Slide 1: Bias in diagnostic studied
Slide 2: Question: you are considering offering fecal immunochemical testing (FIT) for a patient who is reluctant to undergo colonoscopy for colon cancer screening and want to learn more about how well this test preforms.
You read a study that reports a FIT testing sensitivity of 95% and specificity of 97% for colorectal cancer. In the study, all participants underwent FIT testing. Colonoscopy was then preformed only in those with a positive FIT. What type of bias is likely to be present?
A. Spectrum, B. Verification, C. Validation, D. Lead time, E. Ascertainment.
Slide 3: B. Verification bias. Verification bias arises in determining the performance of an index test: when the reference standard (gold standard, e.g. colonoscopy) is not use for all cases (known as partial verification). FIT negative, no colonoscopy. FIT positive, colonoscopy.
when there is more than one reference standard (known as differential verification). Biomarker negative, clinical follow-up. Biomarker positive, pathology.
Slide 4: Verification bias can occur when: only selected patients get the reference standard, such as only those with: a positive test, high pretest probability, or an unrelated indication.
Performing the reference standard in all patient: may be unethical (e.g., not all patients with liver disease need a biopsy!), sometimes cannot be performed if negative result of index test (e.g., how can you biopsy a negative PET?).
The reference standard is burdensome (such as colonoscopy) and some patients are lost to follow up).
Slide 5: How does partial verification affect test characteristics?
Patients with a negative index test are less likely to receive the reference standard, resulting in a lower number of false negatives and true negative. FIT positive, colonoscopy, polyp. FIT negative, no colonoscopy, unverified disease status.
Recall that sensitivity is the proportion of patient with the disease who are detected by the test, or true positive divided by false negative plus true positive. Specificity is true negative divided by true negative plus false positive.
Slide 6: Decreased false negative will increase sensitivity. Decreased true negative will increase specificity. So in general, studies with partial verification will actually make the test seem more sensitive and less specific than it actually is.
Graph showing true negatives, true positives, false negatives, and false positives. Fewer negatives so the bell curve is smaller for true negatives.
Slide 7: Mitigating verification bias. When appraising a study, ask: was the reference test performed in all participants if possible? If not, are there some statistical methods that correct for verification bias?
Some commonly used correction methods: The Begg and Greenes method extrapolates the results of participants who look the reference to those who didn’t. Multiple imputation, like multiple regression, adjusts for the covariates that make a patient more likely to undergo the reference test.
References
- Lijmer JG, Mol BW, Heisterkamp S, Bonsel GJ, Prins MH, van der Meulen JH, Bossuyt PM. Empirical evidence of design-related bias in studies of diagnostic tests. JAMA. 1999 Sep 15;282(11):1061-6. PMID 10493205.
- de Groot JA, Bossuyt PM, Reitsma JB, Rutjes AW, Dendukuri N, Janssen KJ, Moons KG. Verification problems in diagnostic accuracy studies: consequences and solutions. BMJ. 2011 Aug 2;343:d4770. PMID 21810869.
- Day E, Eldred-Evans D, Prevost AT, Ahmed HU, Fiorentino F. Adjusting for verification bias in diagnostic accuracy measures when comparing multiple screening tests – an application to the IP1-PROSTAGRAM study. BMC Med Res Methodol. 2022 Mar 18;22(1):70. PMID 35300611.
Tags: stats with core im, verification bias