In recent years, we developed and evaluated personalised lifestyle interventions, the BETTER programmes (BETER in Dutch, acronym for Move, Eat, Change). Underlying principle for all BETER programmes is that people with the same condition may have different underlying causes, so-called subtypes. In this follow-up project with a mixed methods design, we aim to evaluate and optimise the subtype-questionnaire/algorithm (study 1, interrater reliabiliy) and evaluate the digitised BETER programme, the BETTER App (study 2, case series design with qualitative and quantitative evaluation). The main questions it aims to answer are:

1. What is the inter-rater reliability of two subtype experts and criterion validity of the symptom questionnaire compared with the experts for identifying overweight subtypes?
2. How is the BETER app used and rated (process evaluation)? To answer question 1, participants complete a questionnaire and have two interviews with two experts. To answer question 2, participants use the BETTERapp for 6 weeks and complete a usability questionnaire after 3 and 6 weeks and participate in 1 or 2 focus group interviews. This study contributes to optimising the Minimal Viable Product of the BETER app to finally reach a mature version.

Backgroud: In the Netherlands, the number of people with chronic conditions has increased significantly in recent years and is expected to continue to grow (2). Treating chronic conditions is expensive and requires a lot of care and attention. This leads to high healthcare costs for both the patient and the government. The Dutch state spent 108 billion euros on medical and long-term care in 2021, the largest part of the 125 billion euros spent on Care and Welfare in total at that time (3). Chronic conditions have a major impact on people's quality of life. They can cause symptoms and limitations in daily life and also affect people's work participation.

Overweight and obesity are related to an increased risk of many chronic conditions, such as cardiovascular disease, type 2 diabetes and certain cancers. Lifestyle plays an important role in the development of overweight and obesity. It is known that an unhealthy diet and lack of physical activity are the most common causes of overweight and obesity. Therefore, it is important to take action and offer programmes to reverse these trends and promote healthy lifestyles (4,5). In contrast to a generalised approach, more personalised interventions match a person's specific needs, characteristics and situations and can therefore potentially lead to better results when it comes to sustainable changes in healthy lifestyles (6-8).

Personalised lifestyle interventions, the BETTER programmes, have been developed and evaluated by Zuyd University of Applied Sciences' lectorate of Nutrition, Lifestyle and Exercise in recent years. BETTER stands (BETER in Dutch) for Move, Eat, Change and the programmes focus on exercise, nutrition and behavioural change. Knowledge and insights from systems biological fundamental research provide the basis for these programmes to make lifestyle recommendations more personalised. There is increasing evidence, that people with the same condition, may have different underlying causes, which also makes lifestyle recommendations and interventions different (9). In the BETTER programmes, this knowledge is practically applied by working with subtypes.

The face-to-face BETTER lifestyle programme has been successfully applied to overweight and obese individuals. After completing the programme, participants lost weight, felt fitter and indicated they felt more in control of their lifestyle (10).

Within the BETTER programme, individual subtyping of participants took place by one or more experts. This is time-consuming and (relatively) expensive, especially for repeated measurements. In a pilot study, we investigated to what extent the subtype determined by means of a digital symptom questionnaire, including algorithm, corresponds to the subtype determined by an expert (criterion validity). First preliminary results showed that the outcomes from the questionnaire corresponded moderately to reasonably well with the expert classification. Based on these data, a machine learning algorithm was developed, capable of increasing the validity of the questionnaire (optimising weightings and dependencies of answers). This algorithm can be trained by adding new data (completed questionnaire and subtyping by expert). The algorithm works on the basis of the Semi-Supervised Classification principle (11).

Building on the results and conclusions from the above studies, the BETTER app is being developed in which the BETTER lifestyle programme for overweight people including the symptom questionnaire is offered in an automated and digitised way. Users can independently use personally relevant programme components in an accessible way, promoting self-management and self-regulation. The automated version of the symptom questionnaire plays an important role, since the programme content is, among other things, tailored to the user's underlying subtype. The automated questionnaire in the app therefore makes it easy to identify a change of subtype and adjust lifestyle advice accordingly. For the development of the content, the process to be followed and the design of the BETTER app, material from the previous BETTER programmes will be adapted. This development process takes place in co-creation with the target group in several iterations. This study contributes to optimising the Minimal Viable Product of the BETER app to eventually reach a mature version.

