Banca de QUALIFICAÇÃO: LAURA CRISTINA DE MEDEIROS SILVA

Uma banca de QUALIFICAÇÃO de MESTRADO foi cadastrada pelo programa.
STUDENT : LAURA CRISTINA DE MEDEIROS SILVA
DATE: 07/01/2026
TIME: 14:30
LOCAL: YouTube (StreamYard)
TITLE:

Accuracy of diets designed using artificial intelligence for patients with chronic non-communicable diseases: A systematic review.

KEY WORDS:

Generative artificial intelligence; Nutritional therapy; Non-communicable chronic diseases; Machine learning; Diet.

PAGES: 69
BIG AREA: Ciências da Saúde
AREA: Saúde Coletiva
SUMMARY:

The advancement of artificial intelligence (AI) has expanded its use in clinical practice, including the automated creation of diets for the management of chronic non-communicable diseases. Despite the potential of these tools, uncertainties remain regarding their accuracy in nutrition. This study evaluated the accuracy of diets generated by AI models for individuals with NCDs, comparing them to official nutritional recommendations. This is a methodological systematic review; the study was conducted according to the Cochrane and PRISMA guidelines, with a protocol registered in PROSPERO (CRD420251056403). The review was structured according to the adapted PICOS model, considering adult patients with NCDs, the use of ChatGPT models (versions 3.5 and GPT-4), comparison with guidelines from official bodies, and accuracy metrics such as Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Searches were conducted in six databases, without language or year restrictions. The selection was conducted by two independent researchers, with resolution of disagreements by a third researcher. Of the 2,120 references identified, four studies were included, evaluating prescriptions for obesity, type 2 diabetes (DM), hypertension (HTN), cardiovascular disease (CVD), and dyslipidemia. The MAPE, which represents the average percentage deviation between values prescribed by the AI and recommended by the guidelines, was used for accuracy: <10% indicates good accuracy, 10-20% moderate, and >20% relevant limitations. For calories, accuracy varied in: CVD and DM, the MAPE was <2% indicating good accuracy, but in HTN and dyslipidemia it reached 55-56%, and in specific diets for obesity it exceeded 98%, demonstrating inadequacy. Proteins showed MAPE <5% in CVD, DM, and obesity, but were inadequate in HTN and dyslipidemia (16-18%). Carbohydrates and lipids showed the greatest deviations, with MAPE >100% for carbohydrates and 50% for fats in hypertension and dyslipidemia. Micronutrients revealed critical MAPE: 59% for calcium, 76% for sodium, and 49% for magnesium. Thus, the observed accuracy was adequate only for cardiovascular diseases and type 2 diabetes, being inadequate for hypertension, dyslipidemia, and specific diets for obesity. It is concluded that AI can act as a tool to assist nutritionists in prescribing diets; however, at the time of the included studies, the observed accuracy was still insufficient to replace the work of a professional nutritionist. Its use requires professional supervision and rigorous clinical validation to ensure the safety of evidence-based recommendations.Accuracy of diets designed using artificial intelligence for patients with chronic non-communicable diseases: A systematic review.

COMMITTEE MEMBERS:
Presidente - 2665457 - GIDYENNE CHRISTINE BANDEIRA SILVA DE MEDEIROS
Interna - 2166798 - ADALA NAYANA DE SOUSA MATA
Externa à Instituição - THAIS TEIXEIRA DOS SANTOS - IFRN