COMPARATIVE CARDIOVASCULAR RISK IN U.S. POSTMENOPAUSAL WOMEN STARTING ROMOSOZUMAB VS DENOSUMAB: A HIGH-DIMENSIONAL INVERSE PROBABILITY OF SELECTION WEIGHTING ANALYSIS FOR MISSING CARDIOVASCULAR RISK FACTORS
Author(s)
Ye Liu, DrPH, MD1, Jingyi Zhang, MS1, Hongke Wu, MD, MS, MPH1, Tarun Arora, MS2, Elsa Strotmeyer, PhD3, S. Bobo Tanner, MD4, Kenneth Lyles, MD5, Andrea Burden, PhD1, Kenneth Saag, MD1, Jeffrey Curtis, MD, MS, MPH1;
1University of Alabama at Birmingham, Birmingham, AL, USA, 2FASTER Medicine, Hoover, AL, USA, 3University of pittsburgh, Pittsburgh, PA, USA, 4Vanderbilt University, Nashville, TN, USA, 5Duke University, Durham, NC, USA
1University of Alabama at Birmingham, Birmingham, AL, USA, 2FASTER Medicine, Hoover, AL, USA, 3University of pittsburgh, Pittsburgh, PA, USA, 4Vanderbilt University, Nashville, TN, USA, 5Duke University, Durham, NC, USA
OBJECTIVES: Romosozumab(romo) and denosumab(dmab) are recommended for postmenopausal osteoporosis at high fracture risk, yet romo carries an FDA boxed warning for cardiovascular(CV) risk. Using Medicare-PCORnet data, we examined how CV risk relates to initiating romo vs dmab and applied high-dimensional inverse-probability-of-selection weighting (hd-IPSW) to address bias from missing CV risk factors.
METHODS: We linked Medicare fee-for-service claims to PCORnet EMR via beneficiary identifiers. We included women ≥65 initiating romo or dmab from 4/1/2019-12/31/2022; metastatic cancer and Paget’s disease were excluded. Comorbidities were captured in claims and EMR(BMI, blood pressure, lipid panel). CV risk was estimated using ACC/AHA PREVENT model, ranked by deciles. Logistic regression assessed associations between CV risk and romo/dmab initiation, adjusting for demographics, fracture risk(FRAX), comorbidities, Charlson score, healthcare utilization, and other clinically-selected covariables. Using a high-dimensional propensity score framework, hd-IPSW selected top 100 predictors by three approaches: (1)strength of association with complete-case status; (2)potential bias from selection and treatment (Bross formula); (3)strength of association with treatment. We estimated hd-IPSW with logistic models and fit hd-IPSW-weighted models otherwise same as complete-case analysis.
RESULTS: We identified 6,825 new romo/dmab users; 2,193 were complete cases, others lacked CV risk information. In complete-case analyses, higher CV risk was associated with lower odds of romo initiation (aOR=0.89[95%CI 0.83-0.96]). Complete cases were less likely to initiate romo than dmab vs those with missing values before weighting (OR=0.57[0.49-0.66]), but was attenuated after hd-IPSW (approach 1: wOR=0.83[0.72-0.95]; 2: 0.92[0.80-1.06]; 3: 0.89[0.77-1.02]). After hd-IPSW, associations between CV risk and romo/dmab initiation were non-significant (approach 1: awOR=0.97[0.91-1.04]; 2: 0.97[0.91-1.04]; 3: 0.95[0.90-1.02]).
CONCLUSIONS: The association between CV risk and initiating romo could be affected by the missingness of CV risk in complete-case analysis. Hd-IPSW could mitigate such bias. Future comparative safety studies should carefully evaluate CV risk in light of missingness patterns.
METHODS: We linked Medicare fee-for-service claims to PCORnet EMR via beneficiary identifiers. We included women ≥65 initiating romo or dmab from 4/1/2019-12/31/2022; metastatic cancer and Paget’s disease were excluded. Comorbidities were captured in claims and EMR(BMI, blood pressure, lipid panel). CV risk was estimated using ACC/AHA PREVENT model, ranked by deciles. Logistic regression assessed associations between CV risk and romo/dmab initiation, adjusting for demographics, fracture risk(FRAX), comorbidities, Charlson score, healthcare utilization, and other clinically-selected covariables. Using a high-dimensional propensity score framework, hd-IPSW selected top 100 predictors by three approaches: (1)strength of association with complete-case status; (2)potential bias from selection and treatment (Bross formula); (3)strength of association with treatment. We estimated hd-IPSW with logistic models and fit hd-IPSW-weighted models otherwise same as complete-case analysis.
RESULTS: We identified 6,825 new romo/dmab users; 2,193 were complete cases, others lacked CV risk information. In complete-case analyses, higher CV risk was associated with lower odds of romo initiation (aOR=0.89[95%CI 0.83-0.96]). Complete cases were less likely to initiate romo than dmab vs those with missing values before weighting (OR=0.57[0.49-0.66]), but was attenuated after hd-IPSW (approach 1: wOR=0.83[0.72-0.95]; 2: 0.92[0.80-1.06]; 3: 0.89[0.77-1.02]). After hd-IPSW, associations between CV risk and romo/dmab initiation were non-significant (approach 1: awOR=0.97[0.91-1.04]; 2: 0.97[0.91-1.04]; 3: 0.95[0.90-1.02]).
CONCLUSIONS: The association between CV risk and initiating romo could be affected by the missingness of CV risk in complete-case analysis. Hd-IPSW could mitigate such bias. Future comparative safety studies should carefully evaluate CV risk in light of missingness patterns.
Conference/Value in Health Info
2026-05, ISPOR 2026, Philadelphia, PA, USA
Value in Health, Volume 29, Issue S6
Code
EPH154
Topic
Epidemiology & Public Health
Topic Subcategory
Safety & Pharmacoepidemiology
Disease
SDC: Cardiovascular Disorders (including MI, Stroke, Circulatory), SDC: Musculoskeletal Disorders (Arthritis, Bone Disorders, Osteoporosis, Other Musculoskeletal)