# Analysis of genetic polymorphisms in sudden sensorineural hearing loss and artificial intelligence-supported individualized precision therapy

## TL;DR

This study explores how genetic differences and AI models can help personalize treatment for sudden hearing loss, finding that a combination steroid injection and specific genetic markers improve outcomes.

## Contribution

The study introduces an AI-based model for personalized SSNHL treatment and identifies key genetic polymorphisms linked to treatment response.

## Key findings

- Combination steroid injections showed better recovery rates than systemic steroid therapy.
- Genetic polymorphisms in GJB2, SLC26A4, TNF-α, and CYP3A4 correlate with treatment outcomes.
- A CNN model outperformed traditional models in predicting treatment efficacy.

## Abstract

Sudden sensorineural hearing loss (SSNHL) is characterized by acute onset, complex pathogenesis, and visible variability in prognosis, making precise treatment challenging. This article focuses on identifying key factors influencing the treatment outcomes of SSNHL.

Clinical data were collected from 200 SSNHL patients, recording treatment regimens including systemic and intratympanic steroid administration. Genetic polymorphisms were analyzed via blood testing, and a personalized treatment prediction model was constructed and validated. An independent external validation set of 100 cases was included to assess model generalizability. Comparative efficacy predictions were performed among multifactorial logistic regression, convolutional neural network (CNN), random forest, and support vector machine models.

As against systemic steroid therapy, intratympanic injection, and combination treatment (corticosteroids combined with retroauricular subtympanic membrane and intratympanic injections) showed superior recovery rates. The distinction between combination treatment and monotherapy was visible (p < 0.01). At the level of key genetic polymorphisms, specific single-nucleotide polymorphism sites in genes such as GJB2, SLC26A4, TNF-α, and CYP3A4 were closely associated with treatment responses, with different genetic profiles corresponding to distinct treatment recommendations. In AI-based treatment efficacy prediction, the CNN model demonstrated significantly higher sensitivity, specificity, and accuracy compared to random forest, support vector machine, and other models (p < 0.05). It consistently outperformed traditional multifactorial logistic regression in both internal and external validation sets, particularly in identifying poor-recovery cases (p < 0.05).

In SSNHL treatment, the combined approach of postauricular subperiosteal and intratympanic steroid injections was significantly more effective than systemic steroid therapy, representing the optimal treatment choice. Specific genetic polymorphisms were closely associated with treatment response and may serve as molecular biomarkers for personalized therapy. The deep learning CNN model exhibited superior performance in efficacy prediction, surpassing conventional models, and could assist in precision treatment decision-making.