Many conditions for which patients are prescribed drugs ultimately lead to negative health outcomes. If a certain condition leads to the prescription of a drug and then causes a negative health outcome, the drug may appear responsible for this outcome to an observer comparing patients who did and did not receive a prescription in a database of patients’ medical records. This phenomenon is known as “confounding by indication.” We will present a method to reduce confounding by indication by restricting the study population to patients with just one “indication” (condition for which a drug is prescribed) that is unrelated to the outcome under investigation. We will demonstrate our method, called “extreme restriction”, using a case study assessing the effect of proton pump inhibitors (PPIs) on hospitalisation for community-acquired pneumonia (HCAP). In unrestricted analyses, PPIs will appear to cause HCAP because gastroesophageal reflux disease (GERD) leads to PPI prescriptions and HCAP. We will first demonstrate confounded analyses, comparing PPIs to five different comparator drugs unrelated to HCAP and to randomly-selected patients from the CPRD. We will then reduce confounding by restricting to non-steroidal anti-inflammatory drug (NSAID) users, some of whom receive PPIs prophylactically and are thus unlikely to have GERD.
We propose restriction to a single indication as a method to reduce confounding by indication in observational pharmacoepidemiologic research when common restriction approaches are insufficient. We will outline our method (called “extreme restriction”) and present its use in a case study of PPIs and the risk of HCAP. PPI use has been associated with an increased risk of pneumonia, but this association is likely the result of confounding by indication due to GERD. Using the CPRD linked to the Hospital Episode Statistics (HES) and Office for National Statistics (ONS) databases, we will estimate the effect of PPIs on HCAP in a series of retrospective cohort studies comparing incident PPI users to new users of active comparators (i.e., histamine-2 receptor agonists, H2RAs), new users of comparator drugs with no known pneumonia effects (i.e., treatments for glaucoma/ocular hypertension, osteoporosis, depression, and hypothyroidism), and randomly-selected candidates for PPI use. We will then repeat our analyses on an extremely restricted cohort of incident NSAID users, some of whom take PPIs prophylactically and are thus unlikely to have GERD. All analyses will use multivariate Cox proportional hazards models adjusted for deciles of high-dimensional propensity score (using >500 empirically-selected covariates) with exposures fixed at cohort entry.
Hospitalisation for community-acquired pneumonia
Samy Suissa - Chief Investigator - Sir Mortimer B Davis Jewish General Hospital
Kristian Filion - Corresponding Applicant - McGill University
Carla Doyle - Collaborator - McGill University
Colin Dormuth - Collaborator - McGill University
Laura Targownik - Collaborator - McGill University
Matthew Dahl - Collaborator - McGill University
Matthew Secrest - Collaborator - McGill University
Robert Platt - Collaborator - McGill University
Rui Nie - Collaborator - McGill University
Sophie Dell'Aniello - Collaborator - McGill University
HES Admitted Patient Care;ONS Death Registration Data