نوع مقاله : مقاله های پژوهشی
1 کارشناس ارشد آمار زیستی، مرکز توسعهی تحقیقات بالینی، بیمارستان امام رضا (ع)، دانشگاه علوم پزشکی کرمانشاه، کرمانشاه، ایران
2 استادیار، گروه ریاضی، پردیس بیجار، دانشگاه کردستان، سنندج، ایران
3 استادیار، گروه ریاضی، دانشکدهی علوم پایه، دانشگاه کردستان، سنندج، ایران
4 کارشناس ارشد اپیدمیولوژی، مرکز تحقیقات بیماریهای عفونی، بیمارستان امام رضا (ع)، دانشگاه علوم پزشکی کرمانشاه، کرمانشاه، ایران
عنوان مقاله [English]
Background: Stomach cancer is the fifth most common disease and the third cause of death in the world. Therefore, in order to prevent and reduce the incidence of stomach cancer, factors related to logistic regression and neural network models were investigated.
Methods: In this study, a survey was conducted on 1,170 people as (n = 390) cases and (n = 780) controls. The data collection tool was based on the researcher's checklist. The samples were selected by available sampling method and their information was collected by face-to-face and telephone interviews. The fitting power in the logistic regression model and neural network was compared with receiver function characteristic curve (AUROC), sensitivity and specificity. By introducing the superior model, significant and related factors with stomach cancer were reported.
Findings: The results showed that the accuracy, sensitivity and specificity of the neural network were 96.4%, 93.7% and 81.9%, respectively. But the accuracy, sensitivity and specificity of the logistic regression model were reported as 95.9%, 91.1% and 84.4%, respectively. The neural network model indicates the variables of age (0.646), fruit consumption (0.713), history of self-medication (0.652), history of gastric ulcer (0.734), family history of cancer (0.852) and Family history of stomach cancer (0.836) were associated with the incidence of stomach cancer.
Conclusion: Considering that in the present study, the fit of the neural network was superior to logistic regression and it does not need any special assumptions, so it is suggested to the researchers that the neural network model can be preferred over logistic regression.