In total, 62 400 records for TB notification were analyzed for the period 2009-2013. The following findings were obtained:
- Predictors of TB incidence in O.R. Tambo district
Figure 1A and 1B show comparisons of mean values of incident TB cases from 2009 to 2013. The incidence curve of TB trends dropped abruptly and significantly from 2010 until 2013 (ρ < 0.00001).
During the univariate analysis of predictors of incident TB cases from 2009 to 2013 (Table 1A and 1B), the study found a significant (ρ < 0.05) bivariate (univariate) positive correlation between population density (r = 0.812; ρ < 0.0001), poverty gap (r = 0.210; ρ < 0.01), and mean number of TB cases. However, there was a significant (ρ < 0.05) negative correlation between supervision rate (r = - 0.173; ρ = 0.030), PHC professional nurse clinical workload (r = -0.164; ρ = 0.021), expenditure per patient day equivalent (r = -0.282; ρ = 0.015), PHC expenditure per capita (r = -0.159; ρ = 0.034), local government expenditure on PHC (r = -0.244; ρ = 0.022), and the mean number of TB cases.
After adjusting for confounding variables using multiple linear regression analysis, only expenditure per patient day equivalent (PDE) and PHC expenditure per capita were identified as the most important, significant and independent predictors (adjusted R2 = 60%; ρ = 0.002) in terms of declining incident TB cases following the equation: Y= (- 209× Expenditure per PDE) + (- 0.191 × PHC expenditure per capita).
- Association between socio-economic deprivation and TB care indicators in O.R. Tambo district
The following TB care indicators were assessed in this study: TB deaths (all TB cases); TB case finding index; TB treatment failure (all TB cases); HIV and TB deaths; TB defaulter rate (all TB cases); and New TB smear positive cases.
The associations between the above TB care indicators and multiple deprivation index from 2009 to 2013 were measured as follows:
(a) Univariate analyses
On single variable analysis (Figure 2A – 2F), TB cases in the most socio-economically deprived group were significantly more likely to have a high TB death rates (Figure 2A), high PTB case finding index (Figure 2B) and high HIV and TB co-infection death rates (Figure 2D) as compared to the least socio-economically deprived group. However, individuals in the least deprived group were significantly more likely to be associated with high rates of TB treatment failure (Figure 2C), high rates of TB defaulter rates among TB smear positive group (Figure 2E) and high rates of new PTB smear positive results (Figure 2F) as compared to the most deprived individuals.
(b) Multivariate analysis
Using canonical discriminant analyses (CDA) on normally distributed variables that showed significant univariate associations, three canonical functions were identified as able to discriminate among the socio-economically deprived groups. The obtained 3 discriminant functions had Eigen values that are summarized in Table 2. The larger the Eigen value represents more shared variance in the linear combination of variables.
Of the 3 identified canonical functions, the first two functions contributed a total of 98.5% of the total variance, which easily stratifies into the 5 study groups (from quintile 1 to quintile 5). The Wilks’ Lambda value was significant: 0.019; χ2 = 137.62; ρ < 0.0001 (Table 3).
Functions’ coefficients were calculated and used to decide which variables predicted group membership. Comparing the values between groups, the higher coefficient means the predictor variable attributes more for that group (Table 4 and Table 5).
Table 4 demonstrated Fisher’s linear discriminant functions according to classification function coefficients. After adjusting for confounding factors (PTB case finding index, TB defaulter new smear positive, TB death in all TB, TB defaulter all TB, and TB failure in all TB), only HIV and TB death and new TB smear positive cases were identified as the most important, significant, and independent indicators able to discriminate the most deprived communities with a deprivation index far from other deprivation-concentration-dispersion quintiles 2-5 (Table 5 and Figure 3).
Standardized canonical discriminant function coefficients were used for predicting functions defining group centroids (Figure 3). The canonical group means (also called group centroids) are the mean for each group’s canonical observation scores. The larger the difference between the canonical groups implies the better the predictive power of the canonical discriminant function in classifying observations.
In conclusion, CDA showed that TB cases in the most socio-economically deprived communities (quintile 1) were significantly more likely to have high HIV and TB co-infection death rates as compared to the least socio-economically deprived group (quintile 5) [Eigen value (12.95), function coefficient (1.49) > (.77); Wilk’s Lambda = .019, p < .0001].