The return to play (RTP) process may occur during longitudinal studies tracking recovery after concussion. This factor, which is often omitted within statistical designs, could affect the fit and overall interpretation of the statistical model. This article demonstrates the difference in results and interpretation between 2 linear mixed-model designs: (1) a between-group longitudinal (GROUP) analysis and (2) a between-group longitudinal model that used an inflection point to account for changes around the time of RTP (RTP analysis). These analyses were conducted on instrumented balance data collected on 23 concussed athletes and 25 controls over 8 weeks following concussion. Total sway area and the range of mediolateral acceleration were used as outcome measures. No significant findings were found in the GROUP design for either outcome measure. In contrast, the RTP analysis revealed significant effects of time (P = .007) and RTP change (P = .007), and group*time (P = .028) and group*RTP change (P = .022) interactions for total sway area, and effects of group (P = .011), time (P = .010), and RTP change (P = .014), and group*time (P = .013) and group*RTP change interactions (P = .013) for range of mediolateral acceleration. For both outcomes, the RTP model fit the data significantly better on comparison of likelihood ratios (P ≤ .027). These results suggest that allowing for an inflection point in the statistical design may assist understanding of what happens around clinically meaningful time points. The choice of statistical model had a considerable effect on the interpretation of findings, and provokes discussion around the best method for analyzing longitudinal datasets when important clinical time points like RTP exist.
Keywords: inertial sensors; linear mixed model; mTBI; return to sport; wearable.
© 2018 John Wiley & Sons A/S. Published by John Wiley & Sons Ltd.