BackgroundThe recent availability of dabigatran, a novel oral anticoagulant, provided a new treatment option for stroke prevention in atrial fibrillation beyond warfarin, the main therapy for years. Little is known about their real‐world comparative effectiveness and safety, even less among patient demographic and clinical subgroups.Methods and ResultsUsing a cohort of non‐valvular AF patients initiating anticoagulation from October 2010 to December 2012 drawn from a large US database of commercial and Medicare supplement claims, we applied propensity score weights to Cox proportional hazards regression to assess the comparative effectiveness and safety of dabigatran versus warfarin. Analyses were repeated among clinical and demographic subgroups using stratum‐specific propensity scores as an exploratory analysis. Of the 64 935 patients initiating anticoagulation, 32.5% used dabigatran. Compared with warfarin, dabigatran was associated with a lower risk of ischemic stroke or systemic embolism (composite adjusted Hazard Ratio [aHR], 95% CI: 0.86, 95% CI: 0.79 to 0.93), hemorrhagic stroke (aHR: 0.51, 0.40 to 0.65), and acute myocardial infarction (aHR: 0.88, 95% CI: 0.77 to 0.99), and no relation was seen between dabigatran and the composite harm outcome (aHR: 0.94, 95% CI: 0.87 to 1.01). However, dabigatran was associated with a higher risk of gastrointestinal bleeding (aHR: 1.11, 95% CI: 1.02 to 1.22). Estimates of effectiveness and safety appeared to be mostly similar across subgroups.ConclusionsDabigatran could be a safe and potentially more effective alternative to warfarin in patients with atrial fibrillation managed in routine practice settings.

Background It is unclear whether gender and racial/ethnic gaps in the use of and patient adherence to β-blockers, angiotensin-converting-enzyme inhibitors (ACEIs)/angiotensin receptor blockers (ARBs), and HMG-CoA reductase inhibitors (statins) post-acute myocardial infarction (AMI) have persisted following establishment of the Medicare Part D prescription program. Methods and Results This retrospective cohort study used 2007-2009 Medicare service claims among Medicare beneficiaries ≥ 65 years who were alive 30 days after an index AMI hospitalization in 2008. Multivariable logistic regression models examined racial/ethnic (white, black, Hispanic, Asian, and Other) and gender differences in the use of these therapies in the 30 days post-discharge and patient adherence at 12-months post-discharge, adjusting for patient baseline sociodemographic and clinical characteristics. Out of 85,017 individuals, 55%, 76%, and 61% used ACEIs/ARBs, β-blockers, and statins within 30 days post-discharge, respectively. No marked differences in use were found by race/ethnicity but women were less likely to use ACEI/ARBs and β-blockers compared with men. However, at 12-months post-discharge compared with white men, black and Hispanic women had the lowest likelihood (approximately 30-36% lower, p <0.05) of being adherent, followed by white, Asian, and other women and black and Hispanic men (approximately 9-27% lower, p <0.05). No significant difference was shown between Asian/other men and white men. Conclusions While minorities were initially no less likely to use the therapies post-AMI discharge compared with white patients, black and Hispanic patients had significantly lower adherence over 12 months. Strategies to address gender and racial/ethnic gaps in the elderly are needed.

Knowledge Distillation (KD) has made remarkable progress in the last few years and become a popular paradigm for model compression and knowledge transfer. However, almost all existing KD algorithms are datadriven, i.e., relying on a large amount of original training data or alternative data, which is usually unavailable in real-world scenarios. In this paper, we devote ourselves to this challenging problem and propose a novel adversarial distillation mechanism to craft a compact student model without any real-world data. We introduce a model discrepancy to quantificationally measure the difference between student and teacher models and construct an optimizable upper bound. In our work, the student and the teacher jointly act the role of the discriminator to reduce this discrepancy, when a generator adversarially produces some "hard samples" to enlarge it. Extensive experiments demonstrate that the proposed data-free method yields comparable performance to existing data-driven methods. More strikingly, our approach can be directly extended to semantic segmentation, which is more complicated than classification and our approach achieves state-of-the-art results. The code will be released.

No abstract