Aims of the study

This study aims to investigate the effect sizes (ES) of lifestyle interventions on body weight and other cardiometabolic risk factors, such as waist circumference, blood pressure, blood lipids, glucose and insulin concentrations, in patients with psychotic disorders. The long-term effects of lifestyle interventions on body weight are included in the analysis as well. In addition, the effects of lifestyle interventions on depressive symptoms are investigated. Furthermore, we attempt to find effective components of the interventions, by comparing the studies with the largest ES and examine potential overlap and differences between the elements used in these successful interventions.

Inclusion criteria

A systematic search for all randomized controlled trials evaluating the effects of lifestyle interventions on weight management in patients with psychotic disorders was conducted until April 2014. The following electronic databases were searched: PubMed, Web of Science, PsycINFO and MEDLINE. Search terms included: schizophrenia or schizophrenic or psychotic or schizoaffective disorder or mental disorder or mentally ill or psychiatric disorders or severe mental illness or antipsychotic AND lifestyle intervention or diet or physical activity or nutrition or lifestyle or body weight or weight loss or weight management or exercise AND/OR weight gain or metabolic or metabolic syndrome or diabetes or health or somatic AND/OR psychological intervention or counseling or directive counseling or cognitive behavioral treatment. We included all randomized controlled trials (RCTs) that examined lifestyle interventions; i.e. interventions either targeting overweight patients in order to help them lose weight, or patients in the early stages of their illness in order to help them prevent antipsychotic induced weight gain. Interventions were considered lifestyle interventions when they had a nutritional element, physical activity and/or a psychological intervention aimed at weight loss or weight gain prevention. In eligible studies, all included subjects were diagnosed with psychotic disorders and study outcomes were either body weight or cardiometabolic risk factors (e.g. waist circumference, blood pressure, blood lipids, glucose and/or insulin). Pharmacological interventions were excluded, as were non-randomized studies and studies that did not qualify as lifestyle interventions. These selection criteria were first applied to the title. When the title did not present exclusion criteria (e.g. non-RCT, no intervention, no psychotic disorders) or was inconclusive, the abstracts of the articles were read and -where necessary- the full articles. Finally, the bibliographies of selected articles were searched for relevant references to be included in our analysis.

Outcomes and calculations

The first outcome was mean body weight change, measured directly at the end of the intervention. An overall ES for lifestyle interventions was calculated as well as separate ES for weight loss interventions and weight gain prevention interventions. The same analyses were performed for the long-term effects. A sensitivity analysis was performed using the ‘Clinical Trials Assessment Measure for psychological treatments’ (CTAM) to assess the quality of the studies [31]. The CTAM determines the quality of a study based on sample size and recruitment method, allocation to treatment, assessment of outcome, control groups, description of treatments and analysis. The psychometric properties of the CTAM were found to be adequate [31]. Only the studies marked as high quality (CTAM≥65) were included in the sensitivity analysis. Next, we calculated the ES of lifestyle interventions on cardiometabolic risk factors and depressive symptoms. With regard to cardiometabolic risk, we examined all available metabolic parameters, which include waist circumference, systolic and diastolic blood pressure, total cholesterol, HDL-cholesterol, LDL-cholesterol, triglyceride concentrations, fasting glucose concentrations and fasting insulin concentrations. For depression we considered the mean changes on the depression scales reported in the studies.

Data extraction

Relevant data that were extracted include a) patient characteristics [gender, age, diagnoses], b) intervention characteristics [duration, components, aim], c) study characteristics [dropouts, number of participants, blind assessments, control condition], d) means and standard deviations of baseline and endpoint or change scores of the outcome body weight, e) means and standard deviations of body weight at follow-up, f) means and standard deviations of baseline and endpoint or change scores of the components of all cardiometabolic risk factors as described above, g) means and standard deviations of baseline and endpoint or change scores of the outcome depressive symptoms.

In a meta-analysis all data is pooled and some variation in the intervention effects are to be expected. When the observed intervention effects are more different from each other than would be expected from random error (chance) alone it is called heterogeneity. Heterogeneity, characterized by the I² statistic, is considered low when I²≤25% and high when I²≥75% [32].

Meta-analytic procedure

To standardize the outcome among studies, Cohen’s d was used as a measure of ES [33]. It was calculated using the following equation:

The M_I indicates the mean pre-post intervention difference in the intervention group, whereas M_c is the mean pre-post intervention difference in the control group. SD_pooled indicates the pooled standard deviation for both groups within one report. The Standard Error of the ES (Cohen’s d) was calculated by using the following equation [34]:

in which n_i and n_c indicate the number of participants in the intervention group and the number of participants in the control group respectively, whereas d stands for Cohen’s d. We calculated separate ES for the intervention and control group in studies that did not report the mean weight change, but only the mean weight at baseline and endpoint. Cohen’s d was then estimated by subtracting the ES of the control group from the ES of the intervention group. Standard Error of the ES was established by the formula mentioned above. Missing data with regard to standard deviations were imputed from the included studies. This meta-analysis was performed and written in accordance with the PRISMA-guidelines [35].

