ABSTRACT
Introduction
A variability in statistical complexity exists across the ophthalmology literature. This review investigates the frequency of statistical methods in ophthalmology articles and aims to determine differences based on study design, subspecialty and journal.
Areas covered
Original articles published in the top three comprehensive ophthalmology journals based on 2017 impact factor were identified. Searches of Ovid MEDLINE were used to elicit relevant literature. In total, 325 articles were identified from top comprehensive ophthalmology journals. The top three statistical methods were contingency tables, t-tests and non-parametric tests. Less than half (n = 136, 41.8%) of included articles did not use statistical methods or relied on descriptive statistics. Uveitis publications had the greatest number of statistical methods per paper, while oculoplastics had the lowest. Randomized controlled trials had the greatest number of methods per paper, while case reports and correspondences had the fewest. The average number of methods was 2.1 in the American Journal of Ophthalmology, 1.5 in Ophthalmology and 1.0 in JAMA Ophthalmology.
Expert opinion
The statistics of approximately half of all ophthalmology articles in top comprehensive journals are accessible to readers with an understanding of four common statistical methods: descriptive statistics, contingency tables, t-tests and non-parametric tests.
Article Highlights
The statistics of approximately half of all ophthalmology articles in top comprehensive journals are accessible to readers with an understanding of four common statistical methods: descriptive statistics, contingency tables, t-tests and non-parametric tests
Uveitis and immunology publications had the greatest number of statistical methods per paper, while oculoplastics had the fewest
RCTs had the greatest number of statistical methods per paper, while case reports and correspondences had the fewest
The average number of statistical methods was 2.1 in the American Journal of Ophthalmology, 1.5 in Ophthalmology and 1.0 in JAMA Ophthalmology.
Contributorship Statement
Research Design: Popovic, Schlenker.
Data Acquisition and/or Research Execution: Popovic, Yissar, Joksimovic, Thang, Schlenker.
Data Analysis and/or Interpretation: All authors.
Manuscript Preparation: All authors.
Approval of Final Article: All authors.
This manuscript has not been published anywhere previously and is not simultaneously being considered for any other publication.
Declaration of interest
The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties.
Expert Opinion
This review showed that approximately half of ophthalmology articles in top comprehensive journals are currently accessible to readers with an understanding of four common statistical methods: descriptive statistics, contingency tables, t-tests and non-parametric tests. All four of these tests are not complex and are interpretable to a lay audience with basic epidemiologic training. With increasing statistical complexity over time, the highest frequency tests will likely change in the future. Nonetheless, the rate of change of statistical complexity in the ophthalmology literature is a slow process, and thus for the forseeable future it is advised that readers of the literature be comfortable with the interpretations of the currently identified highest frequency tests. It is our hope that our findings will be useful to teachers of ophthalmic epidemiology courses, who may be able to incorporate teaching of the high frequency methods identified above.
In our study, there was a significant difference in the number of statistical methods per paper between the top comprehensive ophthalmology journals. We believe this effect is similar for the wider ophthalmology literature. At the same time, we believe that the ‘gap’ in statistical number and complexity per paper between various journals has narrowed over time and will likely continue to narrow in the future. This is likely the result of ever increasing interdisciplinary collaboration in studies that lead to statistical consultation for more projects, irrespective of journal. As well, there is now more emphasis than ever on statistical understanding in academic medicine, given the necessity of critical appraisal of the medical literature to a truly evidence-based practice. Together, this has led to a higher quality of statistical testing across all ophthalmology journals.
In subgroup analysis, uveitis and immunology publications had the greatest number of statistical methods per paper, while oculoplastics had the lowest. These estimates are dependent on the number of low complexity articles that are present for each study design, such as case reports and correspondences. For example, retinal publications had a relatively low number of statistics per article, presumably due to a high number of ophthalmic imaging articles and case reports. As expected, RCTs had the greatest number of statistical methods per paper, while case reports and correspondences had the lowest.
Errors in statistical analysis is another area that should be considered in the evaluation of statistical methods in ophthalmology. Thus far, our study has shown that certain statistical methods like sample size calculations remain underrepresented in the literature, which can lead to issues such as type II error. Greater awareness of common statistical errors encountered will be of benefit to authors, readers, peer reviewers and journal editors.
Reviewer Disclosures
Peer reviewers on this manuscript have no relevant financial or other relationships to disclose.
Meeting Presentation
None.