401076 Introduction to Epidemiology

Subject Code :
401076
Country :
AU
University :
Western Sydney University

Overview

Reflected in the title and abstract, Golley et al. (2007) conducted a randomised control trial (RCT) to assess the effectiveness of parental focused programs to help reduce childhood overweight in response to the obesity epidemic. The public health importance of this is not detailed by the authors, however, an epidemic, by definition, suggests response measures are needed (Webb & Bain, 2011).

Golley et al. (2007) remark only one previous study investigating a parental program on child weight conducted in Israel. This justifies conducting the RCT, as consistency across study populations is needed to clarify causation and improve external validity (Rothwell, 2005; Webb & Bain, 2011). Moreover, the central element of family involvement to child weight is highlighted, noting that despite parents having such immense responsibility, they often receive little support. This justifies the rationale of the study intervention; providing parental skills appointing caregivers as the agents of change to their child’s weight. The authors hypothesize that parents may have healthy lifestyle knowledge, but lack the skills to implement them with their children. The authors examined this through a three arm parallel RCT; the first arm was a combined parenting skills training with lifestyle education program, the second was parenting skills training alone and the last was the wait list control group.

Although the authors don’t specify a target population other than children, a clear inclusion/ exclusion criteria was applied. Children included in the study were 6 to 9 years of age, overweight, and at Tanner stage 119 with caregivers willing to attend sessions and English literate. Children excluded had a BMI z score over 3.5 with a diagnosed syndrome cause of obesity or chronic illness or a sibling enrolled in the study. These exclusion criteria make sense as i) the intervention is targeted around modifiable behaviours and ii) siblings in the study would violate independent observations and can cause potential intervention contamination (Elwood, 2007; Webb & Bain, 2011).

The main outcome, children’s weight was measured proximately through BMI and waist circumference z scores. Measurements were taken at baseline, 6-month (intervention groups) and 12-month follow up (all participants). Allocation of participants was computer generated 3-block design randomization and stratified by gender and site of recruitment.

Children’s BMI z score, displayed no statistical difference between groups (p >0.05), meaning any differences between group outcomes could be due to chance variation. Despite this, the authors reported BMI z score reduced by 9%, 6% and 5% for those in the combined, parenting skills only and control group respectively (p= 0.76) and the proportion of BMI z score increase in each group was, 45%, 24% 19% in the control, parenting skills and combined group respectively (p= 0.03). The following primary outcome was children’s waist circumference z score. Similarly, when comparing groups, no significant difference between groups was found (p >0.05). Despite this, within group differences were reported displaying 9%, 5% and 0.6% z score reductions for those in combined (p= <0.01), parenting skills (p= 0.05) and control (p >0.05) group at 12 months compared to baseline. Sub-group gender analysis was also performed, which found no significant differences between either intervention or control group at baseline, 6 month or 12 month follow up.
The authors concluded parenting skills and promoting healthy lifestyle may be an effective approach to children’s weight management with a clear gender effect. This

conclusion could be misleading as no significant differences were found between intervention groups overall or by sub gender analysis. The authors recommend future research to consider accounting weight reductions in control group and primary and secondary analysis by sub-gender. This recommendation makes sense as the authors observed some differences within groups in results when stratified by gender, but did not account for this prior and sample size calculation was not powered for BMI reduction in the control group.

Internal Validity

To examine whether the results are accurate inferences, the overall presentation of results and examination of possible threats to validity, such as bias and confounding need to be considered (Mark & Reichardt, 2001).

Volunteer and consent bias are potential issues in the validity of this study. Participants were recruited via public media and school newsletters. Additionally, following telephone and medical screening of the 193 eligible, 111 attended baseline measurements (57.51%). Those who volunteer and consent to participate in health studies, generally tend to be more interested in health than those that do not (Brassey et al., 2017; Webb & Bain, 2011). Thus, those who participated would likely be more   interested in weight status of their children than those who did not volunteer. Additionally, parents who feel negative about their child being overweight may be less likely to participate. Therefore the lack results may be partially explained that parents who volunteered, regardless of intervention or control assignment, were interested in reducing their children’s weight and may have been engaged in other measures to manage their child’s weight and thus producing no significant results between groups.

Another potential source of bias is participant attrition. Of the 111 who started the trail, a total of 20 (18.02%) dropped out (5, 8 and 7 for control, parenting skills and combined group respectively). Older age, higher BMI z score and waist circumference z score were associated with non-participation in follow up. As drop out fairly distributed across groups this reduces the potential for bias. Additionally, although some systematic differences were observed for those who dropped out, intention to treat analysis was performed to adequately control for those differences and thus minimal bias occurred (Elwood, 2007).

