Originally published by our sister publication Pain Medicine News

Evidence from a new study suggests that artificial intelligence can aid clinicians, with and without expertise, in effectively utilizing patients’ drawings to predict headache surgery outcomes.

Patients’ drawings of pain have been shown to help predict poor outcomes in headache surgery. However, interpreting these drawings requires training and experience that many nonspecialized clinicians lack.

The study included 131 headache surgery patients (77% female, mean age, 46.4 years) (Plast Reconstr Surg 2023;151[2]:405-411). Researchers trained a random forest machine learning algorithm on pain drawings provided prospectively by each of the patients prior to trigger-site deactivation surgery. They selected 24 features to describe the anatomic distribution of pain on each drawing for interpretation by the algorithm.

After an initial pain drawing and a Migraine Headache Index (MHI) score was recorded, study participants underwent trigger-site deactivation surgery for headaches. Trigger sites included the greater occipital nerve (73.3%), supraorbital and supratrochlear nerves (48.9%), zygomaticotemporal nerve (31.3%), lesser occipital nerve (15.3%), and auriculotemporal nerve (7.6%). Mean follow-up was 13.7 months.

The researchers stated the algorithm weighted diffuse pain, facial pain and pain at the vertex as strong predictors for poor surgical outcomes.

The algorithm was consistently more accurate than trained clinical evaluators (TCEs) at correctly predicting patients’ responses to surgery. AI was most effective at predicting if a surgery was likely to have a poor outcome, defined as less than a 20% improvement in a patient’s MHI score. The algorithm had a higher predictive accuracy for poor surgical outcomes at 94% compared with the accuracy of TCEs at 79%.

AI’s rate of accuracy was also higher (91%) compared with the accuracy of TCEs (82%) when predicting more than 50% improvement in a patient’s MHI score (P<0.05). The predictive accuracy was similar when predicting more than 80% improvement in a patient’s MHI score (84% and 82%, respectively; P<0.05).

According to the researchers, diffuse pain, pain at atypical sites and facial pain were strong predictors of poor surgical outcomes.

“Patients diagnosed with migraines and headaches can suffer from nerve pain. It is important to recognize nerve pain, as patients may benefit from nerve decompression surgery. AI may help us screen patients for this surgery,” said researcher Lisa Gfrerer MD, PhD, an assistant professor of surgery at Weill Cornell Medicine, in New York City. “Larger scale clinical testing has to be done in order to verify our findings.”

—Myles Starr

Gfrerer reported no relevant financial disclosures.