This was a secondary analysis of data from the Doctor Referral of Overweight People to Low Energy total diet replacement Treatment (DROPLET) trial. The DROPLET trial recruited 278 participants seeking to lose weight and with a BMI ≥ 30 kg/m2. Participants were then randomly allocated to a total diet replacement programme or usual care. One hundred and forty participants were allocated to the usual care group, and of this group, 53 participants had their consultations audio-recorded. We report on data from all 53 individuals in the usual care group that had their consultations recorded to avoid introducing further selection bias.
Usual care comprised consultations with a GPN for the explicit purpose of weight loss [12]. The GPNs conducting the consultations had self-identified as confident in delivering a weight management programme. They were asked to continue consultations in their usual manner but additionally received the British Heart Foundation booklet ‘So you want to lose weight for good’ to use at their discretion. Consultations took place in general practices in the UK. Consultations from the trial were audio-recorded, and participants’ weights were recorded at baseline, 3, 6, and 12 months [12].
Audio recordings
Participants provided written informed consent for their consultations to be recorded. The recordings were not anonymised but were stored securely. MP3 audio recordings of weight management consultations were classified as ‘introductory’ or ‘follow-up’. ‘Introductory’ sessions were the initial meeting between the GPN and each participant, whilst ‘follow-ups’ were subsequent consultations intended to monitor and support the participants. Two audio recordings could not be linked to individual participants, so they were included in descriptive coding but not the analysis of weight change.
Taxonomy development, use, and coding
Weight management consultations given by GPNs were coded against three components: BCTs using the BCT taxonomy version 1 (BCTTv1) [13], dietary recommendations using a modified version of the Stok 2018 diet taxonomy [14], and physical activity recommendations using an investigator-designed physical activity taxonomy.
The BCTTv1 was chosen as it is an extensive consensually used taxonomy, allowing the identification of BCTs in weight management consultations. The NICE behaviour change guidelines use this taxonomy and include examples of how to use BCTs in the context of weight loss [15]. BCTTv1 contains 93 BCTs grouped into a hierarchy of 16 domains, with examples of a realistic use of each [13]. Three researchers (HT, EM, DAK) coded the content using the established BCTTv1 definitions, available in additional file 1, and examples following completion of a standardised online training for BCTTv1.
The Stok 2018 diet taxonomy was the only diet-specific taxonomy identified within the literature [14]. After listening to randomly selected sample data, the taxonomy was revised to capture the recommendations in the context of weight management consultations more thoroughly. Five recommendations were added: ‘ability to pay’, ‘rate of eating’, ‘food and drink substitution’, ‘total energy intake’, and ‘alcoholic drink intake’. The recommendation ‘dieting’ was removed, so that the specific recommendations used to encourage dieting, a purpose of the consultations, could be quantified. The adapted taxonomy had 30 recommendations within the domains of ‘food choice’, ‘eating behaviour’, or ‘dietary intake and nutrition’. The dietary taxonomy used is available in additional file 2. Researchers coded the content against the adapted taxonomy with pre-set definitions and as examples were not present in the original taxonomy, these were discussed and developed in a code rulebook, available in additional file 3.
A pre-existing physical activity taxonomy could not be identified in the literature, so we developed a taxonomy of 10 recommendations based on the FITT principle (frequency, intensity, time, and type) and a review of existing physical activity interventions [16]. The FITT principle was chosen because it highlights the 4 main planning stages of physical activity. Domains ‘lifestyle activity’ and ‘structured activity’ were chosen as Public Health England (PHE) recognises these as forms of physical activity [17]. The third domain with the single category ‘sedentary time’ was included because sedentary time is independently associated with mortality and morbidity and new PHE guidelines recommend reductions in sedentary time [17, 18]. The physical activity taxonomy used is shown in additional file 4.
