Previous studies have found evidence suggesting paracetamol, as compared to ibuprofen, increases the risk of several health conditions, namely stomach ulcers, heart attack, stroke, and kidney failure. Given an apparent relationship between paracetamol and subsequent adverse health events, one question is whether paracetamol causes these conditions or whether it reflects the characteristics of patients before getting the drug. An important possible reason for the latter pattern occurs when patients are either much more or much less likely to be prescribed certain medications because of a pre-existing health condition. An existing health condition may cause physicians not to prescribe a medication because of side effects associated with the condition. People with a stomach ulcer might not be prescribed ibuprofen because it can cause stomach bleeding and they are already at risk. Instead, they would get paracetamol for mild pain and would likely continue treatment for their ulcer. Then the use of paracetamol would appear related to subsequent stomach ulcer, too. The objective of this study is to determine whether there is a relationship between paracetamol versus ibuprofen and serious health events after adjusting for patient characteristics before getting the drug.
In a prior study, we observed evidence of channeling bias in paracetamol versus ibuprofen prescription. We developed an adjustment for channeling bias using outcomes known not to be causally associated with adverse outcomes of either medicine. We could do this since the exposure and outcome should not be associated. For this study we use the new method to assess whether adverse outcomes in the published literature are associated with paracetamol versus ibuprofen.
In users first prescribed paracetamol alone or ibuprofen alone, we will use large scale propensity score matching to obtain relative risks of interest (myocardial infarction, stroke, gastrointestinal bleeding or renal disease) after paracetamol versus ibuprofen prescription. Negative controls will be used to assess potential bias and to calibrate p-values. Positive controls will used for confidence bound estimates.
Two propensity score models will be used. One based on models typical in the literature using hand-picked covariates. The other a model including all possible (1,000 +) covariates in the database. One-to-one matching in on-treatment Cox models of for the outcomes of interest and negative controls will be implemented. Visualizations of risk ratios versus standard errors will give traditional and calibrated regions for assessing associations.
Myocardial Infarction
- Stroke
- GI Bleeding
- Renal Disease
Rachel Weinstein - Chief Investigator - Johnson & Johnson ( JnJ - USA )
Rachel Weinstein - Corresponding Applicant - Johnson & Johnson ( JnJ - USA )
Daniel Fife - Collaborator - Johnson & Johnson ( JnJ - USA )
Jesse Berlin - Collaborator - Ortho-McNeil-Janssen Pharmaceuticals - Johnson & Johnson Company
Joel Swerdel - Collaborator - Janssen US
Martijn Schuemie - Collaborator - Janssen US
Patrick Ryan - Collaborator - Janssen Research & Development LLC