The complex relationship between obesity and mental illness

People with severe mental illness have higher odds of obesity and die younger than the general population, with many dying from preventable physical diseases where obesity and overweight contribute (Walker Reference Walker, McGee and Druss2015; Liu Reference Liu, Daumit and Dua2017; Afzal Reference Afzal, Siddiqi and Ahmad2021). Obesity or overweight increase the risk of mental illnesses (Amiri Reference Amiri and Behnezhad2019; Steptoe Reference Steptoe and Frank2023) and vice versa (Luppino Reference Luppino, de Wit and Bouvy2010). Shared factors affect both conditions; adverse childhood experiences may increase risks of obesity (Schroeder Reference Schroeder, Schuler and Kobulsky2021) and mental illness.

Interventions for overweight and obesity

There are multiple interventions, with the mainstay being lifestyle change, which is difficult to achieve and sustain (Gadde Reference Gadde, Martin and Berthoud2018). Behavioural change interventions are complex; for reliable replication in research, systems for classifying behavioural change techniques (BCTs) have been developed, such as Michie et al's (Reference Michie, Richardson and Johnston2013) Behavior Change Technique Taxonomy Version 1 (BCT Taxonomy v1) (Box 1).

The role of mHealth

The field of mobile technologies for health (mHealth) is a rapidly evolving and globally important one (World Health Organization Reference Williamson, Bray and Ryan2021), but app development may outpace research. The integration of digitally enabled care into mainstream practice is a key objective of the National Health Service's long-term plan (NHS England 2019), making it imperative to understand the evidence base for mHealth interventions. However, the content of apps aiming to change health behaviour is variable (Lancaster Reference Lancaster, Sweenie and Noser2023; McKay Reference McKay, Chan and Cerio2024). Previous systematic reviews have taken broader perspectives, considering mHealth interventions for other conditions (McKay Reference McKay, Cheng and Wright2018; Jung Reference Jung and Cho2022) or including more electronic interventions (Hutchesson Reference Hutchesson, Rollo and Krukowski2015; Wang Reference Wang, Min and Khuri2020).

The ‘PICO’ of the Cochrane Review

The objective of the review in this month's Cochrane Corner (Metzendorf Reference Metzendorf, Wieland and Richter2024) was to assess the effects (outcome) of integrated smartphone applications (intervention) compared with no, minimal or any other active intervention, a different smartphone application or usual care (comparison), for adolescents and adults with overweight or obesity (population). Here we explore the ‘PICO’ in greater depth.

Studies were included if participants were adults (≥18 years of age) or adolescents (13–17 years) with overweight or obesity. Studies solely on people with conditions that might result in a different motivation for behaviour change, such as pregnancy or depression, were appropriately excluded because common components of smartphone interventions for overweight and obesity target motivation.

The review authors emphasise the importance of inclusion of adolescent studies; arguments for this emphasis include the increasing prevalence and long-term effect of childhood obesity (Stabouli Reference Stabouli, Erdine and Suurorg2021; Jebeile Reference Jebeile, Kelly and O'Malley2022). However, the review identified only two studies involving adolescents (L'Allemand Reference L'Allemand, Shih and Heldt2018; Vidmar Reference Vidmar, Salvy and Pretlow2019), with only one being usable for data analysis. This scarcity of data limits the generalisability to broader adolescent populations.

The mode of delivery of the intervention is well-defined, with appropriate exclusions. Included studies involved up to one monthly in-person session to reinforce an app-delivered intervention. The inclusion of studies with multiple in-person sessions initially suggests that the app component might not be the main intervention. However, multicomponent behavioural interventions for obesity ideally include 14 sessions in 6 months (Elmaleh-Sachs Reference Elmaleh-Sachs, Schwartz and Bramante2023), more than twice the cut-off for inclusion in this review. The decision to include studies with in-person components was also pragmatic: exclusion would more than halve the number of available studies. As the authors include studies with in-person components in the main analyses, the assumption appears to be that up to monthly in-person sessions do not alter the effectiveness of the apps. However, this may be incorrect, and exploring this via subgroup analysis would have been appropriate.

