Post-concussion symptom burden and dynamics: Insights from a digital health intervention and machine learning

Individuals who sustain a concussion can experience a range of symptoms which can significantly impact their quality of life and functional outcome. This study aims to understand the nature and recovery trajectories of post-concussion symptomatology by applying an unsupervised machine learning approach to data captured from a digital health intervention (HeadOn). As part of the 35-day program, patients complete a daily symptom diary which rates 8 post-concussion symptoms. Symptom data were analysed using K-means clustering to categorize patients based on their symptom profiles. During the study period, a total of 758 symptom diaries were completed by 84 patients, equating to 6064 individual symptom ratings. Fatigue, sleep disturbance and difficulty concentrating were the most prevalent symptoms reported. A decline in symptom burden was observed over the 35-day period, with physical and emotional symptoms showing early rates of recovery. In a correlation matrix, there were strong positive correlations between low mood and irritability (r = 0.84), and poor memory and difficulty concentrating (r = 0.83). K-means cluster analysis identified three distinct patient clusters based on symptom severity. Cluster 0 (n = 24) had a low symptom burden profile across all the post-concussion symptoms. Cluster 1 (n = 35) had moderate symptom burden but with pronounced fatigue. Cluster 2 (n = 25) had a high symptom burden profile across all the post-concussion symptoms. Reflecting the severity of the clusters, there was a significant relationship between the symptom clusters for both the Rivermead (p = 0.05) and PHQ-9 (p = 0.003) questionnaires at 6-weeks follow-up. By leveraging digital ecological momentary assessments, a rich dataset of daily symptom ratings was captured allowing for the identification of symptom severity clusters. These findings underscore the potential of digital technology and machine learning to enhance our understanding of post-concussion symptomatology and offer a scalable solution to support patients with their recovery.Author summary: Individuals who sustain a concussion can experience a range of post-concussion symptoms that can significantly impact their quality of life. In this study, we used digital technology to capture daily symptom ratings from 84 patients over a 35-day period as they recovered from a concussion. This provided a rich dataset of over 6,000 individual symptom assessments, one of the largest collections of post-concussion symptom data to date. We found that fatigue, sleep disturbance, and difficulty concentrating were among the most prevalent and severe symptoms reported by patients. While the overall symptom burden declined over the 35 days, there was variation in how quickly different symptoms improved, with physical and emotional symptoms showing early recovery. By applying machine learning techniques, we were able to identify three distinct clusters of patients based on their symptom severity profile. These symptom clusters correlated with outcomes like depression and concussion symptomatology scores at 6 weeks. Our findings underscore how digital tools and data science approaches can provide novel insights into post-concussion symptomatology. Capturing granular, daily symptom data allowed us to characterize patient trajectories and identify differing symptom burden phenotypes. This scalable approach could help guide more personalized management strategies for patients recovering from concussion.