(149) Inverse Probability Treatment Weighting Versus Propensity Score Matching in The Sentinel System: A Test Case

Friday, August 26, 2022

8:00 AM – 6:00 PM CEST

Location: Hall C

Publication Number: 480

Presenting Author(s)

Marie C. Bradley, PhD
– Dr, US Food and Drug Administration, Silver Spring, Maryland, United States

Co-Author(s)

Talia Menzin
– Department of Population Medicine, Harvard Pilgrim Health Care Institute and Harvard Medical School, Boston, MA, Boston, MA, United States
Joy Kolonoski
– Harvard Pilgrim Healthcare Institute, United States
Mayura Shinde, DrPH
– Harvard Medical School, Boston, Massachusetts, United States
Rongmei Zhang
– FDA, United States
Efe Eworuke, PhD, MS, BPharm
– IQVIA, Falls Church, Virginia, United States
David Graham, MPH, MD
– Associate director, US Food and Drug Administration, Silver Spring, Maryland, United States
Laura Hou
– Research Analyst II, Harvard Pilgrim Health Care Institute, United States
Adebola Ajao
– Epidemiologist, US Food and Drug Administration, Food and Drug Administration, Silver Spring, Maryland, United States
Po-Yin Chang
– U.S. Food and Drug Administration, United States
John Connolly
– Harvard Pilgrim Health Care Institute, Boston, MA, United States

Background: Propensity score matching (PSM) is widely used for confounding adjustment in real world evidence (RWE) studies evaluating outcomes associated with drug use in routine care. However, it has an important limitation of removing unmatched observations from the analysis, potentially leading to increased covariate imbalance, reduced sample size and limited generalizability. In contrast, weighting on the propensity score has several advantages, including increased precision through retaining most observations and flexibility for targeting specific populations for inference. The FDA’s Sentinel System recently added inverse probability treatment weighting (IPTW) for confounding adjustment in comparative safety and effectiveness studies.

Objectives: To evaluate IPTW in a distributed data network by comparing adjusted effect estimates obtained using a PSM approach versus an IPTW approach in the Sentinel System for a study of stroke and bleeding risk in patients aged 65 years or older initiating nonvitamin K oral anticoagulants (NOAC) for atrial fibrillation.

Methods: Patients aged 65 years and older initiating NOACs (rivaroxaban, dabigatran, apixaban) between 10/19/2010 and 9/30/2015 in the Sentinel Medicare data partner were identified. PSM (1:1) and IPTW with stabilized average treatment effect weights were used separately for each pairwise NOAC-NOAC comparison. Cox proportional-hazards regression was used to estimate adjusted hazard ratios (HR) and 95% confidence intervals (CI) for the outcomes of thromboembolic stroke (stroke), intracranial hemorrhage (ICH), major extracranial bleeding (MEB), and GI bleeding (GIB) comparing each NOAC with each other.

Results: Overall, the point estimates and 95% CIs were similar between the analyses using PSM and IPTW. For example, there was no difference in the risk of stroke comparing rivaroxaban versus dabigatran after adjustment using PSM (HR [95% CI]: 0.89 [0.74, 1.07]) or IPTW (0.90 [0.76, 1.06]). Similarly, rivaroxaban use was associated with a numerical but non-significant increased risk of ICH compared to apixaban in both analyses [PSM HR (95% CI) 1.28 (0.99, 1.67) and IPTW 1.23 (0.96, 1.58)]. An increased risk of MEB and GIB was observed comparing rivaroxaban and apixaban users after both PSM and IPTW adjustment.

Conclusions: This study, which evaluated outcomes associated with NOAC use, demonstrated the feasibility of a newly implemented distributed approach to IPTW in the Sentinel System which produced similar effect estimates to PSM. Further studies to compare IPTW and alternative PS adjustment approaches in the Sentinel System are needed, particularly for cases to which PSM’s limitations apply.
The views expressed are the authors’ and not necessarily those of the U.S. Food and Drug Administration.