Evidence based laboratory medicine for beginners: diagnostic accuracy studies and sources of bias

• Published in: Pathology Vol. 45; p. S15
• Main Author: Florkowski, Chris
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 2013
• ISSN: 0031-3025; 1465-3931
• DOI: 10.1097/01.PAT.0000426763.54496.5e

Diagnostic accuracy studies are typically cross-sectional studies comparing results of an index test with those of a reference standard in a consecutive series of patients for their ability to identify a condition of interest. From 2 × 2 tabulation of the results, it is possible to derive performance characteristics, including likelihood ratios and which may be further refined by ROC curve analysis. Bias may arise when a number of study conditions are not fulfilled.1 Spectrum bias is incurred when very sick patients are compared with healthy subjects. Differential verification bias is where subjects undergo different reference standards and partial verification bias is where not all subjects are submitted to the standard. Other sources of bias include selection bias, where non-consecutive subjects are recruited, lack of blinding and where there is inadequate description of the test or population. Emergence of the STARD criteria (standards for reporting of diagnostic accuracy)2 has provided a 25 point check-list to ensure rigour of diagnostic accuracy studies and has led to an improvement in the quality of such studies reported in the literature.