Modeling the survival of patients with tuberculosis based on the model of artificial neural networks

Kazempour Dizaji, Mehdi and Marjani, Majid and Tabarsi, Payam and Varahram, Mohammad and Zare, Ali and Ali Emamhadi, Mohammad Ali and Roozbahani, Rahim and Abedini, Atefe and Baghaei Shiva, Parvaneh and Moniri, Afshin and Madani, Mohammadreza (2022) Modeling the survival of patients with tuberculosis based on the model of artificial neural networks. Health Science Monitor, 1 (1).

Text 8 Kazempour Dizaji A-10-76-1.pdf Download (399kB)

Abstract

Background & Aims: The development of treatment methods and increasing the survival of patients with tuberculosis (TB) has led to the complication of relationships between independent and dependent variables associated with this disease. Therefore, it is important to use new methods to model the TB process that can accurately estimate the current situation. This study aimed to model the survival of patients with tuberculosis based on the model of perceptron artificial multilayer neural network (MLP-ANN). Materials and Methods: In this retrospective cohort study, the data was collected from 2366 TB patients who were treated in Dr. Masih Daneshvari Hospital in Tehran from 2005 to 2015. To model the predictive power of survival in TB patients, an MLP-ANN model consisting of three layers was applied. Results: The results of this study showed that based on the MLP-ANN model, the correct predictive power of survival in TB patients is 88.4%. In this study, the variables of patients' age and family size as very effective variables also variables of patients’ gender, marital status, education, adverse drug effects, exposure to cigarette smoke, imprisonment, pulmonary tuberculosis, and AIDS as effective variables in predicting the survival of patients were diagnosed. Conclusion: In the model of artificial neural networks, no restrictions are considered for the data structure and the type of relationship between variables. Therefore, these models with their flexibility and high accuracy can be one of the best methods for modeling health data.

Actions (login required)

View Item