The purpose of the questionnaire was to assess maternal infant milk feeding practices, mothers' decision making regarding how much and how often to feed their babies, their attitudes to infant feeding and growth, and beliefs about following infant feeding recommendations. The 57-item (19 questions), self-administered questionnaire was developed after conducting literature reviews[18–21] and a qualitative study using a flexible semi-structured interview schedule involving 38 parents (n = 35 mothers). Three main themes emerged on parents' decision-making regarding volume and frequency of formula-milk feeds; i) baby's appetite (if the baby finished the bottle, or cried between feeds, more was added to the next feed), ii) instructions on formula milk tins/packets (if the baby did not take what it said on the tin, they were offered a feed again after a short gap) iii) baby's growth (as baby's weight increased, feeds were increased). Parents got information on bottle-feeding from various sources -family, friends, other mothers, 'parent support groups', books, magazines, the internet, formula-milk packets, supermarket shelves, health visitors and midwives. The main barriers to reducing formula-milk feeds were concerns that the baby would cry, be hungry, wake up at night and demand more frequent feeds.
An iterative process was used and numerous revisions were made in response to input from groups of researchers and mothers. Content validity was assessed by extensive pilot testing with mothers (n = 60) participating in a birth cohort study (The Cambridge Baby Growth Study). Questions covered four domains: 1) type of milk feeding, decision making, and sources of advice, 2) frequency and quantity of feeds, 3) attitudes to infant feeding and growth, and 4) theory-based beliefs about following recommendations to reduce formula-milk feed quantities.
Type of milk feeding, decision making, and sources of advice
Questions related to the type of milk feeding: breastmilk, expressed breastmilk, formula-milk feeds, and type and brand of formula-milk feeds. Regarding decision making about frequency of milk feeds, the question was phrased 'When deciding how often to feed your baby, do you usually... feed your baby on demand, or follow a routine, or do a combination of both?' Regarding decision making on quantity of milk feeds, our qualitative study showed that mothers either followed the guidelines printed on the formula-milk packaging, or based their decisions on the baby's appetite or growth. Hence our question was phrased 'When deciding how much to feed your baby, do you usually...follow guidance, or depend on baby's appetite, or depend on baby's growth?'
Our systematic review reported that mistakes in preparation of formula-milk feeds with formula-milk powder were common. Parents sometimes heaped or tightly packed the scoops or added powder to the bottle first which resulted in an over-concentrated feed. Furthermore parents reported that they did not receive sufficient advice from healthcare providers. Consequently, we included questions on how feeds were prepared, mothers' sources of advice, and which advice was followed. These questions would also be of particular value to the process evaluation of any trial to change parents' infant feeding behaviour.
Frequency and quantity of feeds
Amount of milk intake may change often during infancy and therefore to quantify the association between milk intake and infant growth, it may be necessary to assess milk intake frequently (every 4-6 weeks in the first six months). A 4 day diet diary would be burdensome for mothers of newborn infants to complete frequently, hence the following questions were developed as a pragmatic substitute. Example questions include 'In a typical 24 hour period how much formula milk does your baby have? Amount of formula milk per feed? Number of formula feeds per day? Number of scoops of formula milk powder per feed? Duration of typical daytime feed?' Similar questions were included for breastfeeds, water and other drinks. In order to assess whether milk feeds were replaced by solid/semi-solid foods, we included questions relating to these, for example 'What was your baby's age in months when you started solid/semi-solid food?'
Attitudes to infant feeding and growth
A literature review identified validated questionnaires on breastfeeding self-efficacy[18–21] and self-efficacy in infant care[23, 24]. These were used to develop questions to assess mother's confidence (self-efficacy) in infant growth monitoring and feeding so that her baby would not gain too much weight. The questions included eight items, each scored on a five-point Likert-type scale from 'strongly agree' to 'strongly disagree'. For example 'I am confident that I can feed my baby so they do not gain too much weight'. A single question on perception of size -'Do you think your baby is... underweight, OR about right, OR overweight?', was derived from a published study.
