Main Study: Methods and Results
7.1 Research questions
7.2.2 Sample size
M any researchers have reported an inverse association between cow ’s milk
consumption and iron status (see Appendix 3). As discussed in section 5.2.1.1,
increased consumption o f cow ’s milk potentially displaces foods rich in iron and other foods that might promote its absorption. The net effect is reduced iron intake (and possibly reduced bioavailability) and so reduced availability o f iron for erythropoiesis. It is reasonable to assume that given this evidence, children in this study would be more likely to be anaemic if they were higher milk consumers. The prevalence o f anaemia in the recruitment area was known to be about 20% (Lawson et al 1997). It might be expected that a similar relationship would be noted in this study. It was decided that detecting about a 10% difference between the proportions o f anaemic children consuming less than the median cow ’s milk intake and those consuming above the median, would be nutritionally relevant. For the purposes o f calculating a sample size, the statistical hypothesis is that three quarters o f the cases
o f anaemia will occur among those who drink over the median for milk intake and a quarter o f the cases in those who drink less. The sample size was estimated using the following equation (Kirkwood 1988).
n = F \p % , (100 - f % , ) + /7%2 (100 - p% ^ )
7 .8 5 [5 (l0 0 -5 ) + 15 (l0 0 -1 5 )]
10
'= 107.5
= 108
Each group should contain 108 subjects therefore a total sample size o f 216.
Where F= 7.85
p"^]=5% (ie Va, o f the cases o f anaemia estimated to be in the sample) p^2=15% (ie Va o f the cases o f anaemia)
d= 10
For practical reasons, (largely due to the finite budget for processing blood samples), the sample size was reduced slightly to 100 in each group, a total o f 200 children. This number o f subjects will detect an 11.6% difference in the proportions o f anaemic children between the two groups at the 5% significance level with a power o f 80%.
7.2.3 Subjects and recruitment
In the East London and City Health Authority area, invitations were sent out via GP practices to children o f suitable age registered with the practices. Families were sent reminder letters three weeks after the original letter. In each case the invitation pack contained a letter, reply form, reply paid envelope and a parent information sheet (see Appendix 5 for copies). Attempts to recruit volunteers from three drop-in child health clinics in the Camden & Islington Community Trust area were abandoned as recruitment rates were negligible.
Table 7.1 Study 1 inclusion and exclusion criteria
Inclusion criteria Exclusion criteria
Aged 1.5-2.0 years old <1.5 years or >2.0 years
Apparently healthy Undergoing clinical investigations or ill
Singleton M ultiple birth
7.2.4 Study design
Parents and children who volunteered to take part in the study and met the inclusion criteria were visited at their home. The researcher went through a questionnaire asking about the child’s infant feeding practices, current milk drinking habits and other dietary and socio-economic information with the parent (see Appendix 5). In addition the parent/earer completed the food frequency questionnaire about their child’s food and drink intake. Birthweight was recorded as recalled by the parent or taken from the parent held child health record. W here possible, the child’s weight (Seca Scales Model No 834) was measured. At the end o f the interview a capillary blood sample o f approximately 0.2-0.5ml was collected from each child into a paediatric EDTA blood tube. The child’s response to the procedure was recorded. All the capillary blood samples were collected by the author who was trained in blood sampling techniques prior to the study. There were a few occasions when the blood sample was unusable. This was usually due to the sample clotting or insufficient quantity for analysis. In these instances parents were asked if they would give permission for another sample to be collected. No more than two attempts to collect blood were made. The blood samples were immediately taken to the Great Ormond Street Hospital Haematology Laboratory where they were analysed for full blood count (Bayer H3) and screened for haemoglobinopathies (Biorad Variant). Parents and their children’s GPs were informed o f the blood test results. Any child who had a haemoglobin concentration o f <10.5 g/dl was referred to their GP.
7.2.5 Statistical analysis
Odds ratios were calculated to make an initial assessment o f the relationship between the various dietary, anthropometric and lifestyle variables obtained from both the main data collection questionnaire and the FFQ w ith haemoglobin concentration <11.0g/dl and <10.5g/dl. An odds ratio is the ratio o f the number o f times that an outcome o f interest occurs to the number o f times that it does not (Bland & Altman
2000). Variables from the FFQ, which were originally in non-linear categories ie “ 1- 3 times a month”, “5-6 times a week” and so on, were reclassified in a linear fashion ie all categories were expressed as frequency per week. These variables and other continuous data variables were reclassified as binary for these odds ratio calculations. In addition, intake frequency scores were calculated for certain food groups and their relationship with anaemia assessed. For example, a total meat consumption frequency score was calculated. Each category o f the FFQ was assigned a score, the lowest category o f intake scoring 0 and the highest, 8. If a subject’s consumption o f beef was in the lowest intake category, then the score for that food item was 0 and so on. The final score for each subject was the sum o f scores from all the meat items.
Variables found to be significantly related with the two anaemia cut-off points by univariate analysis were entered into separate logistic regression models to identify factors which were independently predictive o f anaemia. It is arguable that certain variables such as gender, social class, ethnicity and volume o f cow ’s milk drunk each day, (see Chapter 5), might be related to risk o f anaemia. Therefore, although none o f these variables proved to be significantly related to anaemia in the univariate analysis, they were also entered into the models. Univariate analysis provides a basis for deciding which o f the many variables generated from a data set would be useful for building a model to predict presence o f anaemia. However, it is also true that as Altman (1991) comments, “ ...variables may contribute to a multiple (logistic) regression model in unforeseen ways due to complex interrelationships among the variables.” Therefore it is reasonable to include in the model variables that on the
basis o f existing evidence m ay influence iron status. On a practical note, any
statistical model that does not include assessment o f social class, a factor often reported to be associated with iron status, is open to criticism. Volume o f cow ’s milk consumption was not associated with anaemia in the univariate analysis in any way, but as there is ample evidence o f an association o f cow ’s milk intake and poorer iron status in other studies, (see Appendix 3, Table 1), it was included in the model. A cut-off point o f 600ml was selected as the current government recommendations are that children consume this amount each day. It is logical to assume that milk intake greater than 600ml/day might prove to be important in the regression model.
All variables were entered simultaneously into the logistic regression models. At each stage the variable that was least significant was removed and the model rerun until the remaining variables were significant at the 5% level.
The ability o f these factors to identify children with and without anaemia was assessed by calculating their sensitivity, specificity, positive predictive value (PV+)
and negative predictive value (PV-). The predictive value o f the variables was
assessed both individually and in combination. The variables were scored according to the magnitude o f the regression coefficients to produce a combined, scored test. The ability o f each possible test score to predict which children had anaemia was assessed. Children in the sample were also categorised as having a dietary deficiency as defined by Boutry & Needlman (1996). They defined dietary deficiency as either drinking more than 16oz o f cow ’s milk per day or eating meat, cereals, bread, vegetables and fruit less than five times a week or taking fned food, crisps, sweets or soft drinks once a day or more. The predictive ability o f this summary measure to predict poorer iron status in the sample was also assessed.
A few other analyses were performed, differences in mean haemoglobin concentration between children who were within, above or below the age inclusion criteria were
compared using one way analysis o f variance (ANOVA). An appropriate
transformation was carried so that the haemoglobin concentration data approximated normal distribution. Proportions in two groups were compared using a chi-squared test. The analyses were carried out using Statistical Package for Social Sciences, version 10.0.
7.3 Results