Objective: To analyze factors predictive of noncompliance with pulmonary tuberculosis treatment. Methods: A historical cohort study involving 481 of the 529 active tuberculosis patients registered with the Tuberculosis Control Program in the city of Cuiabá, located in the state of Mato Grosso, during the 1998-2000 period. Data were obtained by analyzing medical charts and registration records. In the bivariate analysis, the chi-square test was used to calculate noncompliance rate ratios (relative risk), 95% confidence intervals were determined, and Fisher's exact test was used. The choice to estimate the rate of incidence was the method of density and a multivariate logistic regression model was constructed in order to identify the variables that were most predictive of noncompliance, using a level of statistical significance of p < 0.05. Results: The global rate of incidence was 27.3%, equivalent to 5.1 noncompliant patients/100 patients/month, increasing in the second and third months of treatment. In the final logistic regression model, the following were considered predictors of noncompliance: unsupervised treatment (odds ratio: 2.58; 95% confidence interval: 1.64 - 4.06; p < 0.001); having been treated during the 1998-1999 period (odds ratio: 1.43; 95% confidence interval: 1.14 - 1.80; p = 0.002); being male (odds ratio: 1.39; 95% confidence interval: 1.10 - 1.76; p = 0.005) and having been out of compliance with previous treatment regimes (odds ratio: 1.37; 95% confidence interval: 1.06 - 1.78; p < 0.017). Conclusion: The results indicate an elevated incidence of noncompliance and show that unsupervised treatment, year in which treatment was received, male gender and prior noncompliance were predictors of future noncompliance.

Keywords: Tuberculosis, pulmonary/therapy; Treatment refusal; Cohort studies