Evaluation of Survival Analysis Models for Predicting Factors Infuencing the Time of Brucellosis Diagnosis

Abstract

Background:Brucellosis or Malta fever is one of the most common zoonotic diseases in the world. In addition to causing human suffering and dire economic impact on animals, due to the high prevalence of Brucellosis in the western regions of Isfahan province, this study aimed to analyze effective factors in the time of Brucellosis diagnosis using parametric and semi-parametric models and to evaluate the goodness of fit of these models. Methods:This historical cohort study, 412 patients with Brucellosis in Fereydunshahr, Iran who had referred to hospital, rural & urban health centers and physicians' private clinics in Fereydunshahr between 2006 and 2016 were recruited through census sampling. The failure (or event) in this study, was diagnosis of Brucellosis based on positive immunologic tests (2-ME test ≥1:40 and Wright serology ≥1:80). In order to eliminate confounding variables, effective factors of the time of Brucellosis diagnosis were determined using univariate (P≤0.20) and multivariable (p <0.05) analysis according to Cox semi-parametric model and five parametric models (weibull, exponential, log-logic, log-normal and gompertz) and the best fitted model was identified. Data were analyzed using R software version 3.2.3. Results: According to the results of this study, occupation (farmer and livestock breeder), place of residence (urban), having a history of direct contact with livestock, simultaneous infection in other family members, and the newness of the disease (vs. recurrence) were identified as predictors of early detection of the disease. Conclusion: Despite the researchers' tendency to use Cox method in survival analysis, in this study, according to AIC, “Gopmpertz” parametric model was recognized as the best fitted regression model in the analysis of the effective factors in the definitive time of Brucellosis diagnosis.

Keywords

• survival analysis
• cox proportional hazard model
• parametric models
• kaplan-meier
• brucellosis