Fig. 2

Application of a diagnostic strategy to rule out iron deficiency. [Hb]: haemoglobin concentration, expressed in g/L; [whole blood ZPP]: whole blood ZPP content, expressed in μmol/mol haem. The diagnostic strategy in a screen-and-treat survey is based on two criteria: a) the probability of correctly diagnosing iron deficiency should exceed 90%; and b) iron deficiency can be ruled out if the probability of a negative test result being correct (negative predictive value) exceeds 90%. To meet the first requirement, a cut-off point for each diagnostic test result (top) is selected to yield a sensitivity of 90%; the corresponding specificity value is obtained from the ROC curves (Fig. 1). For haemoglobin concentration and whole blood ZPP, the cut-off points were 122 g/L and 99 μmol/mol haem; the corresponding specificity values were 14.8% and 36.0%, respectively. When these markers were combined in a single diagnostic rule, 0.038689 [Hb] + 0.00694 [whole blood ZPP] > 5.93120 had a specificity of 53.7%. The negative predictive value (top panels, blue lines) depends on sensitivity and specificity values thus fixed, and the prevalence of iron deficiency. The second diagnostic criterion, i.e. the negative predictive value should exceed 90%, applies only within a limited prevalence range (top panels, red rectangle); at prevalence values exceeding this range, the negative predictive value will be below 90% and iron deficiency cannot be ruled out with diagnostic test applied (top). The percentage of children with a negative test result declines linearly with the prevalence of iron deficiency (middle panels, blue lines). The percentage of children for whom iron deficiency can be ruled out (middle panels, Y-intercepts of red rectangles) depends on the prevalence range in which the negative predictive value exceeds 90% (top panels, red rectangles). With fixed sensitivity and specificity values, the positive predictive value (bottom panels, blue lines) increase monotonically with the prevalence of iron deficiency. Within the prevalence range in which the negative predictive value exceeds 90% (top panels, red rectangles), the highest positive predictive value is 54% (for combined use of haemoglobin concentration and whole blood ZPP), indicating that additional tests (i.e. other than haemoglobin concentration and whole blood ZPP) are required to accurately determine iron status. For example, haemoglobin concentration > 122 g/L (left panels) has 90% sensitivity of detecting iron deficiency; at a true prevalence of iron deficiency <14.1%, a negative test result obtained with this decision rules out iron deficiency with a probability of 90% (upper left panel, red rectangle). Depending on the true prevalence of iron deficiency, such a cut-off for haemoglobin concentration would result in iron deficiency being ruled out in 14.1%–14.8% of children who are tested (middle left panel, red rectangle). Similarly, within a prevalence range < 28.6%, whole blood ZPP > 99 μmol/mol haem rules out iron deficiency in 28.6%–36% of children tested. Within a prevalence range of <37.4%, 0.038689 [Hb] + 0.00694 [whole blood ZPP] > 5.93120 rules out iron deficiency in 37.4%–53.7% of children