Statistical software applications used in health services research: analysis of published studies in the U.S
1 The Ohio State University College of Public Health, 202 Cunz Hall, 1841 Neil Avenue, Columbus, Ohio 43210, USA
2 Abbott Vascular, 3200 Lakeside Drive, Santa Clara, CA 95054, USA
3 The Ohio State University College of Public Health, The Center for Health Outcomes, Policy & Evaluation Studies, 5049 Smith Laboratories, 174 West 18th Avenue, Columbus, Ohio 43210, USA
BMC Health Services Research 2011, 11:252 doi:10.1186/1472-6963-11-252Published: 6 October 2011
This study aims to identify the statistical software applications most commonly employed for data analysis in health services research (HSR) studies in the U.S. The study also examines the extent to which information describing the specific analytical software utilized is provided in published articles reporting on HSR studies.
Data were extracted from a sample of 1,139 articles (including 877 original research articles) published between 2007 and 2009 in three U.S. HSR journals, that were considered to be representative of the field based upon a set of selection criteria. Descriptive analyses were conducted to categorize patterns in statistical software usage in those articles. The data were stratified by calendar year to detect trends in software use over time.
Only 61.0% of original research articles in prominent U.S. HSR journals identified the particular type of statistical software application used for data analysis. Stata and SAS were overwhelmingly the most commonly used software applications employed (in 46.0% and 42.6% of articles respectively). However, SAS use grew considerably during the study period compared to other applications. Stratification of the data revealed that the type of statistical software used varied considerably by whether authors were from the U.S. or from other countries.
The findings highlight a need for HSR investigators to identify more consistently the specific analytical software used in their studies. Knowing that information can be important, because different software packages might produce varying results, owing to differences in the software's underlying estimation methods.