Thread by @cap1024

Now peered reviewed at @IntJEpidemiol: challenges (& a solution) for the test-negative study design when data are gathered during public health response in a population with clustered vaccine uptake (work w/ @TJHladish, W. John Edmunds, & @rozeggo)

academic.oup.com/ije/advance-ar…

Briefly, this design is biased when mixing different testing data streams-e.g. tests for sick people & tests for *contacts* of sick people. That bias is exacerbated when vaccine coverage is homophilous-i.e. if you're vaccinated, likely your contacts are too & vice versa.

This work focuses on Ebola outbreaks, where e.g. vaccine coverage might be logisitically constrainted to particular areas.

But, it's likely applicable to COVID-19: lots of testing, but limited documentation of why, and pro/anti-vax social grouping (& age-based, during rollout).

The solution? Collect the reason for testing!

With that, the design can be stratified or modified (we proposed a hybrid test-negative / cohort design) by testing reason and the bias due to the issue we analyzed vanishes (though others certainly remain).

Hopefully, this insight will influence future planning for public health responses to enable better real-time assessment of interventions when trials may not be practicable.

@threadreaderapp unroll please!