Although appropriate analysis was performed, not all study results are clearly presented. Participation rate, through flow chart display, and descriptive measurements are adequately displayed, however statistical inferences are not clearly and objectively reported. In all tables displaying analysis, 95% confidence intervals are not used; instead the standard deviation is displayed (albeit can be used to manually calculate the 95% confidence interval). Additionally not all p-values are displayed for statistical analysis. This makes it difficult to judge the precision of results and significance of all findings. The authors tend to report on the significant post hoc within group associations found. This can be potentially misleading for the reader as no significant differences were observed between groups and differences from baseline to outcome within each group could be due to confounders and/or chance findings.

Although group differences could be a potential confounder, strong control for this is reflected in the study design (RCT). Random allocation to either intervention or control ensures known and unknown confounders are distributed evenly among groups. Therefore the only known difference between groups is the intervention, meaning that difference in outcome are attributed to receiving the intervention (Kendall, 2003; Webb & Bain, 2011). Randomisation was stratified by gender (a well-known confounder) and site of recruitment (potential confounder as people tend to be systematically different by their geographic location), which allow even distribution of these known confounders among groups (Australian Institute of Health and Welfare, 2016; Kendall, 2003). Reportedly, the groups were similar in measured characteristics at baseline.

Despite this significant within group outcomes cannot be attributed to the intervention, as outcome measurements were not significant when groups were compared and thus significant within group findings could be chance findings (Alosh et al., 2015). Blinding was an additional mechanism in the study design used to control for bias.

Single, rather than double blinding was implemented and thus some bias could have   occurred. A single trained blinded assessor took baseline and follow up measurements.

This meant that the assessor, knowingly or unknowingly, did not give favorable outcome measurements to the intervention groups (Kendall, 2003). Participants, on the other hand were not blinded to receiving the intervention. In this context, it was less warranted, as a the nature of placebo effect, from the parent’s receiving intervention to the child’s objective measurements is less conceivable (Australian Department of Health, 2004). Additionally, blinding participants was not feasible, as parents will know whether they attend sessions or not. Consequently, parents in the control may not have wanted to feel their child was worse off and may have engaged in other programs, behaviours or paid more attention to their child’s weight. This could have been prompted through the healthy lifestyle pamphlet received by those in the control and parenting skills group. One control the authors took against prompting was minimal contact with control participants to reduce potential therapist placebo effect that may have encouraged control parents to help their child’s weight status. Despite this measure, control was not implemented in the case that parents were attending other health focused family programs. Thus parents in either intervention or control group may have been engaged in other health promotions. This may have confounded the significant results observed within intervention groups from baseline to follow up. It may also explain the lack of significant results between groups if control participants were also engaged in other external measures to help their child’s weight status.
An additional explanation for lack of significant results is possible intervention contamination. Randomisation was applied at individual level and not clustered, thus some contamination is possible. For instance, if parents were recruited from the same     school but allocated to different groups, information from intervention programs may have been shared amongst parents to either intervention groups and/or control. This may have reduced the point estimate of either intervention group effectiveness, which could have caused a type II error explaining the non-significant results (Torgerson, 2001).

Potential for measurement error was very adequately controlled for in the study. The study had clear pre-defined measurements taken objectively by a single trained assessor with standardized procedure and instruments. This allowed for reliable and valid measurements to be obtained (Webb & Bain, 2011). A single assessor promotes reliability and consistency of results and mitigates potential for self-report bias by parents (Kendall, 2003; Van de Mortel, 2008). The authors however fail to mention if the anthropometric tools used have been tested for validity and accuracy which can lead   to invalid study results (Kendall, 2003). However with the mechanisms used the likelihood for measurement error is low and the measure of results are arguably valid and reliable. Additionally, a single trained provider conducted the parental program meaning that bias or error in outcome could not occur from receiving varied intervention.

Causal Inference

The authors assessed whether parenting skills program was just as effective as a combined program compared to no intervention. Overall, no significant differences were found between groups and thus either intervention programs were not proven to be any more effective in reducing children’s BMI or waist circumference z scores than no intervention.

Despite null findings, a failure to detect true effectiveness (type II error) of the program is plausible. Volunteer bias, non-blinding of participants, potential intervention contamination and lack of control for potential confounding effects of other health program involvement may have masked the true effects of the study. However, quality control measures counter against this. Standardised, valid and reliable measurements, assessor-blinded measurements, randomisation with stratification and intention to treat analysis safeguard against bias and a type I error and thus the interventions having no significant effect is also plausible.

Further to examining causal inference is consistency (Webb & Bain, 2011). Although the authors note only one prior compatible study conducted, a number of comparable parent led or parent involved studies exist, which show a mix of positive significant and non-significant results to children weight (Fagg et al., 2014; Hofsteenge et al.,

2014; Israel et al., 1985; Okely et al., 2010; Smith et al., 2013). Thus overall lack of consistency exists. Without combined systematic appraisal of the evidence, it is difficult to determine if bias and confounding can explain study results and the overall mix of findings and whether true causality of parent led programs and child weight exists and if so to what extent.

