Tuesday, June 6, 2023

Twelve-month effectiveness of telephone and SMS support to mothers with children aged 2 years in reducing children’s BMI: a randomized controlled trial – International Journal of Obesity

Study design

We conducted a 2-arm parallel RCT during March 2019 and October 2020 with a three staged nurse-led telephone and SMS support intervention that targeted mothers of children aged 2 years. The study protocol was published prior to the commencement of the study, together with the trial registration [14]. The protocol was implemented with some amendments mainly for stage 3 intervention content (e.g., covering some COVID-19 related information) and measurement of height and weight due to the COVID-19 pandemic prohibiting face-to-face data collection.


The study was built directly on the existing 3-arm CHAT trial [13] conducted in metropolitan Sydney, New South Wales (NSW), Australia, with recruitment from antenatal clinics in eight hospitals of four local health districts [11, 12, 15]. Briefly, the existing trial recruited women (n = 1155) from late pregnancy with follow up until their children were aged 2 years [11, 12, 15].

Participants and recruitment

For this current study we only recruited mothers (n = 666) who completed the 2-year assessment of the previous trial, including a telephone survey and measurement of child’s height and weight at their homes. Informed consent to this current study was obtained for 662 mothers at the time of their 2-year survey, which became baseline for this current trial. We then randomly allocated the participating mother-child dyads to the intervention group (i.e., receiving combined telephone and SMS intervention) or the control group. The original recruitment criteria of the previous trial included women aged >18 years at 28–34 weeks of pregnancy, were able to communicate in English, had a mobile phone, lived in the recruitment areas, were able to give informed consent and did not have any severe medical conditions.


We used a stratified randomization method based on participants’ group allocation within the previous trial (see Fig. 1) so that any ‘carry-over’ effect of the previous trial was balanced between the groups. A web-based randomization plan was generated using randomly permuted blocks (n = 6) (http://www.randomization.com/).

Fig. 1: CONSORT diagram.

A flowchart of the study participants [12].


We developed a 3-stage intervention guided by the Health Belief Model [16], and motivational interviewing techniques as per protocol [17]. The intervention aimed to improve mothers’ parenting behaviors and their own healthy behaviors. Telephone support consisted of protocol-based sessions based on the Australian Dietary Guidelines, early childhood developmental guidelines and the Australian 24-Hour Movement Guidelines for the Early Years (Birth to 5 years) [18]. Each stage of the intervention started with a mailed intervention booklet, followed by a telephone support session and then SMS twice a week for four weeks at three time-points (24–26, 28–30, and 32–34 months of child age). A Child and Family Health Nurse delivered the telephone support session of 30–60 min in duration by going through main intervention messages from the mailed booklets, and then text messages using a 2-way automated SMS system were sent at a predetermined time (10am–1pm) to reinforce the intervention information and key messages in the booklets. A summary of the intervention content be found in the Supplementary Document 1.


Mothers in the control group received usual care from the local health districts. We also sent out two booklets on information not related to the obesity prevention intervention such as toilet training, language development and sibling relationships as a retention strategy.


A market survey company used a computer-assisted telephone interview (CATI) to collect baseline measures at 2 years and outcome measures at 3 years. The interviewers were blinded to the research hypotheses and treatment allocation. Participating mothers were also blinded to the specific details of the research hypotheses.

Outcome measurements and data collection

Primary outcome

The primary outcome was children’s BMI at 3 years of age. We planned to directly measure weight and height by four research assistants (RAs) via home visits. However, we only managed to measure the height and weight of 30 children (14 intervention, 16 control). Home visiting data collection was stopped due to the COVID-19 pandemic lockdown restrictions in April-October 2020, when most face-to-face health services were suspended. Thus, we had to use the CATI survey to collect children’s height and weight measured by their mothers (n = 440) using the measurement kit which was sent out prior to the survey (Supplementary Document 2 about measurements of child height and weight). The measurement kit sourced from a commercial company included the height ruler and detailed instructions for parents on how to measure and record height and weight of their child. We also modified the instructions to suit our study participants. BMI and BMI z-score were calculated using the WHO AnthroPlus v1.0.4.

Secondary outcomes

The secondary outcomes at 3 years were BMI z-score, children’s dietary and activity behaviors as reported by their mothers via a telephone survey with a questionnaire. Children’s dietary behavior included vegetable, fruit, fast food and soft drink consumptions as well as feeding practices (i.e., eating in front of the TV and using food for reward). Children’s activity behavior was assessed by their outdoor playtime, screen time and sleep duration. We also collected mothers’ vegetable and fruit consumptions and physical activity and sedentary behaviors. Socio-demographic data were collected by CATI from the previous trial at baseline and then updated at the 2- and 3-year surveys. The questionnaires used for assessing outcomes were the same as those used in the previous Healthy Beginnings Trial [19, 20], and can be found in Supplementary Document 3.

Sample size

As described in the study protocol [14], we estimated a sample of 506 (253 per group) at age 3 years would allow us to detect a difference of 0.40 kg/m2 in mean BMI (SD = 1.60) at the 2-sided 5% significance level with 80% power. This effect size was based on the findings from a 6-month home-based intervention study in the US that detected a decrease in BMI of 0.40 kg/m2 with children aged 2–5 years [21].

Statistical analysis

All statistical analyses were carried out as per study protocol [14] and pre- specified statistical analysis plan and using statistical software STATA 16 (StataCorp 2016). All P-values were two sided and statistical significance was set at the 5% level. Both intention-to-treat analysis with multiple imputations (MI) and complete-case analysis were conducted and reported.

Descriptive analysis was conducted to describe mothers’ demographic characteristics, child BMI and BMI z-score, and secondary outcomes (i.e., children’s dietary behavior and activity behavior). Pearson’s Chi-squared tests examined the differences in mothers’ demographic characteristics between intervention and control groups.

Multiple linear regression models investigated effects of the intervention on child BMI and BMI z-score at 3 years of age. Multiple logistic regression models were fitted to investigate effects of the intervention on secondary outcomes. Adjusted odds ratios (AORs) were calculated. Since the randomization at age 2 years was stratified by group allocation in the previous trial [14], all multiple regression models were adjusted for their previous group allocation. Interactions of intervention allocation with family socio-economic status (SES) (based on annual household income) and language spoken at home were tested. When a significant interaction was found i.e., between intervention allocation and annual household income [note: SES was found to be associated with childhood obesity [4, 22]]; we conducted further subgroup analyses.

Missing data analyses were conducted for study outcomes to examine the patterns and mechanisms of missing data. Little’s test was conducted to test if missing was completely at random (MCAR). Models for missingness were also fitted to examine whether missing was at random. Since the missing was at random, MI by chained equations was used to address potential bias due to missing values. We imputed all missing values for the full intention to treat analysis of all 662 randomized participants. The imputation model predicting missing outcome values was based on all plausible observed values of outcomes, dietary and activity behaviors and family demographics at baseline (2 years of age) and at 3 years of age. We used 20 imputations each time which gave a relative efficiency of 99% [23], a similar approach to that used in our previous studies [11, 12, 19, 20].

Source link

Related Articles

Leave a Reply

Stay Connected

- Advertisement -spot_img

Latest Articles

%d bloggers like this: