Comparison Of Commercial Artificial Intelligence Applications For Predicting Siegel Outcomes In IdiopathicSudden Sensorineural Hearing Loss
XXIII. World Congress of Otorhinolaryngology Head and Neck Surgery IFOSTANBUL 2026, İstanbul, Türkiye, 9 - 13 Eylül 2026, (Özet Bildiri)
- Yayın Türü: Bildiri / Özet Bildiri
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu
- Sivas Cumhuriyet Üniversitesi Adresli: Evet
Özet
BACKGROUND & AIM
Idiopathic sudden sensorineural hearing loss is an otologic emergency characterized by rapid-onset sensorineural hearing
impairment and variable clinical outcomes. Although systemic corticosteroid therapy is commonly used, treatment response
may differ considerably among patients. Early prediction of hearing recovery may contribute to patient counseling, treatment
planning, and clinical decision support.
This study aimed to evaluate the ability of commercial artificial intelligence applications to predict post-treatment hearing
recovery according to Siegel criteria using only pre-treatment clinical, audiological, and laboratory data in patients with
idiopathic sudden sensorineural hearing loss.
MATERIAL & METHODS
This retrospective single-center methodological study included patients diagnosed with idiopathic sudden sensorineural
hearing loss at Sivas Cumhuriyet University Department of Otorhinolaryngology between January 1, 2016, and the date of data
extraction. Patients treated with systemic prednisolone therapy and having both pre- and post-treatment audiometric records
were included. Demographic characteristics, pretreatment pure-tone audiometry thresholds, and routine laboratory
parameters were analyzed. Pure tone average (PTA) was calculated using air-conduction thresholds at 500, 1000, 2000, and
4000 Hz, and PTA gain was defined as the difference between pre- and post-treatment PTA values. Real Siegel classifications
were determined according to standardized PTA-based criteria. For artificial intelligence prediction, only pretreatment
anonymized clinical, audiological, and laboratory data were provided to ChatGPT Plus, Claude, and Manus using a standardized
prompt. Post-treatment audiometric findings and actual Siegel classes were withheld during prediction. AI-generated
predictions were compared with actual clinical outcomes using four-class Siegel classification accuracy.
RESULTS
A total of 117 patients were included in the analysis. The mean pre-treatment pure-tone average (PTA) was 49.5 dB, and the
mean post-treatment PTA was 38.5 dB, with a mean PTA gain of 11.0 dB. According to real post-treatment audiometric
outcomes, 54 patients were classified as Siegel I, 8 as Siegel II, 7 as Siegel III, and 48 as Siegel IV. Overall recovery (Siegel I–III)
was observed in 59.0% of patients.
For four-class Siegel prediction, artificial intelligence applications demonstrated limited accuracy. ChatGPT Plus achieved
54.7% accuracy, Manus 49.6%, and Claude 45.3%. In the broader binary clinical classification of good response (Siegel I–II)
versus poor/insufficient response (Siegel III–IV), ChatGPT Plus achieved 71.8% accuracy with 79.0% sensitivity and 63.6%
specificity, while Manus achieved 65.8% accuracy with 82.0% sensitivity and 48.2% specificity. Overall, artificial intelligence
applications showed limited performance in detailed Siegel prediction but demonstrated more clinically useful results in
broader binary response classification.
CONCLUSIONS
Commercial artificial intelligence applications showed limited accuracy in four-class Siegel prediction for idiopathic sudden
sensorineural hearing loss, with accuracies ranging from 45.3% to 54.7%. However, binary response classification produced
more promising results, particularly for ChatGPT Plus and Manus. These findings suggest that AI tools may be more useful for
broad prognostic assessment than precise Siegel class prediction. Missing prognostic variables, including treatment delay,
vertigo, tinnitus, and salvage therapy status, may have reduced predictive performance. Although these systems cannot replace physician judgment, they may support clinical decision-making. Larger prospective multicenter studies are needed to validate AI-assisted prediction models.
KEYWORDS
Idiopathic sudden sensorineural hearing loss; artificial intelligence; Siegel criteria; pure tone average; PTA gain; hearing recovery; corticosteroid treatment.