Banca de DEFESA: Yuri Almeida de Oliveira

Uma banca de DEFESA de MESTRADO foi cadastrada pelo programa.
STUDENT : Yuri Almeida de Oliveira
DATE: 29/05/2024
TIME: 10:00
LOCAL: Online - Plataforma TEAMS
TITLE:

MORBIDITY AND LETALITY IN COVID-19: USE OF ARTIFICIAL INTELLIGENCE FOR PREDICTING CLINICAL OUTCOMES BASED ON CLINICAL DATA FROM HOSPITAL ADMISSION

KEY WORDS:

artificial intelligence, machine learning, COVID-19, outcome prediction, clinical data

PAGES: 76
BIG AREA: Ciências da Saúde
AREA: Medicina
SUMMARY:

Introduction: In 2019, COVID-19 began one of the greatest public health challenges in history, reaching pandemic status the following year. By October 2023, there were 771,151,224 confirmed cases and 6,960,783 direct deaths. The high number of simultaneous cases, coupled with a large number of severe cases, led to the collapse of health systems. Systems capable of predicting individuals at higher risk of progressing to severe forms of the disease could optimize the allocation and direction of resources. Thus, various studies have been published in an attempt to find predictors based on artificial intelligence that, through demographic, clinical, laboratory, and medical imaging data, could identify individuals most susceptible to unfavorable disease progression. Objectives: to analyze the performance of Machine Learning (ML) based algorithms as predictors of clinical outcomes during hospitalization in COVID-19 patients, using hospital admission data. Methodology: data collected from a prospective, multicenter cohort of hospitalized COVID-19 patients were used. Admission and outcome data were pre-processed, taking into account the availability of data among study participants and clinical relevance. Through experimentation, various artificial intelligence based algorithms were applied, and the best were validated using Monte Carlo Cross-Validation. Results: artificial intelligence-based classifiers were successfully developed, achieving an 80% accuracy rate in predicting individuals’ progression towards unfavorable outcomes. Random Forest Classifier stands out with an Area Under the ROC Curve (AUC) of 91%, and positive and negative predictive values of 81% and 79%, respectively. Additionally, it exhibited a sensitivity of 46% and a specificity of 95%. Without significant detriment to accuracy, the Support Vector Classification (SVC) achieved a positive predictive value of 87% and specificity of 97%, rates that are suitable for situations where a small number of false positives is preferable. Conclusion: it is possible to use machine learning-based algorithms to predict unfavorable clinical outcomes during hospitalization for COVID-19 with satisfactory accuracy.

COMMITTEE MEMBERS:
Presidente - 2676451 - CIRO MARTINS GOMES
Externo ao Programa - 404943 - ADSON FERREIRA DA ROCHA - UnBExterna à Instituição - Juliana Tessari Dias Rohr - SES-DF
Externo à Instituição - TULIO FRADE REIS - OO