As one of the states with the highest tuberculosis (TB) burden in the United States, Texas had TB identified as the most distinctive cause of mortality from 2001-2010 with 679 deaths. Texas also had a TB rate of 4.9/100,000 population in 2015, a 4.3% increase from 2014. Lacking a standardized prognostic system, predicting mortality risk during TB treatment remains a challenge for health care providers. The proposed population-based analysis and modeling are initial attempts to address the situation.
De-identified surveillance data from the Centers for Disease Control and Preventions TB Genotyping Information Management System was analyzed. All confirmed TB patients from the state of Texas reported between 01/2010 through 12/2016 who were ≥15 years of age, received TB treatment and had an outcome of either completed or died were included in the analyses. Univariate and multiple logistic regression models were used to determine prognostic factors associated with patient mortality. All analyses were performed with Stata MP14.2 (StataCorp LP, College Station, TX). A p value <0.05 was considered statistically significant.
Among the 6740 patients included in the analysis, 508 (7.5%) died during TB treatment. Age ≥65 years, chronic kidney disease, meningeal TB, positive culture, AFB smear and nucleic acid amplification tests, and culture conversion or unknown conversion status had the strongest association with mortality with an adjusted odds ratio (OR) and 95% confidence interval (95% CI) of 8.01 (5.78, 11.11), 4.07 (2.44, 6.78), 8.65 (4.74, 15.74), 9.60 (6.53, 14.12), and 11.37 (8.76, 14.77), respectively. US birth, homeless, residents of long-term care facilities, pulmonary TB, military TB, abnormal chest radiograph, and having positive or unknown HIV results were also associated with higher mortality. The final model had an excellent performance with a C statistic of 0.88.
The prognostic factors for mortality determined by multivariate modeling provide a practical tool for clinicians in appropriately allocating treatment, follow-up and medical support among patients treated for TB. Our results lay the foundation for the development and validation of a prognostic score system for TB mortality in the future.
Figure 1. Area under the receiver operating characteristic (ROC) curve
D. T. Nguyen,