Transcript of "Groenwold causality leiden - 5 oct2012"
Causality in pharmacoepidemiologic research what are the challenges? Rolf H.H. Groenwold, MD, PhD
Wikipedia:“Pharmacoepidemiology is the study of the use of and the effects of drugs in large numbers of people”“Pharmacoepidemiology is the bridge between pharmacology and epidemiology”“Pharmacoepidemiology studies the effects ofdrugs”http://en.wikipedia.org/wiki/Pharmacoepidemiology
Randomized trials….... are the paradigm to assess the effects of drug,yet…… are very expensive… require a lot of men power… answer (at most) only 1 question… are little informative about (rare) side-effects… have limited duration of follow-up
Non-randomized trials….… are MUCH LESS expensive… require LESS men power… answer MORE than 1 question… are MORE informative about (rare) side-effects… have ENDLESS duration of follow-up, yet…... are NOT the paradigm to assess the effects ofdrug, because …
An example of influenza vaccination“Studies naar het effect van vaccinatie tegen griep deugen vaak methodologisch niet.” (www.nos.nl, 24-10-2011)“Effectiviteit griepprik nooit bewezen” (Elsevier, 24-10-2011)
ConfoundingThe everlasting challenge in pharmacoepidemiology- can be addressed by comparing like with like - E.g. compare treated men with untreated men and treated women with untreated women- all methods to control for confounding are based on this principle
Some challenges:1. (non-)positivity2. Dynamic treatment3. Validity and completeness of registries
(non-)positivity – example Study of OC use in medical students: Oral contraceptive No oral Confounder use contraceptive use Females 95% 5% Males 0% 100% Study of β-blocker use in adults β-blocker use No β-blocker use Confounder Hypertension 30% 70% No hypertension 0.5% 99.5%
1. (non-)positivity in pharmacoepi• non-positivity shortly after drug comes to the market (selective groups switch treatment)• second choice drugs (non-respondents switch treatment)• strict indication (yet unrecorded)
2. Dynamic treatment in pharmacoepiTreatment is often not a constant• time-varying dosage• on demand treatment (e.g. SABA, NSAIDs)• ‘ever vs never’ or dynamic regimes (where does our interest lie?)
3. Registries in pharmacoepi Validity and completeness of registries • prescription ≠ actual drug use • only measured confounders can be adjusted • time-varying confounding (how frequent have confounders been measured?) Shouldn’t we anticipated future research in designing our registries? (which are key confounders?)
Causality in pharmacoepidemiologyPharmacoepidemiology is an ideal playing field tostudy methods for causal inference.A lot of opportunities and challenges.