Abstract
In this issue of Journal of Neurotrauma, we are proud to present a terrific lineup of articles on several important topics related to traumatic brain injury.
I'd like to draw special attention to the topic of statistical rigor in neurotrauma research. In every part of our field, the opportunities for sophisticated, large-scale data collection are growing by leaps and bounds. But there are also pitfalls that must be avoided (or at least acknowledged) if we are going to make progress toward improved outcomes for patients with brain and spinal cord trauma. In an article entitled “Prognostic Research in Traumatic Brain Injury: Markers, Modeling, and Methodological Principles,” Retel Helmrich et al. present a carefully reasoned call for transparent reporting, explicit model specification, and assessment of potential non-linear effects. I heartily agree with their affirmation that collaborations between multiple centers will substantially improve generalizability. In another article entitled “Biomarkers for Traumatic Brain Injury: Data Standards and Statistical Considerations” Huie et al. present a set of management concepts for imaging, genomic, proteomic, and other biomarker research. They clearly lay out three phases: discovery, evaluation, and evidence synthesis. Much of the work in the field is best characterized as “discovery” but I am optimistic that as our field matures, we will see more solid articles in the domains of evaluation and synthesis. A third article entitled “Statistical Guidelines for Handling Missing Data in Traumatic Brain Injury Clinical Research” from Nielson et al. presents an introduction to a ubiquitous challenge faced in virtually every domain of clinical research: missing data. If handled poorly, missing data can undermine the validity or weaken the power of clinical studies. But if handled well, researchers can make optimal use of the data collected to draw sound conclusions. I personally look carefully at the statistical methods in the articles I read to see how missing data have been treated in helping inform my judgement about the robustness of the conclusions.
Two of the regular articles in this issue are also notable for their sophisticated use of statistical methods. In an article entitled “Accelerated Brain Aging in Mild Traumatic Brain Injury: Longitudinal Pattern Recognition with White Matter Integrity” Gan et al. reported the use of a machine learning approach called “relevance vector regression” (part of the relevance vector machine technique) to derive a model from 523 healthy individuals, and then applied it to 116 acute mTBI patients and 63 additional healthy controls. Based on the model, the apparent brain age of the mTBI patients was about 2.5 years older than expected. It will be fascinating to see what this finding implies in terms of neurodegeneration and other long-term outcomes. In another article entitled “Effect of Steroids as an Adjunct to Surgical Treatment in Patients with Chronic Subdural Hematoma,” Lodewijkx et al. used an important statistical method called “propensity score matching” to select two cohorts of patients from a retrospective database that differed in that one was treated with corticosteroids and the other was not treated with corticosteroids, but otherwise had similar baseline characteristics. The propensity score matching approach is meant to attempt to approximate the effects that might be seen in a prospective randomized controlled trial. The approximation can be good if there are no important factors affecting the outcomes that are not measured, but not so good if there are major factors that are not measured. The authors found that there was not a statistically significant difference between groups in several relevant clinical outcomes. This finding matches well with the results of an actual prospective randomized controlled trial reported in Journal of Neurotrauma earlier this year (
Looking toward the future, the use of sophisticated statistical and analytical techniques will be increasingly important and have tremendous potential to improve the quality of research in our field. Even those of us who are not statisticians should read these articles carefully and keep their lessons in mind when designing and reporting our studies.