Predictors of the Efficacy of Focused Ultrasound for Moderate-to-Severe Persistent Allergic Rhinitis: A Retrospective Cohort Study.

BACKGROUND

Focused ultrasound (FUS) has achieved favorable results in the treatment of allergic rhinitis (AR). However, some patients still have poor outcomes, and there are no preoperative, especially machine learning (ML)-based tools to predict them.

OBJECTIVE

This paper intends to explore the predictors of the efficacy of FUS in AR and, for the first time, to construct a ML-based predictive model for poor outcomes.

METHODS

Clinical data of patients with moderate-to-severe AR receiving FUS were collected from the otolaryngology of the Central Hospital of Wuhan from 2019 to 2023. Patients were assigned to training and internal validation sets in a 3:1 ratio. In the training set, LASSO regression was employed for variable screening, and four machine learning models were constructed. The area under the receiver operating characteristic curve (AUROC) was leveraged to assess the prediction accuracy. The calibration curve was plotted to measure the calibration of models. The decision curve and clinical impact curve were plotted to assess clinical applicability.

RESULTS

A total of 330 AR patients were included (247 in the training set and 83 in the validation set). Disease duration, peripheral blood eosinophil (PBE) count, comorbid chronic rhinosinusitis (CRS), nasal endoscopy scores, preoperative TNSS and endoscopy total scores were identified as predictors of efficacy. Logistic regression and XGBoost were the most accurate. The AUROC of logistic regression and XGBoost was 0.732 (95% CI: 0.647-0.816) and 0.811 (95% CI: 0.741-0.881) in the training set, and 0.788 (95% CI: 0.682-0.896) and 0.809 (95% CI: 0.707-0.912) in the validation set.

CONCLUSION

The ML-based models can help clinicians identify patients with refractory AR early, providing a theoretical basis for clinical decision-making. This preliminary model was constructed based on single-center data, and only internal validation was performed. Therefore, its clinical applicability should be interpreted with caution, and external validation is needed. Future research should update and refine the model.