Statistical method

The data were analyzed in RevMan Version 5.0 (Cochrane Collaboration software for meta-analyses) with the Inverse Variance method, using random effects models [36]. The χ² test, based on the Q-statistic, was performed to check for the homogeneity of the effects with I² as a quantifiable measure of heterogeneity [32]. Funnel plots were checked for asymmetry and Eggers’ tests were performed for each outcome to rule out publication bias [37]. Z-scores were calculated to test for overall effects.

Study overview

The search resulted in 656 records in four databases. A screening of the articles and their references initially resulted in 32 eligible studies that were discussed within the research group (JB, FJ, EC, MP), which led to the exclusion of seven more studies: in five of these studies the participants were not randomly assigned to the experimental or the control group [38]–[42], and the two other studies did not meet the criteria of a lifestyle intervention [43], [44]. The remaining 25 RCT’s were included in the meta-analysis (see Figure 1). Twenty-four RCT’s reported on body weight (sixteen on weight loss and eight on weight gain prevention), seven RCT’s had long-term follow-up data on body weight (four on weight loss and three on weight gain prevention), fifteen RCT’s reported on one or more cardiometabolic risk factors and four RCT’s reported depression. Six of the studies had missing standard deviations [45]–[50], which were imputed from the data pool. All articles were published in English up to the last date of the search (April 2014).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Prisma 2009 Flow Diagram.

https://doi.org/10.1371/journal.pone.0112276.g001

Study characteristics

A total of 1518 participants were included in the meta-analysis. Sample sizes across the studies varied from 14 to 291 participants with 52% male patients. The mean age across the studies varied from 26.1 (SD = 10.2) to 54.0 (SD = 9.3). Fifteen of the studies reported the use of mental health professionals, dieticians and/or exercise specialists to implement the intervention. Interventions were performed in the United States of America [22], [25], [50]–[55], South-America [21], Canada [48], the United Kingdom [45], [46], Spain [56], [57], Italy [47], [49], [58], Sweden [59], Switzerland [24], the Netherlands [60], Australia [23], Korea [61], Thailand [62], Taiwan [63] and China [64]. An overview of the study characteristics is provided in Table 1. The classification of weight loss study or weight prevention study was solely based on the aim of the researchers of the study. There were no differences with regard to the content of lifestyle interventions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Table of Characteristics.

https://doi.org/10.1371/journal.pone.0112276.t001

Overall results for interventions on body weight

The overall ES of lifestyle interventions on weight was −0.63 (95% CI = −0.84 to −0.42). Lifestyle interventions had a significant, beneficial effect on weight loss (p<0.00001) (see Figure 2a). The experimental groups in the weight loss intervention studies showed a higher reduction of the mean body weight than the control groups (ES = −0.52 with the 95% CI = −0.72 to −0.31, p<0.00001) (see Figure S1a). In the weight gain prevention studies the experimental groups gained less body weight than the control groups (ES = −0.84 with the 95% CI = −1.28 to −0.40, p = 0.0002) (see Figure S1b). The included studies in this meta-analysis showed moderate heterogeneity. The overall intervention effect had an I² of 70%. Among the weight loss intervention studies I² was 55% while the I² was 76% in the weight gain prevention studies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Forest plot a) describes the effects of lifestyle interventions on bodyweight.

Forest plot b) describes the longterm effects of lifestyle interventions on body weight. Forest plot c) describes the effects of lifestyle interventions on depressive symptoms.

https://doi.org/10.1371/journal.pone.0112276.g002

Study quality assessment

First, the funnel plots did not show asymmetry, but the Egger’s tests showed that publication bias could not be ruled out for diastolic blood pressure (p = 0.078) and long-term body weight (p = 0.027). All other outcomes had p-values above the recommended 0.1 on the Egger’s test, varying from 0.185 to 0.961 (see Table S1) [37]. All studies report randomization, but only 10 studies describe the randomization process [21], [22], [45], [46], [52], [56], [57], [59], [62], [64]. Five studies report the use of assessors independent from the study [55]–[57], [63], [64]; assessors were reported to be blind to group allocation in nine studies [21], [45], [46], [48], [52], [55]–[57], [64]. One study failed to adequately describe the treatment [57]. The mean drop-out rate among the studies was 17%, varying from 0–39%, but only four studies described an acceptable strategy for investigating drop-outs [51], [52], [56], [64]. Intention-to-treat analyses were reported in twelve studies [21], [45], [46], [51], [52], [56], [58]–[62], [64]. Four of the studies were considered to be of sufficient quality and had a CTAM score of at least 65 [52], [56], [59], [64]. Sensitivity analyses including these studies resulted in a decreased but still statistically significant ES of −0.55 (95% CI = −0.96 to −0.14, p = 0.008) of lifestyle interventions on body weight (see figure S2a).

