Abstract Text: Most Traumatic Brain Injuries (TBI) are classified as mild, and symptoms are thought to resolve between 2 – 4 weeks post injury. Recent literature suggests that approximately 10-15% of mTBI patients experience chronic symptoms (> 1 year) post injury onset. Currently, objective indicators for mTBI symptoms or biomarkers do not exist, and patient self-reporting for symptom severity and recovery is relied on heavily. Symptoms are also often highly heterogeneity across patients. Post traumatic stress disorder (PTSD) is prevalent in chronic mTBI patients and chronic PTSD disrupts activities of daily living and quality of life. At present, it is unknown why some individuals recover completely, and why others develop chronic PTSD symptoms. The objective of this retrospective cohort study was to: (1) Explore factors that influence chronic post-traumatic stress (PTS) outcomes 1 year post injury in mild TBIs patients and (2) Examine the feasibility of predictive modeling in determining PTS recovery at 12 months post injury using 3 and 6 months symptoms.
Data (N = 533 subjects) for this study was was collected in the Transforming Clinical And Research Knowledge in Traumatic Brain Injury (TRACK-TBI) project obtained from the FITBIR repository. Subjects included based on the following criteria: (1) having a TBI, (2) normal brain imaging results and (3) a classification of mTBI. Data from the post traumatic check list DSM 5 (PCL-5), Patient Health Questionnaire 9 (PHQ9), Insomnia Severity Index (ISI), Rivermead Post Concussion Survey (RPCS), Brief Symptom Inventory18 (BSI18) and imaging findings were compiled. Using PCL-5 baseline and 12 months scores, subjects were divided into 4 groups - (1) Reduced symptoms (N=326) (2) No change, symptomatic (N=11), (3) Increased symptoms (N=184) and (4) No change, asymptomatic (N=12). All data analysis were performed using the SPSS software package. A one-way ANOVA was performed to identify measures of significant difference. Exploratory correlation analysis was performed to assess relationships across variables (emotional distress vs. neurological symptoms). The 12 month PCL-5 results were further divided into (1) asymptomatic group (total score <=20, N=419) and (2) symptomatic group (total score >20, N=108), and we used linear discriminant analysis to model predictive outcomes of PTS symptoms at 12 months using the Rivermead Post Concussion Survey Scores and Insomnia severity index scores at 3 months and 6 months respectively. The model was run for 1000 iterations with a leave-one-out cross validation.
Using data from the 3 month visit, the total accuracy was 84% with a positive predictive value of 93% and negative predictive value of 49%. The variables that contributed significantly to the model were ISI total score, blurred vision, irritability and sleep disturbance. Data from the 6 month visit provided 87.2% overall accuracy with a positive predictive value of 94.5% and a negative predictive value of 59%. The variables of nausea, irritability, double vision and dizziness were the most significant contributors to the model.
Results suggest 36% of subjects with chronic mTBI have PTS symptoms, 1 year post injury, higher than previously reported. Some subjects had worsening symptoms. The model weighted highly several non-emotional measures (i.e. insomnia, double vision, dizziness, nausea, etc.) at 3 and 6 months for predicting PTS outcomes at 12 months. This suggests that non-emotional symptoms have a potential impact on emotional symptom outcomes such as PTS, and has utility with predicting the progression of chronic symptoms in mTBI subjects. The results support the need for a multi-symptom treatment approach, in parallel that incorporates active management of non-emotional symptoms, that could potentially reduce PTS in chronic mTBI subjects. Some key limitations of this study include; (1) potential bias in the self-report measures, (2) small sample size, (3) missing data, and (4) low negative predictive value of the model. Future work will include the incorporation of biomarker, genetic and blood work results to identify potential predispositions that may exist or other hidden factors that impact recovery or the onset of chronic symptoms. Future work will refine and optimize this and compare the predictions with other methods such as neural networks or random forest for increased accurate, predictive power and translation for generalization of predictive ability.
