Fewer than half of masseter trigger point injections succeed, but an algorithm can now flag who will benefit before the needle enters the muscle.
The cost of trial and error
Trigger point injections for masseter myofascial pain syndrome are widely used, yet outcomes remain erratic. In a multicenter study of 1,181 patients, only 43.1% achieved composite treatment success at three months. That leaves nearly six in ten patients undergoing invasive interventions without gaining meaningful clinical relief.
That high failure rate wastes clinical resources and delays effective care.
To reduce trial-and-error care, researchers trained Random Forest and XGBoost models to predict multidimensional success. Treatment success required meeting three criteria simultaneously: a pain reduction of ≥ 2 points on the Visual Analog Scale, a mouth opening increase of ≥ 5 mm, and a quality-of-life improvement of ≥ 5 points on the OHIP-14 scale.
Validation drops outside internal cohorts
On internal validation, the algorithms achieved high diagnostic accuracy. Random Forest scored a ROC-AUC of 0.914, while XGBoost reached 0.888. However, accuracy declined when tested on a geographically independent external cohort, yielding ROC-AUC values of 0.771 for Random Forest and 0.787 for XGBoost.
That performance drop highlights why internal validation alone cannot justify clinical adoption.
Despite the accuracy loss, external PR-AUC values remained solid at 0.759 for Random Forest and 0.761 for XGBoost. SHAP analysis identified baseline mouth opening, quality-of-life scores, patient age, pain intensity, and the chosen injectable modality as the primary drivers of success. These findings align with prior research on predictive modeling in myofascial pain syndrome, confirming that initial disease severity strongly dictates therapeutic response across cohorts.
Translating predictions to the clinic
The team deployed the validated XGBoost model as a web-based clinical decision support tool. Decision curve analysis demonstrated net clinical benefit across relevant threshold probabilities, giving practitioners a practical way to evaluate six injectable options before initiating treatment.
Key quantitative findings from the validation process include:
- 43.1% overall composite success rate across 1,181 patients receiving masseter injections.
- Internal ROC-AUC of 0.914 (Random Forest) and 0.888 (XGBoost).
- External validation ROC-AUC of 0.771 (Random Forest) and 0.787 (XGBoost).
- External PR-AUC of 0.759 (Random Forest) and 0.761 (XGBoost).
Retrospective model validation demonstrates statistical discrimination, but it is not proof of improved patient outcomes or reduced healthcare costs. Clinicians should view this tool as decision support rather than an automated decision-maker until prospective trials confirm that model-guided selection reduces treatment failure rates in real-world practice.
Read the full study in BMC Oral Health.
