A comparison of logistic regression and classification tree to assess brucellosis associated risk factors in dairy cattle

Abstract

Machine learning approaches have been increasingly utilized in the field of medicine. Brucellosis is one of the most common contagious zoonotic diseases with significant impacts on livestock health, reproduction, production, and public health worldwide. Therefore, our objective was to determine the seroprevalence and compare the logistic regression and Classification and Regression Tree (CART) data-mining analysis to assess risk factors associated with Brucella infection in the densest cattle populated Egyptian governorates. A cross-sectional study was conducted on 400 animals (383 cows, 17 bulls) distributed over four Governorates in Egypt's Nile Delta in 2019. The randomly selected animals from studied geographical areas were serologically tested for Brucella using iELISA, and the animals' information was obtained from the farm records or animal owners. Eight supposed risk factors (geographic location, gender, herd size, age, history of abortion, shared equipment, and disinfection post-calving) were evaluated using multiple stepwise logistic regression and CART machine-learning techniques. A total of 84 (21.0%; 95% CI 17.1-25.3) serum samples were serologically positive for Brucella. The highest seroprevalence of Brucella infection was reported among animals raised in herd size > 100 animals (65.5%), with no disinfection post-calving (61.7%), with a history of abortion (59.6%), and with shared equipment without thorough cleaning and disinfection (57.1%). The multiple stepwise logistic regression modeling identified herd size, history of abortion, and disinfection post-calving as important risk factors. However, CART modeling identified herd size, disinfection post-calving, history of abortion, and shared equipment as the most potential risk factors for Brucella infection. Comparing the two models, CART model showed a higher area under the receiver operating characteristic curve (AUROC = 0.98; 95% CI 0.95 - 1.00) than the binary logistic regression (AUROC = 0.89; 95% CI 0.73 - 0.92). Our findings strongly imply that Brucella infection is most likely to spread among animals raised in large herds (>100 animals) with a history of abortions and bad hygienic measures post-calving. The CART data-mining modeling provides an accurate technique to identify risk factors of Brucella infection in cattle.

Keywords: Brucellosis; Dairy cattle; Decision tree; Risk factors; Seroprevalence.