Correcting the sources of error

If the experiment were to be repeated at some later date, a number o f changes would have to be made to the study design to avoid the errors o f this experience.

First, recruitment needs to be maximised. Almost a 1000 letters to parents produced only 34 responses and then only 16 children met the inclusion criteria. Recruiting from Health Authority child health records was more effective than recruiting from individual GP practices. The best approach would be to recruit children from more than one Health Authority. Clerical help would be required to manage this procedure and this would have budgetary implications. A related issue is adhering to the age inclusion criterion. The children involved the present study were in the correct age range at the time o f recruitment but because the investigations were carried out at times convenient for the parents, (rather than the researcher), all but two children were above the upper end o f the age inclusion criteria by the time o f the intervention. An anticipatory approach would be appropriate here. Children could be recruited aged 11-14 months, there would then be more chance o f completing the intervention before they reached the upper age limit. Age is relevant as larger children require larger doses o f isotope to produce a measurable isotopic enrichment.

Second, the children could be given the total isotope doses divided across a number o f days rather than the whole dose with one meal. Both Engelmann et al (1998) and

Fairweather-Tait et al (1995a) used a double isotope technique to measure erythrocyte incorporation in infants (as in the study reported here). In both cases each isotope dose was divided across a number o f test meals, 3 meals in the 1995 study and 2 meals in the 1998 study ie a total o f 6 and 4 meals respectively. Administering the isotope as a divided dose has the twin advantages o f averaging day to day inter­ individual differences in iron absorption and reducing the amount o f dose given with each meal. The disadvantage is that the intervention period is increased to six days, no doubt this would dissuade some parents from volunteering to take part in the study.

Third, the children should be offered a small amount o f cow ’s milk with the test meals. In retrospect it is clear that the widely varying amounts o f milk consumed by children could have affected the interpretation o f the results (had they been obtained). M ilk intake ranged from 50-240ml. The study design would be improved by offering each child a small, but consistent amount o f milk, perhaps about 50-60ml as it seems most children could consume that amount with no problems. However, there is a dose response effect o f calcium on non-haem iron bioavaiiability to be taken into account. Doses o f calcium ranging from 75-300mg when added to bread rolls have been shown to decrease iron bioavaiiability compared to control rolls in adult subjects (Hallberg 1991). It was also observed that a dose o f 40mg o f calcium had no effect. These experiments involved adding calcium chloride rather than increasing content with calcium rich foods and it is not clear what the threshold dose, if any, is in childhood. However, it would seem prudent to offer an amount o f milk containing about twice the minimum amount shown to influence iron absorption in adults. This would be about 125ml o f cow ’s milk.

Fourth, it would be sensible to collect venous rather than capillary blood samples. This would have the advantages o f providing larger and probably more representative blood samples and there would be sufficient for reanalysis if necessary. Note that these advantages are gained at a cost. It would adversely affect recruitment rates as it is a more involved procedure than capillary sampling and parents may be less willing to volunteer for the study. Any factor that adversely affects recruitment rates may prolong the data collection period and this has financial implications. Samples must also be mixed properly before the sub-samples are taken for analysis.

Fifth, other, more sensitive measurement techniques could be considered. Crews et al (1996) reported that an inductively coupled plasma mass spectrometer (ICP-MS) is able to measure ^^Fe/^^Fe and ^^Fe/^^Fe ratios in prepared blood samples more precisely than thermal ionisation mass spectrometry (TIMS). The isotope ratios have lower % RSD measurements (ie more precise measurements) when measured by ICP, less than 0.6% (Crews et al 1996). A more recent study reported measuring iron isotope ratios to an ever greater level o f accuracy and precision using multi-collector ICP-MS (0.01% RSD) (Beard et al 1999). Analysing prepared blood samples on an ICP-MS rather than a TIMS machine may improve the outcome in any future study. (Note that analysis by TIMS was, at the time, the only option at the centre where the study samples were analysed). Theoretically if the technique is more sensitive, the amount o f isotope dose can be reduced. This is a useful financial saving as the isotopes are very costly. It is interesting to note that in one o f the few studies o f iron bioavailability from foods (rather than baby milks) in young children, Engelmann et al (1998) used a different thermal ionisation mass spectrometry technique. It involved the formation o f negative ions instead o f positive ions and reportedly this method has good precision, for example the % RSD for the ratio o f ^^Fe/^^Fe is 0.02% (Walczyk 1997). Again, better sensitivity means that the isotope dose can be reduced and in fact Engelmann and colleagues were able to use small doses o f label, both the total ^^Fe and ^^Fe doses were less than Img.

