Figure 1.
Sampling and sample analysis flow chat.
Table 1.
Descriptive results comparing DZM, DFM and culture on LJ for clinical specimens obtained from 344 patients at Mubende referral hospital, Uganda stratified by patient’s age, sex and HIV status as well as type of specimen and bacilli load in specimen.
Table 2.
Diagnostic performance of DZM&DFM against culture on LJ (gold standard) in detection of bacilli in clinical samples obtained from patients at Mubende referral hospital, Uganda.
Table 3.
Shows factors associated with DZM false negative results.
Table 4.
Drug susceptibility profiles for DZM false negative cases at Mubende regional referral hospital, Uganda.
Figure 2.
Kaplan-Meier plots showing the Time to detection (TTD) of bacilli on MGIT960 (Y-axis = s(t) which is the survival function, X-axis is the TTD in days).
Top row first graph shows the time to detection of all the 344 patients without any covariate structure considered. Second, third and fourth show the TTD when HIV status, Smoking and alcohol consumption are considered as covariates respectively. Bottom row: The first, second, third and fourth graph show the TTD for the combination of (patients who consume alcohol and smoke), (patients who smoke, consume alcohol and are HIV positive), (HIV negative patients who neither consume alcohol nor smoke) and bacillary load levels respectively.
Table 5.
Cox proportional hazard model of the time to detection of M. tuberculosis bacilli on MGIT960 based on 344 patients at the Mubende referral hospital.
Figure 3.
The diagnostic algorithm chart showing the case proportions formed by clinical and diagnostic combinations with data from Mubende regional referral hospital, AUC is the area under the curve, PP = predicted probability of being TB case, TS = total number of true TB cases in each covariate pattern based on the gold standard, N = the total number of individual in a specific covariate pattern (sub-population formed by combining clinical factors), TCC = total number of TB-cases accurately classified my the model.
The adjusted case predictions are made at a probability cut off P = 0.199 as established in figure 3, the broken lines shows pathways that are most likely to increase numbers of false positive cases.
Figure 4.
The receiver operating characteristics curve (ROC) showing the lowest cut-off probability with the highest specificity and sensitivity for the diagnostic/screening algorithm, this probability is used to adjust the case classification in figure 2.
The Solid line represents DZM on its own at AUC = ∼66%, the dotted line shows the model with DZM and the three predictors at AUC = ∼ 88%.
Table 6.
The logistic regression model shows the diagnostic utility of using DZM within the screening algorithm in (Figure 2).