Fuzzy sets theory and logistic regression in decision making for parathyroid scintigraphy investigation

Author(s)

Clóvis Arlindo de Sousa

• Document type: Master's Dissertation
• Press: São Paulo.
• Institution: Universidade de São Paulo (USP). Faculdade de Saúde Pública (FSP/CIR)
• Defense date: 2007-06-15

Examining board members

Julio Cesar Rodrigues Pereira; Laécio Carvalho de Barros; Neli Regina de Siqueira Ortega

• Advisor: Julio Cesar Rodrigues Pereira

Abstract

Introduction - A great number of parathyroid scintigraphies performed before the first surgery (parathyroidectomy) results negative suggesting an inadequate use of this test. Objective - The aim of this study was to develop and compare two mathematical models based on logistic regression analysis and fuzzy sets theory to select patients to be submitted to parathyroid scintigraphy based on serum calcium and parathormone values. Methods - From the data registrered on parathyroid scintigrafy database from Fleury - Medical Diagnostic Center, São Paulo, within the period between January 2000 to December 2004, 194 patients with available serum calcium and parathormone were identified. Two mathematical models based on logistic regression analysis and fuzzy sets theory were developed to recognize patients for parathyroid scintigraphy on the basis of laboratorial exams. The logistic regression model was developed using SPSS and the fuzzy model using MatLab. The performances of the models were compared using ROC curves with the test of Hanley e McNeil. The accuracies of both models were also analyzed and compared using a new database (test database) of 139 parathyroid scintigraphies performed between 2005 and 2006. Results - The performances of the models were statistically different (p=0.026). The area under the ROC curves were 0.86 (CI 95%: 0.81-0.91) for the logistic regression model and 0.89 (CI 95%: 0.84-0.93) for the fuzzy model. As an advantage over the logistic model, the latter has shown the ability to draw decision from parathormone information within a not discriminating interval of calcium values. In the test database the ROC area for the logistic regression model the ROC curve area was 0.89 (CI 95%: 0.83-0.95) and for the fuzzy model was 0.91 (CI 95%: 0.85-0.96) (p=0.128). Conclusions - The mathematical model based on fuzzy sets theory fared better than the logistic model in decision making for scintigraphy indication. Nevertheless, one should take heed that inferences should be kept to what concerns model comparison and not the parathyroid scintigraphy itself, once the data analysed was not representative of any population. (AU)