Sixth, parents would not be asked to keep diet diaries for their children. If the dose administration period is extended from two to six days then the additional burden o f keeping a weighed record o f food intake is probably more than most volunteers are willing to commit to. It would be interesting to examine the relationships between iron bioavailability and nutrient intake but it is secondary issue and resources should be focussed on obtaining an answer to the main research question. If the requirement to keep the diaries was retained, then the number o f days sampled would have to be increased. The diet diaries in the study were kept for four days. As discussed in the previous chapter, this is too short a period o f time to accurately characterise usual intake o f some nutrients.

12.4 Comment

One can only speculate as to what the outcome o f this study might have been. To the author’s knowledge, there are only two studies that have investigated the effect o f calcium or cow ’s milk on non-haem iron bioavailability in children outside o f

infancy. If Study 2 had concluded that there were no differences in iron

bioavailability, this would fit with the observations o f Ames et al (1999). These authors investigated whether increasing habitual calcium intake in children aged 3-5 years would increase calcium absorption without compromising iron absorption. Eleven children were adapted to low or high calcium diets for five weeks in a cross

over study design. The average calcium intake on the low calcium diet was

502mg/day and 1180mg/day on the high calcium diet. Note that the low calcium diet is rather higher than the UK RNI for calcium intake. The RNI for 1-3 year olds is 350mg/day and 450mg/day for 4-5 year olds whereas in the USA where this study was conducted, the AI or adequate intake is 500mg/day and 800mg/day for the same age groups. At the end o f each five week period, the children were given three meals which each included a bread roll labelled with ^^Fe. Blood samples were taken two weeks later for isotope analysis. Mean percentage bioavailability was 6.9% from the low calcium diet and 7.9% from the high calcium diet, a non-significant difference. Although higher intakes o f calcium proved not to influence non-haem iron bioavailability the study only had sufficient power to detect a 40% decrease in iron absorption from the higher intake (at 80% power and 5% significance). It is quite feasible that smaller percentage decreases would be nutritionally relevant.

The hypothesis o f Study 2 was that cow ’s milk would inhibit non-haem iron absorption from a meat meal. If Study 2 had concluded this was the case, there are no studies in children with which to directly compare the finding. There is evidence that cow ’s milk, compared to fhiit juice, does reduce the bioavailability o f non-haem iron in children, but from supplemental rather than food iron (Abrams et al 1996). Ten children aged 12-14 months were recruited and using a dual isotope technique given a 5mg dose o f one isotope with 120ml cow ’s milk plus non iron fortified breakfast cereal and two hours later a 5mg dose o f the other isotope with 120ml fhiit juice and additional vitamin C. The bioavailability o f the supplement was significantly greater with juice than with cow ’s milk, 13.7% compared to 6.4%, p<0.01. As discussed in Chapter 4, there is insufficient evidence to draw firm conclusions on the effect o f

calcium on non-haem iron bioavailability in adults. As there is little on the subject in young children over one year, had results been obtained in Study 2, they would have made a useful contribution to the discussion about calcium’s effects in children at an age where calcium and dairy foods are an important part o f the diet.

If the hypothesis outlined above had proved to be true, that consumption o f cow’s milk with meat containing meals may compromise the bioavailability o f non-haem iron, this would suggest that a separation o f meals and milk drinking may be prudent. Such a finding would have to be replicated in other studies. It should be noted that any resulting public health message would have to be carefully phrased and disseminated so that milk intake and consequently calcium intake is not adversely affected. Calcium intake during childhood is particularly important for achieving peak bone mass (PBM). Although PBM is reached by about the age o f 30 years, 90- 95% is contributed during growth ie childhood (DH 1991). Bone loss starts within a few years o f reaching PBM. Age related decreases in bone mineral density (BMD) are associated with osteoporotic fractures (Cummings & Melton 2002) and although there are many other risk factors, it would be logical to assume that failing to reach an adequate PBM might put an individual at risk o f fracture in later life. This example highlights an important issue, namely that nutrients are not taken independently o f one another. Manipulating intake o f one might affect intake o f another and this could have long term adverse consequences.