Skip to main page content

• HOME
• CURRENT ISSUE
• PAST ISSUES
• MOBILE
• ABOUT
• ACP
• SUBSCRIPTIONS
• E-MAIL ALERTS
• SUBMIT

RSS Feeds

1. Reporting meta-analyses of diagnostic accuracy studies: take a walk on the wild side

   • Levente Kriston, P.h.D.
   • Lars Hölzel, and Martin Härter

   University Medical Center Hamburg-Eppendorf, Germany

   TO THE EDITOR: In a recent meta-analysis we compared the diagnostic accuracy of two tests for detecting unhealthy alcohol use (1). We applied the currently recommended hierarchical summary receiver operating characteristics curve (HSROC) analysis to synthesize findings of primary research (2). However, we were confronted with uncertainties how to present our findings.

   First, for a comprehensive research report the choice of what should be reported is essential. Our analysis provided numerous results, from which a manageable amount had to be selected. Point estimates, such as pooled sensitivity and specificity, may be preferred by clinicians. Furthermore, under certain assumptions they can easily be converted into predictive values, which are even more familiar to several practitioners. But as the included studies applied different thresholds for defining test positivity, these estimates are affected by the primary authors’ subjective trade-off between sensitivity and specificity, which showed substantial heterogeneity. Therefore, we focused our interpretation mainly on the characteristics of summary receiver operating characteristics (SROC) curves.

   Second, to describe SROC curves for a general medical readership did not prove to be straightforward. It was unclear how to weight information on the statistical model, estimation procedure, numeric output, SROC curve forms, precision of the curves, crossing of two curves, and the extent of overlapping prediction regions. Our decision to report as much information as possible may satisfy biostatisticians but is likely to impair comprehensibility and ease of communication to a more general readership.

   Third, although the two investigated tests did show differences in their diagnostic accuracy, these differences rarely reached statistical significance, were frequently opposite for different measures, and the assessment of their clinical significance remained subjective. Due to missing reporting guidelines our way of presentation is only one of several possible alternatives, and it is unclear how it affects interpretation of the findings by the reader.

   In summary, we faced several challenges while presenting the results of a meta-analysis of diagnostic accuracy studies. In accordance with recent publications we feel that there is still more development needed in this area (3, 4). We welcome the work of the Cochrane Diagnostic Test Accuracy Working Group on a detailed guideline dealing with how to perform meta-analyses of diagnostic accuracy studies (4, 5) and hope that standardized reporting guidelines soon become available.

   References

   1. Kriston L, Hölzel M, Weiser A-K, Berner MM, Härter M. Meta-analysis: are 3 questions enough to detect unhealthy alcohol use? Ann Intern Med. 2008;149:879-88.

   2. Rutter CM, Gatsonis CA. A hierarchical regression approach to meta -analysis of diagnostic test accuracy evaluations. Stat Med. 2001;20:2865-84.

   3. Cornell J, Mulrow CD, Localio AR. Diagnostic test accuracy and clinical decision making. Ann Intern Med. 2008;149:904-6.

   4. Leeflang MM, Deeks JJ, Gatsonis C, Bossuyt PM; Cochrane Diagnostic Test Accuracy Working Group. Systematic reviews of diagnostic test accuracy. Ann Intern Med. 2008;149:889-97.

   5. Deeks JJ, Bossuyt PM, Gatsonis C (editors). Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy Version 1.0.0. Oxford, UK: The Cochrane Collaboration, 2009. [Forthcoming]. Avaliable from: http://srdta.cochrane.org/en/authors.html.

   Conflict of Interest:

   None declared

   Submit response
   Published January 13, 2009
2. One Question Enough?

   • David O. Parrish, MS MD

   Bayfront Family Medicine Residency

   I read with interest the AUDIT Meta-analysis article. Simply asking how long the patient has been an alcoholic can be useful. This puts the patient and physician in the position of determining how long, rather than whether, and allows planning for treatment to commence. This can be used in hospitalized patients where it is a bit easier and more appropriate than in an outpatient setting. The PSQT can be useful too.

   Reference:

   1. Parrish DO Single Question Screening for Problem Drinking. The Journal of Family Practice,

   2001. Vol. 50, No. 8

   Conflict of Interest:

   None declared

   Submit response
   Published December 30, 2008