Statistical Inference in Abstracts Published in Cardiovascular Journals.

BACKGROUND

In 2016, the American Statistical Association stated that the use of statistical significance leads to distortion of the scientific process. The principal alternative to significance or null hypothesis testing (NHT) is estimation with point estimates and confidence intervals (CIs).

OBJECTIVES

The aim of this study was to determine the time trend of statistical inference and statistical reporting style in abstracts in major cardiovascular journals.

METHODS

A total of 84,250 abstracts published from 1975 to 2019 in 9 high-ranking cardiovascular journals (Circulation, Circulation Research, European Heart Journal, European Heart Journal: Cardiovascular Imaging, European Journal of Heart Failure, Journal of the American College of Cardiology, JACC: Cardiovascular Imaging, JACC: Cardiovascular Interventions, and JAMA Cardiology) were reviewed; in particular, proportions of abstracts containing statistical inference and its major variants (NHT, significance testing) were compared over time and among journals.

RESULTS

Overall, 49,924 abstracts (59%) contained statistical inference. Among these abstracts, NHT was the most frequent reporting style of statistical inference (79% among all journals). Journals differed considerably in the prevalence of CI reporting (1% to 78% in 2017-2019). With the exception of 2 journals, the proportion of abstracts containing CIs was higher in the more recent period. From 2013-2015 to 2017-2019, the proportion of abstracts containing only CIs increased by 5 (95% CI: 0 to 10), 18 (95% CI: 15 to 21), and 9 (95% CI: 3 to 15) percentage points in the European Heart Journal, the Journal of the American College of Cardiology, and JACC: Cardiovascular Imaging, respectively.

CONCLUSIONS

NHT is still the prevailing reporting style of statistical inference in major cardiovascular journals. Reporting of CIs in abstracts of major cardiovascular journals appears to be growing more popular.