Truth in Numbers: Biostatistics for the Otolaryngologist Clinician
One need not be proficient in running statistical software, nor have a master’s degree in epidemiology or public health, to understand most of what is needed for critical interpretation of clinical data.
Diego A. Preciado, MD, PhD
Further, comfort with data interpretation is paramount to improving the quality of our research, and more importantly, to informing appraisal of the medical literature and the process of conducting peer review. Certainly, as we have learned the past two years, the pandemic has brought to light how important it is to recognize differences between true and anecdotal outcomes.
For the past years, the AAO-HNSF has allowed me the great privilege to discuss these topics through a lecture format at the Annual Meeting. Using an interactive case-based approach, the course has covered essential and practical information for critically appraising or conducting research. The material covered has included a discussion of normal distributions, non-parametric data, continuous versus categorical variables, effect size, confidence intervals, ρ-values, time-to-event analyses, correlations, multivariate linear and logistic regression models, and common statistical errors and misconceptions.
Highlighted areas include review of which statistical tests are needed in order to test specific hypotheses—while grasping the true meaning of significance. Although it may be important to know how ρ-values are calculated for different types of data (continuous, categorical, and time), it is even more important to understand what ρ-values actually mean. Undoubtedly, the widespread use of “statistical significance” (generally interpreted as “ρ ≤ 0.05”) as a license for making a claim of a scientific finding (or implied truth) may at times lead to considerable distortion of the scientific process. When looking at ρ-values in a vacuum, these distortions can be magnified.
Instead, ρ-values need to be considered in the context of clinical significance—nominally often represented by clinical effect size (proportional difference, relative risk, and odds ratio). Comprehending the meaning of these effect size measures and how they are calculated is perhaps more critical than being able to determine ρ-values.
A nearly reflexive understanding of different types of clinical data is paramount. As clinicians, we should move away from the lay vernacular description of populational characteristics or outcomes as “averages” to the more appropriate use of “means/medians” that account for common biases or skewing in sample distribution. Moreover, when sampling any cohort, comparatively or descriptively, it is critical to describe the “variation” (standard deviation/standard error/confidence intervals) noted in the sample, perhaps more important than to simply focus on the “mean.” Representing the sample characteristics graphically, employing tools that best depict the entire cohort (mean/median/and variance) is often the best approach.
Consider censoring of loss to follow-up subjects is of importance when discussing time-to-event occurrences. When trying to infer the relevance of a clinical variable affecting an outcome (most usually through an association, with most clinical studies not necessarily designed to determine causality), it is particularly important to consider all other potentially contributing variables in some form of multivariable regression model.
Univariable comparisons tend to be quite reductionist and may also distort inferences by missing contributions from common confounders. From a technical standpoint, running multivariable models is relatively straightforward with many software programs readily available for this purpose. An example of widely versatile, relatively inexpensive, and user-friendly data analysis program is GraphPad Prism, which also produces simple, publication-ready graphs and charts.
In conclusion, the analysis and interpretation of data is something which, like it or not, permeates our practices (and lives!). The process or arriving at “truth in numbers” for us otolaryngologists should actually be as simple and enjoyable as possible. At the end of the day, there should be nothing more rewarding than the confirmation or rejection of conceived impressions through data.
