A Machine Learning Model Including Questionnaire and Structural Imaging Data Predicts Headache Improvement in Patients with Acute Post-traumatic Headache Attributed to Mild Traumatic Brain Injury
Catherine Chong1, Jing Li2, Todd Schwedt1, Visar Berisha3, Devin Nikjou3, Teresa Wu3, Gina Dumkrieger1, Katherine Ross4, Lingchao Mao2
1Mayo Clinic, 2Georgia Institute of Technology, 3Arizona State University, 4Phoenix VA Health Care System
Objective:
To assess the accuracy of a model including clinical questionnaire and brain structural measures for predicting headache improvement in patients with acute post-traumatic headache (aPTH) attributed to mild traumatic brain injury (mTBI).
Background:
Our prior work demonstrated that questionnaires assessing psychosocial symptoms have utility of predicting PTH persistence. Here, we aimed to determine whether classification accuracy was improved when adding structural brain measures to the model.
Design/Methods:
Adults with aPTH (enrolled 0-59 days post-mTBI) underwent T1-weighted brain MRI and completed three selected questionnaires (Sports Concussion Assessment Tool-symptom checklist, Pain Catastrophizing Scale, and the Trait Anxiety Inventory Scale). Total and sub-scores of questionnaires and automated segmentations of regional brain volume, thickness, curvature, and area measures were used to train a dimension reduction model followed by prediction models of headache improvement. In addition, aPTH vs healthy control (HC) brain imaging data were compared.
Results:
43 patients with aPTH (mean age=43.0, SD=12.4) and 61 HC were enrolled (mean age= 39.1, SD=12.8). At three-month follow-up, 26 aPTH patients had headache improvement and 17 did not. The final model trained on questionnaire data and brain measurements achieved cross-validation Area Under the Curve classification accuracy of 0.801 for predicting headache improvement. The most highly contributing model features included questionnaires assessing psychosocial symptoms and brain structural measurements including curvature and thickness in of superior, middle, and inferior temporal, fusiform, inferior parietal, and lateral occipital regions. At baseline, those with aPTH who improved by three months had less severe symptoms of mTBI, pain catastrophizing, and anxiety than those who did not improve. Furthermore, aPTH patients whose headaches did not improve had significantly greater baseline differences in brain structure vs. HC compared to those whose headaches did improve vs. HC.
Conclusions:
A model including clinical questionnaire data and measures of brain structure accurately predicted three-month headache improvement in patients with aPTH.