Rivaroxaban vs. Warfarin Comparative Effectiveness

Patients were described at treatment initiation in terms of demographic and clinical variables. Continuous variables are presented as means and standard deviations or medians and interquartile ranges (IQR). The numbers and proportions of patients in each category are presented for categorical variables. Person-years of follow-up were calculated from the index date to the outcome event of interest, discontinuation of the index treatment, death, or the end of the study period, whichever comes first. Incidence rates were calculated as the number of events over the observed person-time and presented as per 100 person-years.
We used the propensity score (PS) methods to compare the rivaroxaban and warfarin groups (19 (link)). We utilized stabilized inverse probability of treatment weighting (IPTW) approach based on the PS to adjust for potential confounding resulting from imbalances in baseline patient characteristics. The objective of IPTW is to create a weighted sample for which the distribution of either the confounding variables or the prognostically important covariates is approximately the same between comparison groups (20 (link)). PS is the patient’s probability of receiving a treatment under investigation (rivaroxaban) given a set of known patients’ baseline characteristics. PS was calculated using multiple logistic regression on all the available covariates, including demographics, co-morbidities, CHA₂DS₂-VASc score, Charlson Comorbidity Index, and concomitant medication. For the exploratory analysis, health examination variables such as body weight, body mass index (BMI), eGFR, smoking, alcohol consumption, and physical activity were additionally included for PS calculation. Detailed methods of IPTW are described in Supplementary methods. After IPTW, we assessed the balance of the two treatment groups by using absolute standardized differences (ASDs). The PSs and stabilized weights distributions were inspected for initial and synthetic samples. An ASD of 0.1 or less was considered as a negligible difference between the two groups. The weighted event numbers and incidence rates were calculated. We compared treatments using weighted Cox proportional hazards regression with IPTW. Results of Cox analyses are reported as hazard ratios (HRs) with 95% confidence intervals (CIs). Each Cox regression was checked to see if the model assumptions were fulfilled. For the exploratory analysis set, weighted cumulative incidences of the composite of five renal outcomes were estimated by the Kaplan–Meier method and log-rank test.
All statistical analyses were performed using SAS 9.3 (SAS Institute Inc., Cary, NC, United States).