Background: Traumatic brain injury (TBI) has physical consequences (e.g., loss of mobility, muscle strength, balance and walking ability) which often have long-term effect on daily functioning. Physical therapy (PT) plays an important role in the functional recovery and is frequently applied in the acute/sub-acute phases of injury. However, less is known about the use of PT in the later stages of injury and predictors of probability trajectories in a long-term perspective.
Aims: To examine probabilities and baseline predictors of PT use across 10 years after TBI.
Methods: Longitudinal observational study of 97 survivors who sustained moderate-to-severe TBI in 2005 - 2007 in the South-Eastern region of Norway. Socio-demographics and injury characteristics were recorded at baseline. PT use was recorded at 1-, 2-, 5- and 10-year follow-ups. Hierarchical linear modelling (HLM) was applied to examine PT use probabilities across the four time points. Time, gender, age, relationship status, education, pre-injury employment and occupation, cause of injury, GCS, duration of posttraumatic amnesia (PTA), CT findings (Marshall score), and injury severity score (ISS) were considered as predictors, as well as the interaction terms between time and the significant predictors.  
Results: Overall, the use of PT deceased from 55 % at 1-year to 10 % at 10-year follow-up. The first HLM of PT use probabilities showed a significant cubic, or S-shaped, curve, suggesting a steep initial decrease up to 2 years after injury, plateauing, and then a final decrease. Gender, PTA, and CT severity score yielded statistically significant effects on participants’ PT use probability trajectories. Women (p = 0.003) and individuals with a longer duration of PTA (p < 0.001) had higher probability of PT use across the four time points. Finally, individuals with a higher CT severity score (more severe injury) had a lower probability of PT use (p = 0.02); however, on visual examination of the graph it seems this was a statistical suppressor effect due to low-to-moderate multicollinearity among the predictors devoid of substantive interpretation. A second HLM examined whether PT use probability trajectories could be predicted by the previously significant predictors (gender and PTA), as well as their interactions with cubic time. Neither of the interaction terms was statistically significant (all ps > 0.59), suggesting that PT use probabilities did not change differentially over time as a function of either these predictors.
Conclusion: This study presents a novel model for the probability of PT use following TBI. Findings can be applied to inform clinicians and decision makers on patients’ needs for PT from a long-term perspective.

Physical therapy use probabilities across 10 years after moderate-to-severe traumatic brain injury