Of 163 patients, 147 (90%) were enrolled in Uganda and 16 (10%) were enrolled in South Africa. Compared to patients enrolled in the original clinical trial who were not included, patients in this substudy were significantly younger, more often male and of black race, weighed less, and were more likely to be HIV infected (Table ). The median baseline CD4+ cell count of HIV-infected subjects in the substudy was 200 μl−1 (range, 110 to 286 μl−1).
Culture end points and treatment failure.
Of 163 patients with TTD culture results, 146 patients, including 5 of the 6 patients with failure, had data for four cultures in the 2 months of study therapy. In these 146 patients, there was more rapid detection of growth (TTD of four cultures) in the five patients who failed treatment than in the others who did not fail treatment (mean ± standard deviation, 16.3 ± 2.9 days [range, 13.0 to 20.3 days] versus 25.4 ± 6.7 days [range, 6.0 to 42.0 days]; P = 0.003 by t test) (Tables and and Fig. ). When TTD was assessed as a diagnostic end point for treatment failure, using a mean cutoff of less than 21 days (for specimens for culture obtained 2, 4, 6, and 8 weeks after the beginning of study treatment), the accuracy was 75%, the sensitivity was 100%, the specificity was 74%, the positive predictive value (PPV) was 12%, and the negative predictive value (NPV) was 100% (Table , end point label 1). Improved PPVs (18 to 20%) and similar sensitivities and specificities were obtained when fewer time points were used (two time points at week 6 and 8 cultures or one time point at week 8 culture) (Table , end point labels 2 and 3). By comparison, the alternative end point of culture conversion to negative after 2 months of treatment was less accurate and specific (accuracy, 47%; sensitivity, 100%; specificity, 49%; positive predictive value, 6%; negative predictive value, 100; Table , end point label 4).
Clinical, microbiologic, radiographic, and molecular data from six patients with treatment failure (cases 1 to 6) and another with early relapse (case 7)
Range of sensitivity, specificity, positive and negative predictive values, and accuracy of mean TTD with different numbers and weeks of time point data and cutoff values in evaluable patients
FIG. 2. Patients with (, n = 5) and without (, n = 133) treatment failure (all with four time points, relapse case excluded). Unadjusted means (bars, 95% confidence intervals) of time to detection are depicted on the (more ...)
Among 38 HIV-infected patients at initial testing, 7 (18%) had cavitary lung disease, received ethambutol, and were on intermittent therapy; 4 of the 7 (57%) had treatment failure. The accuracy of the TTD end point (four time points with a mean of <21 days) among the 24 HIV-infected patients with cavitary disease was 71% (sensitivity, 100%; specificity, 65%; PPV, 36%; NPV, 100%).
Correlation of TTD with known risk factors for combined end point of treatment failure and relapse.
Significant effects of TTD in univariate analysis were also found in multivariate ANCOVA (Table ). TTD was associated with cavitary lung disease; i.e., the greatest decrease (delayed response) was with large cavities, an intermediate decrease was with small cavities, and the lowest decrease was without cavities (Table ; Fig. ). The adjusted mean TTD was also significantly decreased among patients treated thrice weekly and with ethambutol (versus moxifloxacin). By comparison, 2-month culture conversion rates were similar among 158 patients treated thrice weekly versus 5 days/week (46% versus 51% culture conversion; P = 0.5, chi-square test) or treated with moxifloxacin versus ethambutol (culture conversion, 54% versus 46%; P = 0.42). Although not significant, the direction of proportions with 2-month culture conversion corresponded to the significant differences observed with TTD.
Univariate and multivariate effects on TTDa
FIG. 3. Growth was detected sooner in patients with large cavities and was intermediate in those with small cavities than in patients without cavities (, with cavity of ≥4 cm; ○, with cavity of <4 cm) or no cavity (), (more ...) Other analyses of TTD.
The adjusted mean TTD differed between study sites (Uganda versus South Africa, 25.5 days versus 20.8 days; P = 0.008), reflecting in part differences among patient groups and in the culture processing methods, the broth media, and the monitoring methodologies used. However, differences in the adjusted mean TTD of other cofactors were similar between models of Ugandan patients alone and all patients (Uganda and South Africa). Further, there were no significant differences by site between drugs or treatment frequency; i.e., there were no site interactions. These findings are consistent with an increase in power from a larger sample size with all patients.
From 161 patients, 591 duplicate sputum cultures (of specimens obtained from the baseline to week 8 of treatment) with results of Mycobacterium tuberculosis or no growth were evaluable. TTDs in the duplicate cultures were significantly correlated (P < 0.0001, R2 = 0.69).
Other analyses of the TTD data were performed to explore the operating characteristics of the ANCOVA model. A frequency distribution of TTD data from 1 to 41 days demonstrated a middle peak and two tails with very low frequencies at days 1 and 2 and days 39, 40, and 41. Because about 30% of cultures did not demonstrate growth by the end of the monitored interval, we performed a sensitivity analysis of the effect of censoring at 42 days. Model term estimates and the significance of terms were insensitive to imputed censoring at 39 or at 45 days compared to 42 days.
We also performed GEE analysis to evaluate the effect of scoring all cultures as either positive or negative and compared the results to those obtained with a parallel TTD ANCOVA model. The significance of model terms and the direction of change between model terms were comparable between the different models, with one substantive exception (Table ). The adjusted mean TTD of site differed significantly in the ANCOVA model but not in the GEE model. This demonstrates the similarity of data from the Bactec system, used at the Uganda site, and the MGIT system, used in South Africa, when culture results are analyzed as binary positive or negative results. However, a difference is distinguished between the monitoring systems with TTD data, likely reflecting the greater sensitivity of MGIT medium and monitoring to detect M. tuberculosis.
Comparison between TTD data by ANCOVA and binary data (culture positive or negative) in a generalized model employing estimation equationsa