METHODS

This systematic review was performed in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses)16. The systematic search was conducted by two independent investigators (S.T and A.N.L) using databases of MEDLINE and ClinicalTrials.gov from 1985 to April 2019. A broad search strategy was used to identify RCTs of AF in human subjects: (*Atrial fibrillation OR *AF) AND (*Rate control therapy OR *Rhythm control therapy OR *Ablation OR *Cardioversion OR *Anticoagulation OR *Treatment) (Table supplementary S1). ClinicalTrials.gov was queried using following limits: Adult (≥18 years), interventional, phase II-III and “atrial fibrillation”. We also downloaded meta-analyses related to AF for additional information 17, 18. After removing duplicates, all the articles were scrutinized at the title and abstract level followed by review of full text to include phase II and III RCTs of adult human participants with follow-up of at least 3 months. While there was no restriction on sample size, we preferred 3-month follow-up to include maximum number of RCTs yet ensuring important data because relatively longer follow-up is considered to provide more accurate information13. The following data was extracted: (1) title, (2) year of publication, (3) journal, (4) total number of participants, (5) number of women, (6) mean or median age, (7) number of older patients, defined as ≥75 years, (8) ethnicity and race, (9) location, (10) funding sources, (11) types of RCTs (prevention of stroke or treatment of AF), and (12) baseline population. We reviewed ClinicalTrials.gov for any missing information not obtained from trial’s manuscript. We followed the US Food and Drug Administration (FDA) position statement during data abstraction on ethnicity and race minorities19. Ethnicity was captured as Hispanic/Latino; Race was defined as: (1) African American or Black, (2) American Indian or Alaskan Native, (3) Native Hawaiian or other Pacific Islander, and (4) Asians. If study had characterized the race as non-White or race other than White was not clearly specified, we grouped data as non-White race. Location of RCTs were divided into (1) exclusively North America including United States, Canada, and Mexico; (2) exclusively Western Europe including Austria, Belgium, Bermuda, Denmark, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Luxembourg, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, and United Kingdom; (3) rest of the world-exclusively outside of North America and Western Europe, and (4) Multi-regional. We classified funding sources as per ClinicalTrials.gov designations: government, academic center/university, industry or collaborative trials between non- profit organizations and industries 13. Baseline population was categorized into non- valvular AF, AF in mitral valve disease, AF in rheumatic valve disease, status post catheter ablation in AF and AF in miscellaneous groups (i.e. AF after cardioversion, AF in heart failure or with recent percutaneous coronary intervention). Further categorization of stroke prevention and AF treatment trials was done based on medications or interventions examined in RCTs. To examine proportion of these demographic subsets in our trial level data base relative to their representation in disease population, we selected comparative data from observational and epidemiological studies (Table S2) 4, 20-28. To most closely mirror the target population in AF RCTs, the relevant comparative data was selected to represent global disease burden including regions of North America, Western Europe, Asia and other regions. The data was weighted for sample size, i.e. weighted mean for women, age, ethnic/racial minorities was derived by multiplying the mean of each study by weighted number based on study’s size relative to the total studies included. RCTs were grouped according to publication year starting from 1989-2019, with first and last group consisting of 5-year period and remaining groups of 4-year period. Continuous variables were reported as mean (standard deviations [SD]) or median (interquartile range [IQR]). Categorical variables were expressed as No. (%). Categorical variables were compared using Pearson χ2 testing. One-way analysis of variance (ANOVA) test was used for testing significance of means and α was kept at 0.05. To test for a trend in demographic characteristics of the patient samples across the study period, we used simple linear regression models using year of publication as the independent variable. The dependent variables were the proportion of women, mean age of participants, proportion of >75 years old participants and the proportion of underrepresented minorities. Analyses were performed with IBM, version SPSS 24 (IBM Corporation) and Microsoft Excel (Microsoft Corporation).