Type of bias | Recognising bias | Effect on accuracy | ||
Partial verification workup or referral bias | Only patients tested with gold standard are included; patients with positive index test are more likely to get gold standard. | Falsely increases sensitivity by lowering rate of false negatives | ||
Part 1: Suboptimal patient selection | Spectrum bias through case–control design | Inclusion of ‘sickest of the sick’ or ‘wellest of the well’ | Falsely increases sensitivity and specificity | |
Spectrum bias through dropping indeterminate subjects | Ask ‘did they describe their method for handling indeterminates?’ | Falsely increases sensitivity if excluded indeterminates have mild disease. Falsely increases specificity if excluded indeterminates are not diseased. | ||
Spectrum bias through convenience sampling | Look for screening modality in methods section. | Falsely elevates sensitivity and specificity when sampling excludes difficult, indeterminate or ambiguous patients. | ||
Part 2: Diagnostic and verification test | Interpretation | Indeterminate | When indeterminate results are considered dichotomously as positive or negative. | It can overestimate or underestimate the accuracy of the test depending on how indeterminates are included. |
Review | Occurs when the person interpreting the diagnostic test has access to the gold standard test. | Usually falsely increases the sensitivity and specificity of the index test. | ||
Verification | Incorporation | Occurs when the criteria for a gold standard include the results of the diagnostic test | Falsely increases the sensitivity and specificity of the index test. | |
Double gold standard (differential verification) | Occurs when gold standard test is invasive or expensive, and is only performed when index test result is positive. | Falsely increases the sensitivity and specificity of the index test. |
Bold text denotes the biases discussed in this manuscript, ‘Recognising Bias in Studies of Diagnostic Tests Part 1: Suboptimal Patient Selection’ 1. Greyed out text denotes the biases discussed in ’Recognising Bias in Studies of Diagnostic Tests Part 2: Interpreting and Verifying the Index Test'.