Text mining describes the use of statistical and epidemiological methods in published medical research

To describe trends in the use of statistical and epidemiological methods in the medical literature over the past 2 decades.

We obtained all 1,028,786 articles from the PubMed Central Open-Access archive (retrieved May 9, 2015). We focused on 113,450 medical research articles. A Delphi panel identified 177 statistical/epidemiological methods pertinent to clinical researchers. We used a text-mining approach to determine if a specific statistical/epidemiological method was encountered in a given article. We report the proportion of articles using a specific method for the entire cross-sectional sample and also stratified into three blocks of time (1995-2005; 2006-2010; 2011-2015).

Numeric descriptive statistics were commonplace (96.4% articles). Other frequently encountered methods groups included statistical inferential concepts (52.9% articles), epidemiological measures of association (53.5% articles) methods for diagnostic/classification accuracy (40.1% articles), hypothesis testing (28.8% articles), ANOVA (23.2% articles), and regression (22.6% articles). We observed relative percent increases in the use of: regression (103.0%), missing data methods (217.9%), survival analysis (147.6%), and correlated data analysis (192.2%).

This study identified commonly encountered and emergent methods used to investigate medical research problems. Clinical researchers must be aware of the methodological landscape in their field, as statistical/epidemiological methods underpin research claims.