Figure 4.
Major axes of covariance in clinic-immunological variables and their relationship to blood Xpert Ultra results
Principal components analysis of 32 clinic-immunological variables using varimax rotation done on 447 patients with confirmed tuberculosis. (A) Loadings of the 32 variables on first two principal components, which together capture 48% of total variance. (B) Individual patients' principal components 1 and principal components 2 scores, by day 84 outcome. Density histograms show distributions of patients' principal components 1 and principal components 2 scores by day 84 outcome. ORIQR indicates the OR for mortality associated with a one IQR increase in principal components score (the IQR effect size) with associated p value derived by logistic regression. Scatter-plot shows mortality outcome mapped onto the two-dimensional space defined by principal components 1 and principal components 2 scores. (C) Distribution of principal components 1 and 2 scores by blood Xpert Ultra Ct value, and in patients with a negative blood Xpert Ultra result. A locally estimated scatterplot smoothing fit regressing principal components score on Ct value is shown in the strata of patients with a positive blood Xpert Ultra, with 95% CI indicated by red shaded area, as well as a correlation coefficient and associated p value for a linear regression. Distribution of principal components score in blood Xpert Ultra negative patients is shown, with β coefficient from regressing principal components score on blood Xpert Ultra result (indicating the average difference in principal components score between blood Xpert Ultra positive and negative patients). Ct=cycle threshold.