Theory-based beliefs about following recommendations to reduce formula-feed quantities
In 2004 the World Health Organisation (WHO) and other international bodies reduced their recommendations for energy requirements during infancy from previous (1985) recommendations. Mothers' beliefs about following these new recommendations would influence whether the recommendations were followed. The 11 items in this domain were chosen to measure self-efficacy (confidence in performing a behaviour and overcoming barriers to that behaviour), outcome expectancies (expectation that a positive outcome will occur as a function of that behaviour)-the hypothesised mediators of behaviour change according to Social Cognitive Theory[28, 29], and intentions as informed by Theory of Planned Behaviour. These could be scored on a five-point Likert-type scale from 'strongly agree' to 'strongly disagree'. Through our qualitative study of 38 mothers we identified the most common barriers to reducing formula-milk feed quantities (baby would cry, remain hungry, wake up frequently), and used these to create the items in this scale. We also wanted to measure the three dimensions of outcome-expectancies (physical, social and self-evaluative) which predict behaviour. The items were worded positively and negatively and presented in random order. For example 'If I follow the new feeding recommendation, my baby will wake up frequently at night' (negative physical outcome expectancy).
Thirty one mothers were recruited from an ongoing birth cohort study. To include a more diverse range of participants, we also recruited nine exclusively formula-milk feeding mothers from a focus group study conducted to inform intervention development. To assess test-retest reliability, following receipt of a completed postal questionnaire, the same questionnaire was posted and mothers were asked to complete it a second time (median time interval between completion of the two questionnaires was 6 days (range 2-16 days).
To validate the questionnaire, no 'gold-standard' existed, therefore semi-structured interviews were used to assess criterion validity as previously reported for the validation of a questionnaire covering correlates of children's physical activity. We developed an open-ended, semi-structured interview schedule in which participants could make general comments on the questionnaire, clarify the meaning of the questions, and expand on their answers, giving a richer response (see Additional file 1 for interview schedule). After the first questionnaire was returned, an appointment was made for a face-to-face or telephone interview with the mothers. At the start of the interview, it was confirmed that mothers had completed the second questionnaire and they were asked for general comments on the questionnaire. The median time interval between completion of the first questionnaire and interview was 8 days (range 3-16 days). Interviews were conducted by RL or JL, were tape-recorded and were transcribed by an independent company. The interview data were used as the criterion measure against which to compare the participant's first questionnaire responses. For each participant, two researchers (RL and JL) used the transcripts to fill in a blank questionnaire. This was done independently and blind to the participant's questionnaire responses. For the two questions (questions 16 and 19 with a total of 19 items) which were scored on a five-point Likert type scale, the transcripts were used to mark a collapsed three-point scale 'agree, neutral, disagree'. Once all of the transcripts were coded, the two researcher-completed questionnaires (for each participant) were compared, and in the event of disagreement, consensus was reached by discussion. This final 'agreed' (by both researchers) questionnaire was used as the 'criterion' for comparison against the participant's first questionnaire responses to assess criterion validity.
Approvals for the study were obtained from the local Research Ethics Committee and research governance committees of the local hospital and Primary Care Trust. Demographic details and written informed consent were obtained from all participants.
All analyses were performed using Stata version 10.1 (Stata Corp LP, College Station, Texas). The test-retest reliability (between 1st and 2nd questionnaires) and criterion validity (between 1st questionnaire and interview coded response) were assessed for each item by calculating percentage exact agreement and chance-corrected agreement (Cohen's kappa). A percentage agreement ≤ 66% was used to indicate fair agreement [32, 33] and kappa values were categorized as: poor (< 0.0), slight (0.00-0.20), fair (0.21-0.40), moderate (0.41-0.60), substantial (0.61-0.80), and almost perfect (0.81-1.0) agreements. Weighted kappas were calculated for ordinal categorical variables and Spearman's correlation coefficients for continuous variables.
To assess internal consistency we calculated Cronbach's alpha for items measuring theory-based constructs- self-efficacy, outcome-expectancies and intentions. Negatively worded items were re-coded. A summary score for each construct was calculated by summing the individual item scores and dividing by the number of items in the construct (4 for self-efficacy, 5 for outcome-expectancies and 2 for intentions). The correlations between a summary score for these constructs were calculated.