Because there's so much talk about #causalinference around here.
Computer scientists, economists, statisticians... talk a lot about the merits of #DeepLearning, instrumental variables, or whatever their preferred methodology is.
Need to compare the mortality of women under two dynamic screening strategies using a database of insurance claims with time-varying treatments and confounders.
How'd you do it?
4)
We specified a pragmatic #TargetTrial with sustained treatment strategies (dynamic screening strategies) and emulated it explicitly using #observational data from #Medicare.
We used cloning + censoring + #IPweighting to adjust for time-varying confounders.
And we concluded:
5)
Experts who are very active on Twitter:
Is our answer to this public health question wrong?
What could we have done better? Downplay #epidemiology principles and use more #AI or instrumental variables?
If so, please DO IT and let us know what you find.
Help us or shut up.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
1/ One day everyone will recognize #selectionbias due to a #collider and the world will be a better place.
This time observational studies found a higher risk of omicron reinfection after a 3rd dose of #COVID19 vaccine. As usual, alarms went off.
Can you see the obvious bias?
2/ Those who receive a booster and get infected are, on average, more susceptible to infection than those who don't receive a booster and get infected.
So no surprise than those who receive a booster and get infected are more likely to get reinfected.
1/ Our findings on a fourth dose (2nd booster) of the Pfizer-BioNTech #COVID19 vaccine are now published.
Compared with 3 doses only, a fourth dose had 68% effectiveness against COVID-19 hospitalization during the Omicron era in persons over 60 years of age.
@ProfMattFox 1/
The odds ratio from a case-control study is an unbiased estimator of the
a. odds ratio in the underlying cohort when we sample controls among non-cases
b. rate ratio in the underlying cohort when we use with incidence density sampling
No rare outcome assumption required.
@ProfMattFox 2/
Because the odds ratio is approximately equal to the risk ratio when the outcome is rare, the odds ratio from a case-control study approximates the risk ratio in the underlying cohort when we sample controls among non-cases and the outcome is rare.
But...
@ProfMattFox 3/
... for an unbiased estimator of the risk ratio (regardless of the outcome being rare), we need a case-base design, not a classical case-control design.
Of course, all of the above only applies to time-fixed treatments or exposures.