Limitations of ARISTOTLE Weight Subgroup Analysis (post hoc) Hohnloser et al.

• Results obtained from a post hoc analysis are subject to limitations due to the retrospective nature of the analysis
• The relationship between the changes in weight over time and the outcomes could not be assessed
• Results may not be generalizable to the general population

Limitations of ARISTOTLE BMI Analysis (post hoc) Sandhu et al.

• This is a post hoc analysis that suffers from typical limitations; the distribution of demographics and comorbidities in our study reflects RCT participants rather than real-world clinical practice
• Baseline BMI was used and therefore the impact of weight changes on outcomes was not assessed
• Markers of inflammation were not adjusted in the multivariable analysis

Limitations of ARISTOPHANES 2020

• Only statistical associations could be concluded and not causal relationships
• Although cohorts were matched through PSM, there were potential residual confounders
• Outcome measures could only be based on ICD-9-CM codes without further adjustment with precise clinical criteria
• Body measurements such as weight or lean body mass were not available in the claims data. Classification was based on
  ICD-9 codes
• Laboratory values were not available in the dataset so the time in therapeutic range for patients prescribed warfarin was indeterminable
• Unobserved heterogeneity may exist across the 5 data sources
• Results may not reflect the overall NVAF population in the United States as the study did not include uninsured patients or those solely covered by other public health insurance plans

Limitations of Deitelzweig VA and Medicare 2022

• Findings
  • Only associations
  • Presence of a prescription claim did not indicate that the medication was taken as prescribed or at all
  • OTC medications (e.g., aspirin) and medication samples were not captured in claims data
• Risk of residual confounding exists even after IPTW
• Dataset did not support the longitudinal evaluation of specific INR values in warfarin patients
• Majority of the study population was male, which did not reflect the male to female ratio of the overall US NVAF population
• Results may not be generalizable to the entire US NVAF population, such as women, the uninsured, those with other insurances (e.g., commercial or Medicaid), and those without dual Veterans Affairs-Medicare coverage

Limitations of Briasoulis 2022

• Information on the INR levels of warfarin users and TTR is lacking
• Outcomes for admissions of Veteran Affairs patients to healthcare facilities outside of the VA were not available; however, given the large sample size, the RR of admission outside the VA should not differ by drug type
• There is a possibility of residual confounding from unmeasured comorbidities, other confounders, and off-label dosing of DOACs
• The study only included veterans, so findings might not be generalizable to other populations
• The study included a small proportion of female patients

Limitations of O’Kane 2022

• Components of CHA₂DS₂-VASc and HAS-BLED scores differed between the BMI ≥50 kg/m² and <30 kg/m² groups
  • The BMI <30 kg/m² group scores were driven more by older age
  • The BMI ≥50 kg/m² group scores were driven by hypertension, heart failure, and diabetes
• In the BMI ≥50 kg/m² group, a higher proportion of patients were taking rivaroxaban compared with that in the BMI
  <30 kg/m² group
• The retrospective single-health system design and the inability to assess adherence to medications
• SE events included both VTE and arterial thromboembolism, but the delineation of events was not recorded
  • Since VTE events are rarely associated with AF, the incidence of SE related to NVAF may be overestimated
• The study was underpowered
• The point estimates for ISTH major bleeding and CRNM bleeding flip from protective in the unadjusted analysis to demonstrating an increased risk of bleeding in the multivariate logistical analysis, suggesting that renal function, age, or choice of DOAC therapy may be impacting the safety outcomes
• The study only reviewed apixaban and rivaroxaban, so extrapolation to the other DOACs may be limited

Limitations of Kushnir 2019

• The standard limitations of retrospective non-randomized chart-based cohort analyses
• Attempted to compensate for potential bias in the choice of anticoagulant by adjusting for the CHA₂DS₂-VASc score for patients with AF, and age and Charlson score for all patients, but residual confounding is likely and over adjustment is possible
• Low sample size and a low number of clinical events
• The Bronx has relatively high proportions of African-American and Hispanic patients: The findings may not be generalizable to other morbidly obese populations
• No data on certain thrombotic risk factors, such as active malignancy or history of bariatric surgery
• Events occurring in the study cohorts outside of the Montefiore system or Bronx Regional Health Information Organization Network may not have been identified
• Reliable adherence data were not available. Time in TTR of patients on warfarin could not be calculated and serum concentrations of DOACs were not available

