Predição de síndrome metabólica em indivíduos com doença renal crônica utilizando técnicas de aprendizado de máquina

• Ano de defesa: 2021
• Autor(a) principal: BITTENCOURT, Jalila Andréa Sampaio
• Orientador(a): NASCIMENTO, Maria do Desterro Soares Brandão
• Banca de defesa: NASCIMENTO, Maria do Desterro Soares Brandão , BARROS FILHO, Allan Kardec Duailibe , CHAGAS, Deysianne Costa das , ANDRADE, Marcelo Souza de , SILVA, Mayara Cristina Pinto da
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM SAÚDE DO ADULTO E DA CRIANÇA/CCBS
• Departamento: DEPARTAMENTO DE PATOLOGIA/CCBS
• País: Brasil
• Palavras-chave em Português: Doença renal crônica; Síndrome metabólica; Aprendizado de máquina
• Palavras-chave em Inglês: Chronic kidney disease; Metabolic syndrome; Machine learning
• Área do conhecimento CNPq: Ciências da Saúde
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/3509

The chronic kidney disease (CKD) and metabolic syndrome (MS) are closely linked to overweight, obesity and cardiovascular risk factors. In order to postpone the complications associated with them and due to the increasing incidence in all age groups, the early detection of these pathologies is necessary. Based on this, the study aimed to develop a model to predict the risk for the MS in people with the CKD. This is a cross-sectional study, carried out with patients from the Center for the Prevention of Kidney Diseases (CPDR) of the University Hospital of the Federal University of Maranhão (HUUFMA). The sample was obtained from volunteers of both genders who were 20 years old or over and were classified according to their health status (healthy or with the CKD). The stages of the CKD are classified according to the glomerular filtration rate (GFR) and the suggestive diagnosis of the MS was established according to the proposed by the International Diabetes Federation (IDF). Also, anthropometric, biochemical, hemodynamic, and lifestyle data were evaluated. For the MS tracking, the k-nearest neighbors (KNN) classifier algorithm, that is a supervised machine learning (MA) method, was used. To implement the classifier algorithm, the following entries were used: gender, smoking status, neck circumference (NC) and waist-hip ratio (WHR). The construction of the classifier algorithm and software implementation took place through the MATLAB® program. For the data file and statistical analysis, the SPSS® software was used, and the following statistical tests were applied: Kolmogorov-Smirnov, Student's t, Mann-Whitney U, in addition to the ROC curve. The results were considered statistically significant for p<0.05. This study was approved by the Ethics and Research Committee of the Federal University of Maranhão, with number 67030517.5.0000.5087. A total of 196 adult individuals with a mean age of 44.73 ± 15.96 years were evaluated, of which 71.9% (n=141) were female and 69.4% (n=136) were overweight, and 12.24% (n=24) had CKD. Of the investigated sample, 45.8% (n=11; p=0.006) of CKD patients had MS, with the majority presenting up to 3 altered metabolic components. Of these components, the group with CKD had higher mean/median values in all parameters, with statistical significance in: waist circumference (WC) (94.85±11.7; p=0.02), systolic blood pressure (SBP) [134(123.25- 165.5) mmHg; p<0.001], diastolic blood pressure (DBP) [86.5(76.25-91) mmHg; p=0.019] and fasting glucose (FG) [81(75-88) mg/dL; p=0.001]. The KNN algorithm proved to be a good predictor for MS tracking, as it had 79% accuracy and sensitivity, 80% specificity, having its performance evaluated by the ROC curve (AUC=0.79). Thus, the KNN algorithm can be used as a screening method with high sensitivity and low cost to assess the presence of MS in people with CKD.