Confidence at 100%: Characteristics of Likelihood Ratio Confidence Intervals in the Emergency Medicine Diagnostics Literature. (28th February 2020)

• Record Type: Journal Article
• Authors: Friedman, Ari B.; Berning, Aric W.; Marill, Keith A.
• Editors: Yarris, Lalena M.
• Is Part Of: Academic emergency medicine. Volume 27:Number 9(2020)
• Page Start: 897
• Page End: 904
• Publication Date: 2020-02-28
• Subjects: Emergency medicine -- Periodicals; 616.02505
• DOI: 10.1111/acem.13930
• Languages: English
• ISSNs: 1069-6563

Abstract

Abstract: Objective: We hypothesized that "perfect" 100% sample sensitivity or specificity (PSSS) is common in the emergency medicine (EM) literature. When results yield PSSS, calculating the likelihood ratio (LR) 95% confidence interval (CI) has been challenging. Consequently, we also hypothesized that studies with PSSS would be less likely to report the LR and associated CI, and those that did would use imperfect methods. Methods: We searched PubMed or Scopus for all articles reporting diagnostic test results in the 20 top EM journals from 2011 to 2016 and randomly sampled 124 articles. Trained researchers coded the articles as having PSSS or not ("controls"). We separately sampled 100 articles with PSSS and compared them to 100 controls in terms of their reporting of diagnostic tests and associated CIs. Results: Of the 124 articles, 19.4% (95% CI = 13% to 27.6%) feature a diagnostic test with PSSS. The LR is reported significantly less often in PSSS studies versus control studies: 18 of 100 articles (18% [95% CI = 11.3% to 27.2%]) versus 34 of 100 articles (34% [95% CI = 25% to 44.2%]), with an odds ratio (OR) of 0.43 (95% CI = 0.21 to 0.86). The LR 95% CI is also reported less often in PSSS versus control studies: five of 100 articles (5% [95% CI = 1.9% to 11.8%]) versus 27 of 100 articles (27% [95% CI = 18.8% to 37%]), with an OR of 0.11 (95% CI = 0.02 to 0.44). Five articles with perfect sample sensitivity reported their negative LR CI. The bootstrap method resulted in CIs that were 42.7% smaller on average (range = 16.6% to 63.6%). Conclusion: This analysis provides systematic evidence of diagnostic test reporting in the EM literature. Sample sensitivity or specificity of 100% is common. LRs and their associated 95% CIs are infrequently reported, particularly for PSSS samples. When the LR CI is reported in this scenario, it is overly wide. Improved reporting and methods can enhance the utility and confidence in diagnostic tests in EM.