The authors did not explore the content of the interventions beyond setting the use of two or more BCTs as an inclusion criterion, owing to lack of sufficient details reported in some studies. Literature on using BCTs for health-related outcomes suggests that different BCTs may be associated with a difference in effectiveness of interventions (Black Reference Black, Johnston and Michie2020; Aguiar Reference Aguiar, Trujillo and Chaves2022). App quality and content might be an effect modifier; it is disappointing that information from included studies was insufficient to investigate this. Attempts to assess the content of apps were further hindered as only four randomised controlled trials (RCTs) (Stephens Reference Stephens, Yager and Allen2017; Locke Reference Locke, Falkenhain and Lowe2020; Eisenhauer Reference Eisenhauer, Brito and Kupzyk2021; Wilson 2023) studied commercially available apps or apps that were made available to the public after the trials.

The objective of the review was to understand the effects of smartphone apps, and nine outcomes were considered (please refer to the review itself). The review authors did well to consider adverse effects such as smartphone addiction, as well as more desirable outcomes. Outcomes were assessed in the short, medium and long term. Although this led to the exclusion of studies with follow-ups of less than 3 months, this choice is appropriate based on current understanding of core outcomes for weight management (Mackenzie Reference Mackenzie, Ells and Simpson2020).

Exploring the review's methodology

The review authors performed a thorough search for relevant studies and utilised independent screening of search results from databases to decrease the risk of bias. Although their search strategy is sufficient (Box 2), potentially more studies from low- and middle-income countries might have been retrieved on searching further geographical databases. Grey literature was included and the authors of studies awaiting classification and those of ongoing studies were contacted.

The review authors included only RCTs, which is appropriate for a research question focused on the outcome of an intervention.

Missing data affected the included studies, consisting of both loss to follow-up and missing data in the intervention groups. Multiple outcomes were self-reported and a lack of masking (‘blinding’) might have affected the results. The review authors describe how they critically appraised use of imputation in the studies and where missing data was a problem. If the standard deviation (s.d.) of an outcome could not be obtained, the review authors standardised using the mean of the pooled baseline standard deviation from studies where information was reported. Imputation was planned; however, the review authors do not describe this as being needed.

The COVID-19 pandemic potentially had an impact on individuals’ participation and the effects of interventions. For some studies, complete methodologies were difficult to obtain despite commendable efforts from the review authors to contact the studies’ authors. However, if outcomes were not obtainable, a study was excluded, potentially contributing to the risk of non-reporting bias in this review.

The review authors highlight that there was a high risk of bias regarding included studies. They used the Cochrane RoB 2 tool (Higgins Reference Higgins, Savovic and Page2019; Box 3) to independently rate studies to assess this risk; RoB 2 is a transparent, standardised approach for documenting potential flaws in a study that may result in bias (Sterne Reference Sterne, Savović and Page2019).

The review authors extracted and pooled data from similar studies, although variability in studies limited the amount of pooling possible. They appropriately performed separate pre-specified analyses for studies with different types of comparator groups, time frames or outcomes. Where meta-analyses were done the authors used GRADE (Box 4) to rate the certainty of evidence.

The results

This review found that the effects of apps as an intervention in adults were small compared with minimal or no intervention. For most results, the confidence interval (CI) for the mean difference crosses 0, so the true difference may be positive or negative; it was not statistically significant. For example, comparing an app with no or minimal intervention in adults, the mean difference in BMI at medium-term follow-up was −2.6 kg/m², but the 95% CI was −6 to 0.8 kg/m². This means we can be 95% confident that the true difference lies between −6 and 0.8 kg/m², so the true result of using an app could be an increase or a decrease in BMI.

For the few outcomes with statistically significant effects, they were often of questionable clinical significance (Box 5). When comparing an app with no or minimal intervention in adults in the short term, the mean difference in BMI was −0.5 kg/m² and the 95% CI was −0.3 to −0.8. This is a statistically significant decrease in BMI because we are 95% confident that the true difference lies between −0.3 and −0.8 kg/m². However, the threshold for clinically significant weight loss is commonly considered to be 5% weight loss (Williamson Reference Wilson, Driller and Johnston2015). At a BMI of 25 kg/m², the lower limit for inclusion in the review, a 5% decrease equates to a change of −1.25 kg/m². The difference between −1.25 kg/m², a clinically significant change, and −0.5 kg/m², the mean difference in BMI actually found, illustrates the low clinical significance of this result.

For adolescents only one study could be included for analysis (Vidmar Reference Vidmar, Salvy and Pretlow2019); compared with personal coaching, there was no difference in weight loss for those using apps as intervention.