OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George...
Transcript of OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George...
![Page 1: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/1.jpg)
OHDSI Methods for Causal Effect Estimation
George Hripcsak, David Madigan, Patrick Ryan, Martijn Schuemie, Marc Suchard
“The sole cause and root of almost every defect in the sciences is this: that whilst we falsely admire and extol the powers of the human mind, we do not search for its real helps.”
— Novum Organum: Aphorisms [Book One], 1620, Sir Francis Bacon
http://www.ohdsi.org
![Page 2: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/2.jpg)
When comparisons were made between NOACs, matched rivaroxaban patients had a significantly higher risk of major bleeding (HR: 1.82; 95 % CI: 1.36–2.43) compared to apixaban patients.
![Page 3: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/3.jpg)
Reliability: Analysis Ignores…• Selection bias• Measurement error• Model misspecification• Multiple modeling• Unmeasured confounding
“Grave errors are commonplace, perhaps typical. It does no good to append a claim that you have included in the regression all relevant covariates, a claim that there are no unmeasured confounders and that you could not be mistaken in making this claim. Who are you that you could not be mistaken?”
- Paul Rosenbaum
![Page 4: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/4.jpg)
Observational research results in literature
85% of exposure-outcome pairs have p < 0.05
29,982 estimates11,758 papers
![Page 5: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/5.jpg)
• Reproducible, systematized, open source approach at scale
• Negative controls– Drugs and outcomes “known” to have no causal
association– Literature, product labels, spontaneous reports– Empirical p-values
• Positive Controls– Inject signals onto negative controls with known effect size– Calibrated confidence intervals
A New Approach
![Page 6: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/6.jpg)
![Page 7: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/7.jpg)
Observational research results in literature
29,982 estimates11,758 papers
![Page 8: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/8.jpg)
Calibration Assumptions
8
![Page 9: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/9.jpg)
Many models, many databases
9
Combining calibration with random effects meta-analysis and BMA
![Page 10: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/10.jpg)
Method: Study design (LEGEND)
10
Treatment strategies:• Atenolol• Nebivolol
Causal contrasts of interest:• On-treatment effect• Intent-to-treat effect
Atenolol Atenolol
Nebivolol Nebivolol
PS matching
Follow-up timeMedical history
Eligibility criteria:• Diagnosis of hypertension
during previous 1 year• No prior antihypertensive
drug• No prior cardiovascular
outcome
Outcome (Major Adverse Cardio-Cerebrovascular Event):• Hospitalized myocardial infarction, heart failure, stroke
and sudden cardiac death
https://github.com/OHDSI/LEGEND
![Page 11: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/11.jpg)
Cohort Methods
11
![Page 12: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/12.jpg)
Other Methods: Case Control
12
![Page 13: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/13.jpg)
Other Methods: SCCS
13
![Page 14: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/14.jpg)
Gold Standard Performance: CCAE
![Page 15: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/15.jpg)
Gold Standard Performance: CCAE
![Page 16: OHDSI Methods for Causal Effect EstimationOHDSI Methods for Causal Effect Estimation George Hripcsak, David Madigan, Patrick Ryan, MartijnSchuemie, Marc Suchard “The sole cause and](https://reader035.fdocuments.in/reader035/viewer/2022071021/5fd5e28aaca3cb5f815577ee/html5/thumbnails/16.jpg)
Concluding thoughts
• An international community and global data network can be used to generate real-world evidence in a secure, reliable and efficient manner
• Common data model critically important• Much work remains on establishing (and
improving) actual operating characteristics of current approaches to causal inference
OHDSI: Join the journey