The reflex to indiscriminately stratify and adjust on covariates is strong in biomedical research. I face this so often, especially with clinician-scientist collaborators.
The analyses and models that arise from this, in fact, often do not well represent how exposures and treatments exert their effect in reality.
A #TargetTrial approach crystallizes when, where, and how co-variates should be treated.
Take this great illustration from Target Trial originator @MiguelHernan where he explains the study design choices for study on COVID vaccine booster effectiveness:
https://fediscience.org/@MiguelHernan/109331025655365201
#epidemiology #biostatistics #epitwitter #statstodon #epiVerse @epiVerse
Miguel Hernan (@[email protected])
Attached: 1 image 1/ Using observational data, we estimated the short-term effectiveness of a 4th dose of #COVID19 vaccine: 68% for hospitalization, 74% for death in persons over 60 (compared with 3 doses) https://nejm.org/doi/10.1056/NEJMoa2201688 And then we received an interesting criticism: "You overestimated short-term effectiveness against hospitalization/death because only people infected with #SARSCoV2 during follow-up can be hospitalized or die due to #COVID19 and you didn't restrict the analysis to infected people."