviews and guidelines. This can have a detrimental impact on patients since the proportion of incorrect tests (false negative and false positive cases) will be underestimated and could lead to sub-optimal patient outcomes. Patient/consumer partners did not contribute to this statement.