ABSTRACT
By analyzing the data on patients (n = 17295) who were discharged from 5 regional tuberculosis hospitals of Central Russia, the authors defined and assessed the major factors showing a steady-state strong correlation with early treatment termination through a patient's fault. Descriptive and univariate analyses and multifactor modeling were made to determine interfering factors and systematic errors (logistic regression modeling). For inpatient treatment at a tuberculosis hospital, a logistic regression model was used to develop a predictive decision rule that would determine the probability of treatment discontinuation, the quality control of the predictive rule on the basis of the bootstrap method. The rule makes it possible to form high and very high discontinuation risk groups in order to organize target measures to enhance the efficiency of tuberculosis control.
