Iron absorption: Comparison of prediction equations and reality. Results from a feeding trial in the Philippines

Background: A number of algorithms have been developed that seek to predict the bioavailability of iron from mixed meals and diets, but their direct validity in predicting change in iron status remains questionable. Throughout the course of conducting a large feeding trial in convents in Manila, we collected weighed food-intake data and have the opportunity to directly compare the performance of these prediction equations. Aims: The specific aims of this particular analysis are as follows: (a) to determine habitual intakes of macro- and micronutrients in religious sisters in Manila, (b) to determine the predicted efficiency of iron absorption from each of the published bioavailability algorithms (Monsen, Hallberg, Reddy, Tseng, Barghava, and Du), and (c) to determine which of these equations best predicts the actual "gain" in iron in religious sisters over the duration of the trial. Design: The efficacy of consuming high-iron rice was tested during a nine-month feeding trial with a double-blinded dietary intervention; these results have been published [7]. Religious sisters living in 10 convents around metropolitan Manila, the Philippines were randomly assigned to consume either high-iron rice (3.21 mg/kg Fe) or a local variety of control rice (0.57 mg/kg Fe) within each convent. Results: Religious sisters in convents consumed a diet that appeared to be typical of habitual intakes of macro-and micronutrients in this part of the Philippines. The analysis of the six equations revealed highly significant differences in predicted efficiency of iron absorption. The Hallberg, Monsen, and Reddy equations all predicted similar median efficiency (6.88, 7.92, and 6.42%). In contrast, Bhargava (4.68%), Tseng (3.23%), and Du (2.92%) were significantly lower. The correlation (r = 0.98) of Monsen to Hallberg was highly significant and the slope was not different than unity. The median efficiency of absorption based on the gain in body iron in 114 subjects over nine months, combined with an estimate of daily iron requirements, was 17.2%. Thus, none of these equations approximated the computed iron absorption based on improvement in serum ferritin and suggests alternative approaches to predicting iron accumulation from diet need to be formulated. Conclusions: Inhibitory factors in the prediction equations either had little effect or had too large of an effect on apparent bioavailability as compared to median absorption over a nine-month period. The causes of the lack of agreement between computed iron gain and predicted absorption are open to discussion and will need to be resolved.