Article Text

Download PDFPDF
436 Application of multi-state survival methods to randomized clinical trials of concussion rehabilitation
  1. Stephen Marshall1,2,3,4,
  2. Johna Register-Mikalik1,3,4,
  3. Paula Gildner1,
  4. Michael McCrea5,
  5. Kevin Guskiewicz1,3,4
  1. 1University of North Carolina Injury Prevention Research Center, Chapel Hill, USA
  2. 2Epidemiology Department, Gillings School of Global Public Health, Chapel Hill, USA
  3. 3Department of Exercise and Sport Science, Chapel Hill, USA
  4. 4Matthew Gfeller Sports-Related Traumatic Brain Injury Research Center, Chapel Hill, USA
  5. 5Center for Neurotrauma Research, Medical College of Wisconsin, Milwaukee, USA


Background International protocols for managing sport-related concussion are based on progression across clinically-defined recovery stages. Recent randomized clinical trials have investigated whether controlled rehabilitation exercises accelerate progression through stages of recovery. This presentation compares two statistical techniques for analysis of post-concussion recovery time.

Methods Two key stages in concussion recovery are: post-injury, prior to symptom-resolution (denoted here as State A); and post-symptom-resolution, prior to return-to-play-clearance (denoted here as State B). Statistical analyses can be misleading if researchers ignore (A,B) state transitions. To illustrate the issue, we compared traditional survival models (incorrectly ignoring state transitions) to multi-state survival methods (accounting for state transitions). The example data source was a two-arm cluster-randomized trial of 251 prospectively-registered sport-related concussions at 28 participating sites across three countries.

Results In preliminary analyses, traditional Cox models (ignoring states) falsely suggested a treatment-by-time interaction. Subsequent multi-state Cox models (accounting for states) highlighted the fact that treatment effects were present in one state (Hazard Ratio [HR] 1.88) but not the other (HR 1.05), thereby generating the illusion of time-interaction. Regression models were consistent with multi-state survival curves (obtained by differencing Kaplan-Meier estimators) and time-in-state descriptive statistics. Both traditional Cox and multi-state Cox survival models correctly identified the main effect of treatment arm for overall time (sum of A+B; injury to return-to-play-clearance, HR 1.38).

Conclusion Multi-state survival methods are the preferred method for analysis of post-concussion clinical interventions.

Learning Outcome Multi-state time-to-event models outperform traditional time-to-event models in analysis of recovery-time outcomes for sports-related concussion.

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.