Techniques/recommendations were rated as absent (coded as 0), present in all probability (coded as 1), or present beyond all reasonable doubt (coded as 2), as per the BCT training guidance [19]. Coding BCTs/ recommendations as 1 or 2 was based on the confidence of presence, as it was apparent BCTs/ recommendations met the criteria of definitions to differing degrees. One instance of the category present was sufficient to code the content. Consultations were listened to a minimum of 2 times. Coders were blinded to participants’ baseline weight and overall weight change until all coding was complete.
One researcher coded all consultations. Two further researchers independently coded a random sample (10%) of the consultations against the three taxonomies. Coded in batches of two or three, researchers then met to discuss the coding. Justification of the rationale for coding specific BCTs/ recommendations was discussed. Techniques/ recommendations that any researcher coded in consultations that were not coded by both other researchers were discussed in detail. This process was repeated iteratively until an agreement between all 3 coders was reached and final codes were applied.
A code rulebook was developed to aid coding. This consisted of rules on when to code specific BCTs/ recommendations and examples that should and shouldn’t be coded. For example, the BCT social reward was coded as ‘1′ if the GPN gave general reward (e.g. a GPN said well done to a participant, not directly referring to one behaviour in particular but generally from the overall outcome at the end of consultation). It was coded as ‘2′ if the GPN gave reward about a specific behaviour or outcome of behaviour (e.g. if the GPN congratulated the participant for losing 1 kg after weighing them). General coding rules stated to only code what the GPN said. It was not sufficient to code a BCT or recommendation if a GPN was simply responding to a participant’s statement, unless the GPN developed a discussion about the suggestion for an extended period of time. Guidance also stated that when GPNs questioned a participant about their behaviours, it was not sufficient to code questions asking about the past (e.g. ‘Have you swum before?’), but it was sufficient to code if the GPN asked questions relating to the future (e.g. ‘Will you try swimming next week?’).
Inter-rater reliability
Three random consultations coded for BCTs and 3 for dietary and physical activity recommendations were independently coded by each researcher, following iterative discussion and independent coding in triplicate of 10% of the consultations. Perfect agreement was considered when all coders coded the same values (0, 1 or 2) for a technique/ recommendation. Good and poor agreement, calculated using Krippendorff’s alpha with a SPSS syntax by Hayes, was considered to be an alpha value ≥ 0.600 and < 0.600, respectively.
Data Analysis
Analyses were carried out in Microsoft Excel v15 and SPSS v25.0 (Chicago, IL). Continuous variables are presented as means and standard deviations. Independent t-tests compared weight change from baseline at 3, 6, and 12 months in audio-recorded and non-audio-recorded groups. The Mann–Whitney U test was used to compare lengths of introductory and follow-up consultations.
Percentage usage of BCTs and recommendations across audio-recorded consultations was calculated. A weighted mean of each BCT and recommendation used in consultations was calculated. BCTs and recommendations coded as 1 were weighted half of those coded as 2, allowing for quantification of the fact that some BCTs were coded as present with more confidence than others.
For the following tests, BCTs and recommendations rated as 1 or 2 were deemed present and calculations were based on presence or absence rather than confidence of presence. Associations between lengths of consultations, and BCT and recommendation use with weight loss were assessed with the Pearson’s correlation coefficient.
Comparisons of the average and total length of consultations as well as the average number of BCTs and recommendations used per consultation were made between participants that had lost at least or less than 5% of their baseline weight at 6 months. This was based on the transtheoretical model of intentional behaviour change which argues long-term changes in behaviour may be assessed after 6 months [20]. Percentage usage of individual BCTs and recommendations used in consultations of the subgroup was compared using chi-squared tests and where appropriate Fisher’s exact test (Table S3, additional file 5).
Due to small numbers, we plotted the weight change and the usage of techniques/ recommendations per consultation by GPN, but were unable to perform additional formal exploratory analysis (Additional file 5, figures S1-4).
The level of statistical significance was set at p < 0.05.