Background: The 1994 publication of one of the “Users’ Guides to the Medical Literature” in JAMA promoted using likelihood ratios (LRs) in diagnostic testing. Traditionally, sensitivity and specificity (SN/SP) were reported however LRs are more helpful because they can be used to go from a pretest to a posttest disease probability. SN/SP cannot be used directly to do this. Further, unlike SN/SP, judging whether a test is likely to be helpful is very straightforward using LRs. Predictive values (PVs) generally are of limited utility, as they depend on disease prevalence. We hypothesize that, because of all this, the proportion of manuscripts on diagnostic tests that report LRs has increased in recent years.
Methods: Review of published diagnostic test papers. Protocol: We searched PubMed using the Medical Subject Heading (MESH) term, “diagnosis”. We included all entries from five major emergency medicine journals and from other journals, but limited to papers with the MESH term, "Emergency Service, Hospital". We excluded non-English papers and those without a PubMed abstract. We first examined the abstracts and from those chose full papers that included at least one of LR, PV or SN/SP. We examined papers in 1997- 2000 (1997 alone had too few papers) and in 2017. We tallied the number of papers in which LR, PV and SN/SP were in the entire paper and, also, in the abstract alone because some clinicians read only the abstract. We calculated differences and 95% confidence intervals (CIs) between the two time periods.
Results: In 1997-2000 and 2017, the search yielded 537 and 668 papers, respectively, of which 68 and 78, met inclusion criteria. In 1997-2000, SN/SP, PV and LR were in 92%, 39% and 3% of the abstracts, respectively and in 99%, 71% and 19% of the papers, whereas in 2017, the corresponding values were 89%, 39% and 15% and 97%, 73% and 35%. From 1997-2000 to 2017, the reporting of LRs increased by 12% (95% CI, 2%, 22%) and 16% (95% CI, 1%, 29%) in the abstracts and papers, respectively. SN/SP and PV did not change significantly.
Conclusion: Reporting of LRs increased significantly over the last twenty years, although they are still included in only a minority of manuscripts on diagnostic tests. More needs to be done to educate clinicians about LRs and to encourage researchers, journal editors and reviewers to use them.