A novel cortical biomarker signature accurately and reliably predicts individual pain sensitivity: The PREDICT longitudinal analytical validation study

Background Biomarkers would greatly assist chronic pain management. The present study aimed to undertake analytical validation of a sensorimotor cortical biomarker signature for pain consisting of two measures: sensorimotor peak alpha frequency (PAF) and corticomotor excitability (CME), using a human model of prolonged temporomandibular pain (masseter intramuscular injection of nerve growth factor [NGF]). Methods 150 participants received an injection of NGF to the right masseter muscle on Days 0 and 2, inducing prolonged pain lasting up to 4 weeks. Electroencephalography (EEG) to assess PAF and transcranial magnetic stimulation (TMS) to assess CME were recorded on Days 0, 2 and 5. We determined the predictive accuracy of the PAF/CME biomarker signature using a nested control-test scheme: machine learning models were run on a training set (n = 100), where PAF and CME were predictors and pain sensitivity was the outcome. The winning classifier was assessed on a test set (n = 50) comparing the predicted pain labels against the true labels. Results The winning classifier was logistic regression, with an outstanding area under the curve (AUC=1.00). The locked model assessed on the test set had excellent performance (AUC=0.88). Results were reproduced across a range of methodological parameters. Moreover, inclusion of sex and pain catastrophizing as covariates did not improve model performance, suggesting the model including biomarkers only was more robust. PAF and CME biomarkers showed good-excellent test-retest reliability. Conclusions This study provides evidence for a sensorimotor cortical biomarker signature for pain sensitivity. The combination of accuracy, reproducibility, and reliability, suggests the PAF/CME biomarker signature has substantial potential for clinical translation. Key Points Question Can individuals be accurately classified as high or low pain sensitive based on two features of cortical activity: sensorimotor peak alpha frequency (PAF) and corticomotor excitability (CME)? Findings In a cohort study of 150 healthy participants, the performance of a logistic regression model was outstanding in a training set (n=100) and excellent in a test set (n=50), with the combination of slower PAF and CME depression predicting higher pain. Results were reproduced across a range of methodological parameters, and inclusion of covariates did not improve model performance Meaning A novel cortical biomarker comprised of PAF and CME can accurately distinguish high and low pain sensitive individuals