Furthermore, there are a number of external factors at various levels that can influence child weight as well as parent’s ability to manage their child’s weight. These include food advertising, food prices, nutrition labeling, physical activity programs, availability of recreational spaces and availability of other community initiatives that target healthy eating and physical activity for children (World Health Organization, 2012). With these larger scale, more influential external exposures present among intervention and control group it may explain the lack of significant findings between groups.

On the other hand, a causal strength of the study is temporality. The authors took baseline measurements before the intervention and took follow up measurements thereafter. However the results displayed no significant results. With this and the above arguments for and against type II error it is hard to say whether causal relationship truly exists.

Interaction and/or sub analysis

Interaction and sub-gender analysis was performed. The authors found a significant group by time adjusted by gender interaction effect for BMI z score (p= 0.04), however further analysis showed no significant differences between groups for boys or girls (p >0.05). The authors also performed group by time interaction analysis for waist circumference z score, which was also significant (p=0.03), but don’t mention they did this by gender.

Although finding no difference between groups for BMI or waist circumference z scores, the authors conducted post hoc analysis examining differences within groups between baseline and follow up by gender. Significant differences were reported for boys BMI z score in combined (p= 0.02, p= 0.01) or parenting skills alone (p=0.04, p=0.01) groups at 6 and 12 months respectively, compared to baseline and girls BMI z score in the control group (p= 0.02) at 12 months compared to baseline.

Overall stratifying by gender showed no significant differences between groups. Therefore, although some significant effects are shown for boys in intervention from baseline to 6 and 12 month follow up, the difference compared to no intervention is not significant and casual inference cannot be made to the intervention programs. Additionally the authors did not account for sub-gender analysis in their sample size calculation. Therefore caution needs to be taken in concluding these results as exploratory subgroup analysis are prone to chance findings (Alosh et al., 2015; Elwood, 2007). If a significant effect were found between groups by gender, with pooled analysis non-significant, this would indicate an effect modification relationship with gender as the effect modifier. Significant within group relationship by gender indicate a effect modification by gender, however it is difficult to attribute the programs as the causal exposure as there are no significant differences between groups by gender.

External validity

The authors don’t clearly specify a target population in whom they intend to infer the results back to. Presumably however, this would be to a population of a similar context,

i.e an Australian metropolitan population as the study was conducted at two metropolitan hospitals in South Australia (SA).

As the sample was not a simple random sample some issues around external validity are present. As discussed earlier, volunteer bias means the study sample may be systematically different to that of the broader population. Therefore any study results may not be applicable to the wider population for families who are less interested in health and child weight management.

Furthermore, demographic information suggests the study population was not diverse. Demographics displayed 72% were dual parent families, 98% had parents from white needs to those in metropolitan areas and families from non-white ancestry may have different cultural needs to those in the study group (Australian Institute of Health and Welfare, 2016, 2017; Ghosh et al., 2016). Underrepresentation of groups that would benefit the most means the effectiveness and appropriateness of the intervention programs, to those groups, is unknown and therefore can’t be easily disseminated to other populations.

In the discussion section of the study, the authors briefly list study power, intervention adherence and reduction of effect size via intention to treat analysis as limitations, which may have caused a failure to detect true effectiveness resulting in a lack significant findings between groups. Additionally they note ‘motivation’ of the control generalisability of results.

Conclusion

Overall, the study did not find the intervention of parenting skills alone or parenting skills combined with lifestyle education were significantly effective over no intervention. The potential reasons behind this may be due to volunteer bias, single blinding, potential intervention contamination, lack of control measures for other confounders and external influences and homogeneity in the study sample. Additionally, it may also be that the intervention is not effective in reducing child weight, particularly amongst the plethora of other larger factors that influence weight. The generalisability of study has clear limitations with underrepresentation of important groups. The study adds to the mix of study results in the examination of parent-led programs and additional studies on more diverse populations are needed.

Reflect upon the content of the topics we have covered in this course, and critically appraise the article you have selected by answering the following questions. You may use additional source material as you need to.

The following questions are only a guide to your assignment. These points should be addressed but the organisation and subheading is up to you.

1) Overview of the paper you selected.

-What is the issue addressed in the paper, is it important? Were the objectives clearly stated

- What was the study design? - What was the exposure /or intervention? -What was the main study outcome?

-What was the study population?

-What were the main findings? Are the results clearly presented?

Were the authors advise for future direction or about the study implication?

2) Appraise the internal validity

To what extent can the observed association between the exposure and outcome be attributed to non-causal explanations? Are the results likely to be affected by chance variation? -Are the results likely to be affected by biases, what kind of biases ? corroborate your arguments with details from the paper! are the results likely to be affected by confounding? If yes what are those that authors have missed? Back it up with referencing, has any interaction or sub-group analysis was conducted –bring evidence from paper

3) Do you believe a causal association between exposure and outcome is plausible?

This is based on your conclusion on internal validity but also appraisal of possible causal association based on several criteria we have covered in class

4) Appraise the external validity of the finding

can the findings be applied to the source population from which the study population was derived? -Can the study results be applied to other relevant populations? Appraise the overall quality of the discussion section.