Skip to main content
Figure 6 | BMC Medicine

Figure 6

From: Proxy markers of serum retinol concentration, used alone and in combination, to assess population vitamin A status in Kenyan children: a cross-sectional study

Figure 6

ROC curve of the best model to predict vitamin A deficiency with its decision rules during different prevalence rates of vitamin A deficiency. True prevalence was arbitrarily selected as 6%, 15%, 30% and 40% as the mid-points for the ranges that indicate mild, moderate and severe public health problems (2-10% and 10-20%, >20% respectively). Note that, with the true prevalence decreasing, the optimal cut-point for the linear predictor results in increased specificity even at the expense of sensitivity. For example, at a true prevalence of vitamin A deficiency of 15%, individuals for whom (−15.277 × [RBPμmol/L] - 7.013 × [Transthyretinμmol/L] + 0.367 × [C-reactive proteinmg/L] + 24.714) > 0.496 should be classified as deficient, whereas all others can be classified as replete (in this formula, concentrations are indicated in straight brackets, and expressed in units as indicated).

Back to article page