Dados do Trabalho

Título

Diagnosis Accuraccy and OSA persistence in children using the artificial intelligence.

Introdução

Obstructive sleep apnea syndrome (OSA) is characterized by episodes of upper airway obstruction, with systemic repercussions, such as craniofacial, cardiovascular, metabolic and neurocognitive alterations. The diagnosis is confirmed by polysomnography (PSG), available in only few centers. In addition, Artificial intelligence (AI) has been used in medicine, facilitating the classification and prognosis of diseases, making it possible to assess associations not found with traditional scientific methodology. We hypothesize that AI can help in the recognition of moderate-severe OSA and/or persistence of OSA after adenotonsillectomy, based on age, weight, the questionnaire Obstructive Sleep Apnea 18 (OSA-18) and sleep endoscopy.

Objetivo

To evaluate the diagnostic accuracy of the OSA-18 dataset, age, weight/age z-score and sonoendoscopic examination for the diagnosis of moderate-severe OSA, and persistence of OSA in children, using only AI.

Métodos

Data from the OSA-18, age, weight/age Z-score, PSG and sleep endoscopy of 95 children aged 4 to 9 years who underwent PSG for diagnosis and treatment of OSA at the HCFMB from 2015 to 2021 were included. Children with incomplete data, neuropathic or syndromic were excluded. Using AI, with machine learning, the diagnostic accuracy for normal-mild and moderate-severe OSAS was evaluated in two stages, as follows:
1. from OSA-18 data, age, weight and preoperative sleep endoscopy ;
2. by the set of the same data, but after adenotonsillectomy, evaluating the accuracy for the persistence of OSAS.

Resultados

The results showed high accuracy (93%) for the preoperative diagnosis of OSA, when data from the OSA-18, age and weight/age z-score were used, reflecting high sensitivity, 100% in normal-mild OSA and 88% in moderate-severe OSA, in addition to high specificity, 88% and 100%, respectively. On the other hand, sleep-endoscopy data for diagnosis showed a reduction in accuracy, making its execution unfeasible. However, to assess the persistence of OSA in children, sleep endoscopy proved to be better than the OSA-18 questionnaire, leading to an accuracy of 79%, against 75% combining OSA-18, age and weight.

Conclusões

The use of AI with machine learning proved to be effective in the diagnosis and follow-up of OSA, and may contribute to the replacement or helping the decision on PSG indication.

Palavras -chave

OSA, Sleep Endoscopy, Artificial Intelligence.