John D Seeger1,2, Katsiaryna Bykov3, Dorothee B Bartels4,5, Krista Huybrechts3, Sebastian Schneeweiss3. 1. Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, 1620 Tremont Street, Suite 3030, Boston, MA, 02120, USA. john.seeger@optum.com. 2. Optum Epidemiology, 1325 Boylston St., Boston, MA, 02215, USA. john.seeger@optum.com. 3. Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women's Hospital and Harvard Medical School, 1620 Tremont Street, Suite 3030, Boston, MA, 02120, USA. 4. Corporate Department Global Epidemiology, Boehringer Ingelheim GmbH, Ingelheim, Germany. 5. Hannover Medical School, Institute for Epidemiology, Social Medicine and Health Systems Research, Hannover, Germany.
Abstract
INTRODUCTION: Comparing medications in observational settings requires differences in patient characteristics to be accounted for. Propensity score (PS) methods can address these differences, but PS weighting approaches may introduce bias. METHODS: Within a cohort study of anticoagulant initiators from October 2010 through to December 2012, PS values for dabigatran relative to warfarin were estimated, and study outcomes (stroke or major bleeding) among the cohort were identified. The PS was used to match initiators and results compared with those obtained using inverse probability of treatment weighting (IPTW) and standardized morbidity ratio (SMR) weighting. Hazard ratios (HRs) for study outcomes were estimated using a proportional hazards regression model. RESULTS: There were 23,543 dabigatran and 50,288 warfarin initiators, and matching formed 19,189 pairs (81.5% of dabigatran initiators) which resulted in a pooled stroke HR of 0.77 (95% confidence interval [CI] 0.54-1.09), and a pooled major hemorrhage HR of 0.75 (95% CI 0.65-0.87). The IPTW results for stroke (HR = 0.00; 95% CI 0.00-0.56) and major hemorrhage (HR = 0.08; 95% CI 0.08-0.10) substantially differed, while the SMR-weighted results for stroke (HR = 0.65; 95% CI 0.42-1.03) and major hemorrhage (HR = 0.73; 95% CI 0.61-0.85) differed only slightly from matching. CONCLUSIONS: In this example, different applications of the same PS led to substantially different results, a finding that was particularly apparent with IPTW, and this was remedied by truncating extreme weights. If IPTW is used, information regarding the weights applied along with sensitivity analyses could avoid misrepresentation of study results, and would enhance their interpretation.
INTRODUCTION: Comparing medications in observational settings requires differences in patient characteristics to be accounted for. Propensity score (PS) methods can address these differences, but PS weighting approaches may introduce bias. METHODS: Within a cohort study of anticoagulant initiators from October 2010 through to December 2012, PS values for dabigatran relative to warfarin were estimated, and study outcomes (stroke or major bleeding) among the cohort were identified. The PS was used to match initiators and results compared with those obtained using inverse probability of treatment weighting (IPTW) and standardized morbidity ratio (SMR) weighting. Hazard ratios (HRs) for study outcomes were estimated using a proportional hazards regression model. RESULTS: There were 23,543 dabigatran and 50,288 warfarin initiators, and matching formed 19,189 pairs (81.5% of dabigatran initiators) which resulted in a pooled stroke HR of 0.77 (95% confidence interval [CI] 0.54-1.09), and a pooled major hemorrhage HR of 0.75 (95% CI 0.65-0.87). The IPTW results for stroke (HR = 0.00; 95% CI 0.00-0.56) and major hemorrhage (HR = 0.08; 95% CI 0.08-0.10) substantially differed, while the SMR-weighted results for stroke (HR = 0.65; 95% CI 0.42-1.03) and major hemorrhage (HR = 0.73; 95% CI 0.61-0.85) differed only slightly from matching. CONCLUSIONS: In this example, different applications of the same PS led to substantially different results, a finding that was particularly apparent with IPTW, and this was remedied by truncating extreme weights. If IPTW is used, information regarding the weights applied along with sensitivity analyses could avoid misrepresentation of study results, and would enhance their interpretation.
Authors: Peter M Wahl; Keith Rodgers; Sebastian Schneeweiss; Brian F Gage; Javed Butler; Charles Wilmer; Marshall Nash; Gregory Esper; Norman Gitlin; Neal Osborn; Louise J Short; Rhonda L Bohn Journal: Pharmacoepidemiol Drug Saf Date: 2010-06 Impact factor: 2.890
Authors: David J Graham; Marsha E Reichman; Michael Wernecke; Rongmei Zhang; Mary Ross Southworth; Mark Levenson; Ting-Chang Sheu; Katrina Mott; Margie R Goulding; Monika Houstoun; Thomas E MaCurdy; Chris Worrall; Jeffrey A Kelman Journal: Circulation Date: 2014-10-30 Impact factor: 29.690
Authors: J J Gagne; R J Glynn; J A Rassen; A M Walker; G W Daniel; G Sridhar; S Schneeweiss Journal: Clin Pharmacol Ther Date: 2012-05-16 Impact factor: 6.875
Authors: Lu Wu; Diamantis I Tsilimigras; Katiuscha Merath; J Madison Hyer; Anghela Z Paredes; Rittal Mehta; Kota Sahara; Fabio Bagante; Eliza W Beal; Feng Shen; Timothy M Pawlik Journal: J Gastrointest Surg Date: 2019-04-22 Impact factor: 3.452
Authors: M Lefort; S Sharmin; J B Andersen; M Magyari; T Kalincik; E Leray; S Vukusic; R Casey; M Debouverie; G Edan; J Ciron; A Ruet; J De Sèze; E Maillart; H Zephir; P Labauge; G Defer; C Lebrun-Frenay; T Moreau; E Berger; P Clavelou; J Pelletier; B Stankoff; O Gout; E Thouvenot; O Heinzlef; A Al-Khedr; B Bourre; O Casez; P Cabre; A Montcuquet; A Wahab; J P Camdessanché; A Maurousset; H Ben Nasr; K Hankiewicz; C Pottier; N Maubeuge; D Dimitri-Boulos; C Nifle; D A Laplaud; D Horakova; E K Havrdova; R Alroughani; G Izquierdo; S Eichau; S Ozakbas; F Patti; M Onofrj; A Lugaresi; M Terzi; P Grammond; F Grand'Maison; B Yamout; A Prat; M Girard; P Duquette; C Boz; M Trojano; P McCombe; M Slee; J Lechner-Scott; R Turkoglu; P Sola; D Ferraro; F Granella; V Shaygannejad; J Prevost; D Maimone; O Skibina; K Buzzard; A Van der Walt; R Karabudak; B Van Wijmeersch; T Csepany; D Spitaleri; S Vucic; N Koch-Henriksen; F Sellebjerg; P S Soerensen; C C Hilt Christensen; P V Rasmussen; M B Jensen; J L Frederiksen; S Bramow; H K Mathiesen; K I Schreiber; H Butzkueven Journal: BMC Med Res Methodol Date: 2022-05-30 Impact factor: 4.612