Article Text
Abstract
In this two-part series on sources of bias in studies of diagnostic test performance, we outline common errors and optimal conditions during three study phases: patient selection, interpretation of the index test and disease verification by a gold standard. Here in part 1, biases associated with suboptimal participant selection are discussed through the lens of partial verification bias and spectrum bias, both of which increase the proportion of participants who are the ‘sickest of the sick’ or the ‘wellest of the well.’ Especially through retrospective methodology, partial verification introduces bias by including patients who are test positive by a gold standard, since patients with a positive index test are more likely to go on to further gold standard testing. Spectrum bias is frequently introduced through case–control design, dropping of indeterminate results or convenience sampling. After reading part 1, the informed clinician should be better able to judge the quality of a diagnostic test study, its inherent limitations and whether its results could be generalisable to their practice. Part 2 will describe how interpretation of the index test and disease verification by a gold standard can contribute to diagnostic test bias.
- imaging
- research, methods
- statistics
- ultrasound
Statistics from Altmetric.com
Footnotes
Contributors MKH, BK and RW all conceived the review article and generated the original text. All authors drafted the article and contributed substantially to its revision. MKH takes responsibility for the paper as a whole.
Funding BK was supported by NHLBI K08 (grant number 1K08HL140105-01). MKH was supported by NASA Glenn Research Center and ZIN Technologies; NIDDK (grant number P01DK043881); and Emergency Medicine Foundation. RW was supported by UCSF–Learning Healthcare Systems Award.
Competing interests BK is site investigator for Ortho-Clinical Diagnostics.
Provenance and peer review Not commissioned; externally peer reviewed.
Patient consent for publication Not required.