Limitations of AMPLIFY Post hoc Analysis

• Body weight and BMI were only assessed at baseline¹
  • Clinical outcomes may have been impacted by body weight and BMI changes during the trial
• Apixaban exposure may differ in a real-world population, particularly the obese population, due to more comorbidities and concomitant medications¹
• Small number of patients included in the following¹:
  • ≥120 kg body weight and BMI >40 kg/m² groups
  • Population PK analysis (~5% of AMPLIFY patients)
• Relatively short follow-up (30 days) after the 6-month intended treatment period²

Limitations of Cohen 2021

• The interpretation of results was based on statistical associations, not causality
• Study used an on-treatment approach and did not evaluate the events occurring after the patients switched or discontinued the index treatment, making it difficult to confirm the relationship based on the claims data
• There may be misclassifications of the diagnosis codes
  • Due to the lack of weight-related information in the databases, obesity was identified based on the diagnosis codes for BMI ≥30, ≥40, or codes that had an indication for obesity status
  • Recurrent VTE and MB may be incorrectly coded or included as rule-out criteria than actual disease
• Although the IPTW system was used to control for varying patient characteristics, the IPTW methodology may be subject to some hidden or residual confounding
• Results may not be generalizable to the entire obese VTE population in the United States as uninsured patients or those who had coverage through other insurance were not included

Limitations of Perino 2021

• No adequate and well-controlled head-to-head clinical trials have compared the safety and efficacy of DOACs
• Potential for clinically meaningful differences in patients’ demographics and baseline comorbidities across body sizes and prescribed anticoagulant (DOACs vs warfarin)
• Residual confounding variables (e.g., provoked, unprovoked, and malignancy-associated VTE) cannot be ruled out
• There were fewer patients at the extremes of the weight and BMI distributions were included
• Management failures of VTE in the outpatient setting were not captured
• The results may not be generalizable outside the VA healthcare system, which uses pharmacist-led anticoagulation clinics associated with high rates of on-label DOACs dosing and TTR for patients treated with warfarin, and has a predominantly male population

Limitations of Crouch 2021

• Retrospective design and potential lack of follow-up data to confirm VTE recurrence or bleeding events
  • The reliability of the results dependent on complete and accurate documentation within the electronic health record
• Inability to collect information on duration of anticoagulation, adherence, or outcome events in the outpatient setting, potential confounding comorbid diseases not included in our baseline data, time in therapeutic INR range for those on warfarin, reason for hospital encounters, and may not have captured events that did not result in a hospitalization or an emergency department visit

Limitations of Kushnir 2019

• The standard limitations of retrospective non-randomized chart-based cohort analyses
• Attempted to compensate for potential bias in the choice of anticoagulant by adjusting for the CHA₂DS₂-VASc score for patients with AF, and age and Charlson score for all patients, but residual confounding is likely and over adjustment is possible
• Low sample size and a low number of clinical events
• The Bronx has relatively high proportions of African-American and Hispanic patients: The findings may not be generalizable to other morbidly obese populations
• No data on certain thrombotic risk factors, such as active malignancy or history of bariatric surgery
• Events occurring in the study cohorts outside of the Montefiore system or Bronx Regional Health Information Organization Network may not have been identified
• Reliable adherence data were not available. Time in TTR of patients on warfarin could not be calculated and serum concentrations of DOACs were not available

Limitations of ADVANCE-1, ADVANCE-2 and ADVANCE-3

• Lack of statistical power in potentially important subgroups that included relatively few patients
• Possibility of false-positive findings owing to chance alone because of multiple statistical testing

Limitations/Disclaimers: Real-world evidence of oral anticoagulants

• There are no adequate and well-controlled head-to-head clinical trials comparing the efficacy and safety of DOACs
• Real-world evidence evaluating DOACs should not be assumed to imply comparable efficacy, safety, or product interchangeability
• Limitations of the real-world evidence studies (retrospective database analyses) are primarily those associated with the use of administrative data that rely on accurate and complete ICD-9 coding, CPT coding, EHR, and pharmacy claims
• The interpretation of results is based on statistical associations, not causality
• In observational studies, not all unobserved confounders may have been adjusted for or controlled for
• The use of OTC medications (e.g., aspirin, NSAIDs, etc.) and laboratory values (e.g., INR) are not typically captured in claims data
• The results are applicable only to the populations studied and may not be generalizable to other populations
• Edoxaban has not been included in some studies incorporated in this deck as either the studies were conducted prior to edoxaban market entry, or sample sizes were too small to allow for a meaningful analysis