Dados do Resumo

Título

Artificial Intelligence as a Tool for Predicting Postoperative Complications in Patients Undergoing Major Oncological Head and Neck Surgeries

Introdução

Patients undergoing major oncological surgeries of the head and neck face a high risk of postoperative complications, especially surgical site infections (SSIs). Consequently, these infections prolong hospital stays and increase healthcare costs. Strategies to minimize these events include preoperative decolonization, which involves oral hygiene and chlorhexidine baths with nasal mupirocin for five days before surgery. Studies indicate that these practices can reduce the incidence of SSIs and hospitalization time, improving treatment adherence and other important factors for successful cancer care. With advances in artificial intelligence (AI) techniques, there is a promising opportunity to enhance patient experiences by analyzing large volumes of complex data and identifying critical variables, thus enabling early and personalized interventions to achieve better clinical outcomes.

Objetivo

To develop and validate an AI model to predict postoperative infections in patients undergoing major oncological head and neck surgeries.

Métodos

Retrospective data from electronic health records of patients who underwent major oncological head and neck surgeries at a cancer center in São Paulo from 2014 to 2023 will be used to build an AI model capable of predicting postoperative infections. The model will be trained with anonymized data using machine learning techniques such as XGBoost and evaluated through cross-validation. After building the model, results will be interpreted using SHAP (Shapley Additive ExPlanations), which will determine the variables that most contribute to prediction and are associated with the outcome. This study was approved by the Ethics Committee, approval number: 6.873.237, CAEE: 79192024.1.0000.5432.

Resultados

From 2014 to 2023, 1,028 major oncological head and neck surgeries were performed. Postoperative infections within 30 days occurred in 136 (13.2%) cases. Regarding the financing of the procedures, 748 (72.8%) were privately funded, while 280 (27.2%) were funded by the public health system (SUS). In terms of gender distribution, 666 (64.8%) patients were male and 362 (35.2%) were female. The ASA classification of patients showed that 654 (63.6%) were ASA II, 246 (23.9%) ASA III, 104 (10.1%) ASA I, and 22 (2.1%) ASA IV. These data correspond to the clinical characteristics of the samples that will be used for the construction and validation of the AI model in the present proposal.

Conclusões

Early detection of complications is of particular interest in an oncology hospital. In this research project, we propose to develop an AI method to predict infections following major surgery. As a starting point, we characterized a cohort of patients treated at the A.C. Camargo Cancer Center over 10 years (2014 to 2023). Combined with other clinicopathological information, an AI model will be developed to create a decision support system. Ultimately, the goal is to enhance personalized cancer patient care, optimize resource use, and reduce costs.

Financiador do resumo

CAPES (Coordenação de Aperfeiçoamento de Pessoal de Nível Superior).

Palavras Chave

Artificial Intelligence; Postoperative Complications; Head and Neck Surgery