Thread by @COREIMpodcast

1/ Today on #StatswithCoreIM, let’s talk about #bias

2/ Let’s get started with a question: What type of bias do you think is present in this example?

3/ The answer is Verification bias! Verification bias is present when:

✔️The reference standard (gold standard) is not used for all cases [= emoji] partial verification

✔️there is more than one reference standard 🟰 differential verification

4/ Think about the different scenarios where #verificationbias may occur:

5/ How does partial verification affect test characteristics?

Negative index test → less likely to receive reference standard→ ⬇️number of false negatives and true negatives

How will that affect the sensitivity and specificity?

6/ Putting this all together, if you have a lower # of false negatives, the sensitivity goes⬆️

and a lower # of true negatives means the specificity goes ⬇️

Therefore, studies with partial verification will make the test seem MORE sensitive and LESS specific

7/ To sum up mitigation of verification bias, ask these Qs

Was the reference standard performed in ALL participants?
If not are there some statistical methods that correct for this verification bias?

Consider:
Begg and Greenes extrapolation
Multiple imputation for covariates

8/ Thanks everyone and have a great day. Remember, trust but verify :)

Author: Robert Wharton
Graphics: @ivannatang
Link on CoreIM: coreimpodcast.com/2022/08/09/ver…