The study by Stoltzfus et al. (2003) uses logistic regression to calculate, for each study, an estimate of the risk ratio associated with a one-unit difference in hemoglobin. An overall risk ratio for the six studies was calculated in a random effects model (after testing for heterogeneity), weighting the individual estimates by the inverse of their within-study variances. The combined odds ratio for maternal mortality associated with a 1 g/dL increase in hemoglobin during pregnancy was 0.75 (95% CI: 0.62 - 0.89). In one study in China that was able to adjust for a wide variety of potential confounders in the relationship between maternal hemoglobin and perinatal mortality, the adjusted estimate was attenuated by 20%. Although there is no evidence that such attenuation applies in other contexts or to the relationship between maternal hemoglobin and maternal mortality, confounding is a universal threat and is likely to affect both relationships. In keeping with the recommendation of Stoltzfus et al. (2003) and to be conservative, PROFILES uses the attenuated estimate (OR=0.80; CI: 0.70 - 0.91) to calculate the contribution of anemia to maternal mortality, given the prevalence of maternal anemia.

Estimating the contibution of anemia to maternal mortality. For this calculation a way of estimating the mean hemoglobin concentration for a given prevalence of anemia is needed. PROFILES uses the algorithm suggested by Stoltzfus et al. (2003):

a) Assume a normal distribution of hemoglobin values with a standard deviation that depends on the prevalence of anemia (hemoglobin < 11 g/dL) (such that if prevalence <15%, SD=1.0; if prevalence =15-30%, SD=1.2; if prevalence >30%, SD=1.5);

b) Calculate the mean hemoglobin as a function of the prevalence (p), the cutoff level (c), and the standard deviation (sd) using the standard normal distribution function such that

where Ni(p,sd) is the inverse of the normal cumulative distribution for a probability equal to the prevalence p, a mean of zero and the specified standard deviation, sd.

PROFILES calculates the maternal deaths currently attributable to iron deficiency by comparing the current mean hemoglobin level (Hbc) with the estimated mean hemoglobin in the absence of iron deficiency (Hb0), assuming that some proportion of current anemia is attributable to iron deficiency. This proportion is taken as 50% in the default case (Stoltzfus et al. 2003) but can be modified if evidence suggests that it should be different. As this is a population average and as pregnant women are more likely to be iron deficient, this assumption should be considered conservative.

The calculation is done in 2 steps. First, the proportion of maternal deaths due to anemia -- the population attributable risk (PAR_mda) is calculated as:

Where RR_mda is the relative risk of maternal mortality associated with a 1g/dL increase in hemoglobin, estimated as the odds ratio cited earlier (0.80 per 1g/dL increase in mean hemoglobin), Hb₀ is the estimated mean hemoglobin level at the assumed prevalence of anemia in the absence of iron deficiency, given the assumed proportion of anemia attributable to iron deficiency, and Hbc is the mean hemoglobin level at the current prevalence.

Maternal deaths currently attributable to iron deficiency anemia (MD_ida) are then calculated as:

Where MD are the maternal deaths under current conditions of maternal mortality and anemia prevalence.

The FANTA project funded the modeling of the iron deficiency anemia model.

See Also: the Underweight calculator