Chapter 3: The effects of multi-component weight management interventions on
4.14 Data analysis
4.14.6 Quantitative data analysis
4.14.6.1 Analysis principlesThe primary analysis in this thesis were conducted as a completers intention to treat (ITT) analysis and exploratory analysis was conducted as per-protocol analysis. ITT is considered the gold standard statistical approach in clinical trials (Gupta, 2011; Hollis & Campbell, 1999). ITT aimed to eliminate potential bias associated with drop outs, missing data or deviations from the initial sample size calculation. ITT analysis was defined as analysing participants according to initial randomisation, irrespective of what happened to the participant after this point, i.e., whether they complied with the intervention or dropped out from the study). There are several advantages to ITT as it aims to emulate the effects of treatments in the βreal worldβ (Gupta, 2011). Thus, it provides a more realistic estimate of the intervention effect as it allowed for dropouts and non-compliance of interventions which occurred in everyday life. Furthermore, ITT analysis, maintained the integrity of the randomisation process and thus the validity of the results. This allowed interpretation of any post-treatment differences between interventions to be attributed to the treatments and not due to differences in participant characteristics.
An additional analytic approach in clinical trials is per-protocol analysis. This can be defined as participants considered to have complied with the intervention protocol and further defined as having adhered to a specific dose of the intervention, i.e. attendance at 75% of intervention sessions. Participants were defined as completing the intervention if they attended 75% of the total sessions. This statistical approach can be used to further explore
conditions which participants fully adhered to the intervention. Descriptive statistics were used for participant demographics and all outcome measures. Results have been presented as means and SD for baseline continuous variables (Anthropometric outcomes, physical activity outcomes and health related quality of life) and frequencies and percentages (%) for categorical variables (Demographics, Ability and development, Health conditions).
Mixed linear models were used to examine the potential efficacy of primary and secondary outcomes. Mixed linear models were selected due to their capabilities over general linear models as mixed models may have accounted for correlated data or data with unequal variances which can arise from hierarchical data. For example, and relevant to this study mixed models can analyse data for individuals who have been selected from a cluster of participants.
To differentiate the proportion of the variance that is due to between cluster variation an interclass correlation (ICC) was calculated. The ICC was calculated as the proportion of total variance that is due to between cluster variation from the following equation:
ππ2/(ππ2+ππ€2)
Where ππ2 is the variance due to differences between clusters and ππ€2 is the variance due to differences between individuals within clusters. An ICC of less than 5% indicates there is no meaningful difference between clusters (Tabecjnick & Fidell, 2007).
Analysis was conducted to assess normal distributions of the data. Each variable was assessed graphically using a histogram with normal distribution curves, boxplots and Q-Q residual plots. In addition to visual inspection of these plots for normal distribution, skewness and kurtosis were tested using z-scores with < 1.96 representing normally distributed data. In circumstances that data were considered not normally distributed, factors affecting this were explored. Outliers were identified by examining residual Q-Q plots. Data points classified as outliers were assessed for their potential to influence results and therefore study conclusions. Sensitivity analyses were conducted to compare results with and without outlier(s). Discrepancies between the two analyses are reported and discussed in chapters five and six. In addition to examining the data with and without outliers, in some cases transformation of the data were assessed using logarithmic and square root transformations and normality reassessed.
All statistical analyses were carried out in accordance with a pre-specified SAP. The objective of the SAP was to provide detailed analysis of each outcome; this is presented in Appendix vii. All statistical data were analysed using SPSS 21 IBM statistical package (SPSS IBM, New York, NY, USA).
4.14.6.2 Primary outcomes
A mixed effects model was used to determine the mean difference in weight loss at the end of the intervention period (~12 months) from baseline, between TAKE 5 and WWToo. The mixed effects model accounted for the effects of clustering of participants (stratified by level of intellectual disability, number of participants within a cluster and presence of Down syndrome) and was adjusted for baseline weight. Within group analysis was also investigated from the mixed models. The ICC, adjusted mean difference (95% confidence interval (CI) and p-value are reported.
4.14.6.3 Secondary outcomes
Continuous secondary outcomes were analysed and reported similarly as described above. A logistic regression model was fitted for the categorical outcome, weight loss of 5% or more of initial body weight taking account of clustering and baseline adjustments listed above.
4.14.6.4 Process outcomes
Descriptive statistics were used to assess process measures such as attendance to sessions for each treatment group. Independent sample t-tests were performed to determine if there were any differences between the treatment groups.
4.14.6.5 Serious adverse events
Serious adverse events (SAE) were defined as an adverse event (an injury or newly diagnosed health condition) that induced hospitalisation or prolonged hospitalisation, results in persistent/significant disability or incapacity or is life-threatening or fatal. SAE were recorded after baseline visits by the researcher at follow up visits. Incidence of SAE were recorded for each treatment group.
4.14.7 Qualitative data analysis
In addition to the quantitative data collection methods, qualitative interviews were conducted with the research dietitians. The interviews with the dietitians were conducted retrospectively on completion of delivery of the intervention sessions and 12 month data collection. The interviews were conducted by an independent researcher with experience in conducting qualitative research and not otherwise involved in the study. The main aim of the qualitative interviews was to elicit the research dietitianβs views of the interventions, the practicalities of delivering the intervention and any challenges to implementing the intervention as intended. The interview consisted of semi-structured questions and is presented in appendix vii. The interviews were audio-recorded using Olympus DSS player 2300. The interviews lasted between 45 minutes and one hour. The interviews were transcribed and analysed guided by the thematic analysis framework by Braun & Clarke, (2006). This consists of an outline of the following six steps:
1. Familiarisation of transcribed data:
The researcher familiarised oneself with the data through repetition of reading the transcribed data and noting initial ideas that emerged.
2. Initial coding of data:
Initial coding of data, identifying relevant data extracts to support codes 3. Theme searches:
The data were grouped together into potential themes and sub-themes with supporting evidence.
4. Theme revisions
Themes were reviewed against codes and relevant themes were modelled in relation to each other in a thematic map.
5. Theme definitions
The specifics of each theme were defined, themes were discarded if no longer relevant in order to form a coherent collection of themes.