Evaluation of hemoglobin performance in the assessment of iron stores in feto-maternal pairs in a high-risk population: receiver operating characteristic curve analysis

Abstract

ObjectiveBy applying receiver operating characteristic curve analysis, the objective of this study was to see whether hemoglobin levels reflect body iron stores in a group of pregnant women at term who, by using serum ferritin as the reference test, had a high pre-test probability of having iron deficiency anemia. Likewise, we evaluated the ability of hemoglobin and maternal serum ferritin levels to predict iron deficiency anemia in newborns.MethodsHemoglobin and serum ferritin were measured in 187 pregnant women at term belonging to a group with a high pre-test probability of iron deficiency anemia and their newborns. Women with Hb <11.0 g/dL and newborns with cord Hb <13.0 g/dL were classified as anemic. A serum ferritin <12.0 μg/L in women and a cord blood serum ferritin <35.0 μg/L were considered to reflect empty iron stores. Receiver operating characteristic curve analysis was applied to select the cut-off points that better reflected iron stores.ResultsThe Hb cut-off point selected by receiver operating characteristic curve analysis in women was <11.5 g/dL (sensitivity: 60.82, specificity: 53.33%, Youden Index: 0.450). Most of the newborns had normal Hb which precluded this analysis. Maternal Hb <11.0 g/dL was the cut-off point that best reflected iron deficiency anemia in newborns (sensitivity: 55.88%, specificity: 57.24%, Youden Index: 0.217). The best cut-off point of maternal serum ferritin to reflect empty iron stores in newborns was <6.0 μg/L (sensitivity: 76.47%, specificity: 31.58%, Youden Index: 0.200).ConclusionHemoglobin concentration performed poorly to detect iron deficiency anemia in women at term with high risk for iron deficiency and their newborns.

Could the erythrocyte indices or serum ferritin predict the therapeutic response to a trial with oral iron during pregnancy? Results from the Accuracy study for Maternal Anaemia diagnosis (AMA)

“…Previous studies, which validated the erythrocyte indices against SF at 2 nd and 3 rd gestational trimesters, report similar low diagnostic performance with very low Se [ 14 , 21 , 22 , 34 ]. Accordingly, in the National Health and Nutrition Examination Survey of United States (NHANES) 1999-2006, the Hb detected only 16 % (Se) of the pregnant women classified as iron-deficient by the ‘total body iron model’: {– [log 10 (sTfR × 1000 ÷ ferritin) – 2.8229] ÷ 0.1207} < 0 mg/Kg [ 35 ].…”

“…Accordingly, in the National Health and Nutrition Examination Survey of United States (NHANES) 1999-2006, the Hb detected only 16 % (Se) of the pregnant women classified as iron-deficient by the ‘total body iron model’: {– [log 10 (sTfR × 1000 ÷ ferritin) – 2.8229] ÷ 0.1207} < 0 mg/Kg [ 35 ]. ROC curves in a moderately (0.7 < AUC ≤ 0.9) or less accurate ranges [ 33 ] for both indices of erythrocyte mass and morphology were observed by Tam et al [ 10 ] ( versus SF < 20 ng/mL), Casanova et al [ 9 ] ( versus SF < 10 ng/mL) and Jaime-Pérez et al [ 14 ] (versus SF < 12 ng/mL). On the other hand, Tiwari et al [ 13 ] found a highly accurate (0.9 < AUC < 1.0) curve for Hb, with Se of 91 % and Sp of 79 % for a cutoff of 9.7 g/dL ( versus SF < 12 ng/mL), differing from ours and other authors results.…”

“…Clinical studies have showed a low accuracy of these tests to diagnose maternal iron-deficiency [ 9 – 11 ]. Hb and HTC [ 9 , 12 – 14 ], as well as the other erythrocyte indices [ 9 , 13 ], do not correlate with the iron status in pregnant women, probably due to the physiological haemodilution [ 15 , 16 ] and ‘emergency hemopoiesis’ [ 17 , 18 ], which not only modify the cell indices in the erythrogram, but also the serum ferritin and other biomarkers of the body iron [ 6 , 7 ]. The cutoff points of erythrocyte indices and serum ferritin were based on the distribution curves of populations from developed countries in the 70’s and 80’s [ 19 , 20 ]; therefore, they were not defined by diagnostic validation studies.…”

“…The cutoff points of erythrocyte indices and serum ferritin were based on the distribution curves of populations from developed countries in the 70’s and 80’s [ 19 , 20 ]; therefore, they were not defined by diagnostic validation studies. Pregnancy validation studies, which compared erythrocyte indices with the serum ferritin [ 9 , 10 , 14 , 21 , 22 ] or the iron content in the bone marrow [ 11 , 12 ] describe very low sensitivity (Se) and/or specificity (Sp) for the erythrocyte mass indices (Hb and HTC) [ 10 , 14 , 21 , 22 ] and morphology indices (mean corpuscular volume - MCV, mean corpuscular haemoglobin - MCH, mean corpuscular haemoglobin concentration - MCHC, red blood cell distribution width - RDW) [ 9 , 11 , 12 , 22 ]. On the other hand, only two studies compared the ferritin levels with the iron in pregnant women’s bone marrow [ 11 , 12 ], and there is no consensus about its cutoff during pregnancy [ 7 , 23 – 25 ].…”

Sparse learning of maximum likelihood model for optimization of complex loss function

“…This comparison can be conducted by using some multivariate performance measures. For example, prediction accuracy, F-score, area under receiver operating characteristic curve (AUROC) [7,24,25,2], and precision-recall curve break-even point (RPBEP) [37,19,29,22]. A problem of such machine learning procedure is that in the training process, we optimize a simple loss function, such as hinge loss, but in the test process, we use a different and complex performance measure to evaluate the prediction results.…”