Doria AS; Tomlinson G; Beyene J, et al. Research Methods in Radiology: A Practical Guide; Thieme;2018;328 pp; 114 ill;$99.99.
The statistical analysis of study data is often a kind of a “black box” for those of us without much formal training in biostatistics or epidemiology. We take the statistical validity of published data for granted, and focus on how the results will impact our clinical practice. The critical reader will seek to identify potential flaws in the study design, or in the implications of the results, but the appropriateness of statistical tests performed are generally accepted prima facie.
Faith in rigorous peer review notwithstanding, some familiarity with the more frequently performed statistical tests fosters a deeper understanding of a study. Unfortunately, medical school, residency, and fellowship curricula infrequently dedicate formal instruction time to this vitally important facet of training, and few resources are devoted to the specific kinds of study designs and statistical analyses that commonly arise in imaging studies.
This shortcoming is where Research Methods in Radiology: A Practical Guide aims to step into the breach. Written by a team of radiologists and statisticians from the University of Toronto and McMaster University, this approximately 300-page volume is packed with well-organized concepts focused on the types of issues in research design, data analysis, and publishing that are most pertinent to our field. This makes it a valuable resource for students and residents who are looking to better understand the literature or prepare a manuscript draft of their own, and a particularly useful text for fellows or junior faculty by emphasizing sound research design principles and explaining the statistics appropriate for answering well-formulated research questions.
The book begins with an overview of research designs, with topics including cross-sectional, case-control, retrospective and prospective cohort, and randomized-controlled studies. It follows with chapters discussing descriptive statistics (e.g. histograms, measures of centrality and dispersion, and diagnostic test performance, sensitivity, specificity, likelihood ratios, and receiver operating characteristic [ROC] curves.) True to the author’s focus on radiology, this latter chapter includes important contextual discussion on how prevalence impacts the positive and negative predictive values of any diagnostic test.
Moreover, the authors provide an explanation of biases such as lead-time and length-time bias, an understanding of which is critical to a nuanced interpretation of the many studies assessing the clinical impact of population-based imaging screening programs, and the ensuing controversies that have arisen from the various conclusions drawn from the data purporting to show survival or quality of life benefit (or lack thereof) in those screened populations. Of course, much of this controversy owes to the different standards applied in the weighing of healthcare costs and benefits—this topic receives further attention in chapter 9, which is focused on economic evaluation in radiology.
Chapter 4 focuses on the high yield topic of intra- and interobserver variability, assumptions about which either underlie or are the research question to be answered in many diagnostic imaging studies. The authors explain how to quantify the variability or disagreement given by two readers (or two imaging modalities) making the same measurements on the same subjects, invoking Cohen’s kappa, intraclass correlation coefficient (ICC), and Bland-Altman plots.
The book’s middle chapters offer more detail on concepts first introduced in the opening chapter: observational design and randomized controlled trials. It goes on to include a chapter on systematic reviews that also tackles issues pertaining to translation of research findings to clinical practice. There is a timely discussion of decision analysis in Chapter 8, which touches on radiologists’ use (and some might say dependence) of heuristics in image analysis, and the susceptibility to bias engendered by those “mental shortcuts.” There are also chapters focused on how to comply with institutional and national safety and ethics regulations, the draft and publishing of manuscripts, and selecting appropriate funding mechanisms that can help one navigate a path to academic promotion.
The book’s final chapters deal with more advanced statistical operations, including constructing confidence intervals and conducting hypothesis tests, regression modelling, sample size estimation, and the components of a meta-analysis. Some of these topics could be challenging for readers without a background in statistics, but there is an online appendix containing additional data sets and problems with solutions worked through using R, a free statistical software package.
Overall, this text is well-worth reading if embarking on an academic career or wishing to improve one’s ability to critically appraise the radiology literature. I could easily see certain key chapters, such as those covering research design, diagnostic test performance, measurement validity/reliability, and manuscript drafting as forming the core of a brief primer for residents or medical students as they initially engage in radiology research. More advanced readers will find the book’s comprehensive nature and radiology-specific focus make it a practical reference to turn to for guidance in selecting appropriate statistical analyses and enhancing their clarity in communicating their research findings.