In this follow-up project, we aim to evaluate and optimise the subtype-questionnaire/algorithm (study 1, interrater reliability) and evaluate the digitised BETER programme, the BETTER App (study 2, case series design with qualitative and quantitative evaluation). The main questions it aims to answer are:

1. What is the inter-rater reliability of two subtype experts and criterion validity of the symptom questionnaire compared with the experts for identifying overweight subtypes?
2. How is the BETER app used and rated (process evaluation)?

Method Study design and measurements Sub-study 1: Symptom questionnaire To assess the inter-rater reliability and criterion validity of the subtype measurements, participants will be invited to a measurement session of up to 60 minutes. These measurement sessions will be offered on five different days and at different locations, and participants will be supervised by a researcher/research assistant. Participants go through three measurement sessions in separate rooms: (1) completion of the digital symptom questionnaire and demographic data (approx. 10 minutes), (2) interview with expert 1 (approx. 20 minutes), (3) interview with expert 2 (approx. 20 minutes).

Sub-study 2: Process evaluation BETER app This sub-study uses both quantitative and qualitative research methods in a case-series design. The baseline measurement (T0) takes place before the start of the app use, an intermediate measurement (T1) 3 weeks after the start and a final measurement at the conclusion of the app use (T2) 6 weeks after the start. At baseline, demographic data and expectations towards the app are collected. At T1 and T2, usage, experiences and ratings are mapped through a questionnaire (mHealth App Usability Questionnaire (MAUQ)) and focus group interviews. Log files automatically record how often a person logs in, which parts of the BETER app are used by participants and the duration of app use.

Study population Inclusion criteria: Persons aged 16 years or older who are overweight or obese (body mass index (BMI) of 25 or higher); exclusion criteria: Insufficient mastery of Dutch language, insufficient basic smartphone skills.

The BETTER app The BETTER app offers a 'tailor-made' lifestyle programme including the option of personal coaching for a duration of 6 weeks. The BETTER app is an automated and digitised lifestyle programme based on the previously developed and evaluated BETER programme. The BETER app was developed by researchers from the centre of expertise of Nutrition, Lifestyle and Exercise from Zuyd University of applied Sciences and the software developer HelloSunshine B.V. in co-creation with the target group. The app offers support for lifestyle behaviour change.

Data analysis To assess inter-rater reliability between the two experts, the Cohen's kappa (k) with standard error and percentage agreement is calculated between the two experts. A 95% confidence interval is used. Both the unweighted Kappa and the linearly weighted Kappa are calculated. By means of the weighted kappa, in case of difference, it can be examined whether this difference mainly occurs between certain subtypes and can be corrected for this (17).

For determining criterion validity, the degree of agreement is expressed as a correlation coefficient (r ≥ 0.8 is assessed as 'good' and used as a cut-off point). In addition, sensitivity, specificity and F1 score are assessed using a 5x5 table and the five "One versus Rest" ROC curves (18,19).

Process evaluation: all measured variables from quantitative measurements T0, T1 and T2 and logfiles are reported at group level for each measurement time point using descriptive statistics.

During the focus group interviews, data collection and data analysis take place partly simultaneously. We apply thematic coding and categorization already during the data collection process.

Data management and ethical considerations All data is stored on secure network drives and the anonymity of participants is guaranteed. Informed consent will be obtained prior to the measurements of substudy 1 and prior to completing T0 of substudy 2.

Inclusion Criteria:

• Persons aged 16 years or older who are overweight or obese (body mass index (BMI) of 25 or higher)

Exclusion Criteria: : Insufficient mastery of Dutch language, insufficient basic smartphone skills.

-

Persons aged 16 years or older who are overweight or obese (body mass index (BMI) of 25 or higher). N = 50 (sub-study 1); N= 15 (substudy 2)