Long term effects

Seven studies reported follow up data between two and six months after the completion of the intervention [20]–[26]. Their combined ES was −0.62 (95% CI: −0.93 to −0.31, p<0.0001) as is described in Figure 2b, which is in favor of the intervention. When we analyzed weight loss and weight gain prevention studies separately, we found that interventions aimed at weight gain prevention have a large long-term effect (ES = −0.85, 95% CI = −1.29 to −0.41, p = 0.0002) (see figure S3b) while the long-term effect of the weight loss interventions was moderate (ES = −0.46, 95% CI = −0.83 to −0.08, p = 0.02) (see figure S3a). With regard to heterogeneity, I² was 63% for the overall long-term effects of the intervention, 51% in the weight gain prevention studies and 62% in the weight loss studies.

Cardiometabolic risk

Ten studies reported cardiometabolic risk factors and in these studies there were no distinct differences between the weight loss interventions and the weight gain prevention interventions with respect to the content of the intervention. Therefore, in the analyses of the metabolic parameters, the two types were taken together. The lifestyle interventions demonstrated significant effects for waist circumference, triglycerides, fasting glucose and insulin. The ES for cholesterol (total, HDL-C and LDL-C) and systolic and diastolic blood pressure were not significant. An overview is provided in Table 2. Three of the high quality studies also reported on waist circumference and insulin [52], [59], [64]. A sensitivity analysis that only included the three high quality studies resulted in slightly decreased effect sizes that were no longer statistically significant for both waist circumference (ES = −0.30, 95% CI = −0.63 to 0.03, p = 0.08) (see figure S2b) and insulin (ES = −0.26, 95% CI = −0.64 to 0.12, p = 0.18) (see figure S2c). Regarding heterogeneity, only in the studies with fasting glucose I² was 0%. The remaining components had moderate to high heterogeneity with I² varying from 51% to 91% (see Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Effects of lifestyle interventions on cardiometabolic risk.

https://doi.org/10.1371/journal.pone.0112276.t002

Depressive symptoms

Four studies reported the effects of lifestyle interventions on depressive symptoms based on a continuous depression scale. Skrinar et al. (2005) used the depression scale of the SCL-90-R, Scheewe et al. (2013) used the Montgomery Åsberg Depression Rating Scale (MADRS) and the remaining studies reported the depression score of the Hospital Anxiety and Depression Scale (HADS) [45], [46]. The overall ES for lifestyle interventions on depressive symptoms was −0.95 (95% CI = −1.90 to −0.00, p = 0.05). An overview of the results is shown in Figure 2c. None of the studies reporting depression had a CTAM score of 65 or above, so these results should interpreted with caution. Heterogeneity among the studies with outcome depression was high (I² = 93%).

Intervention characteristics

Most of the studies had an intervention period of three months or less [21], [23], [24], [45]–[47], [49], [50], [56]–[58], [61], [62], [64]. Seven studies had an intervention period between three and six months [25], [51], [53]–[55], [60], [63] and four studies had an intervention period of twelve months or more [22], [48], [52], [59]. There were also differences between the intensity and duration of the supervised exercise sessions. Exercise sessions varied from two hours per week [48], [59], [60] to 45 minutes supervised exercise four times a week [50] or daily 30 minutes sessions [22]. We studied the interventions with large effect sizes for corresponding intervention elements. However, we did not find any element present in all or most of the interventions with the largest effect sizes, nor did we find corresponding elements for the interventions with the smallest effect sizes. We did find a difference in the ES of interventions depending on whether group- or individually based interventions were used. Five studies used an individual approach in their intervention [23], [45], [49], [56], [61]. Their combined ES was −0.67 (p = 0.0004) (see figure S4a). Ten studies presented their patients with a group intervention [21], [22], [24], [25], [50], [51], [55], [57], [59], [63].The group interventions had an overall ES of −0.36 (p = 0.002) (see figure S4b). A combined approach of a group interventions accompanied by individual sessions was used in five studies [48], [52], [53], [62], [64]. They showed the largest overall ES (ES = −0.99, p = 0.002) (see figure S4c).

Limitations

There was significant heterogeneity among the studies, suggesting that there were differences in the effectiveness of the lifestyle interventions that could not be a result of chance alone. Therefore, Cohen’s d has to be interpreted with caution. Based on the funnel plots and Egger’s test for asymmetry, we could not rule out the possibility of publication bias with regard to long term effects on body weight and the effects on diastolic blood pressure. This could indicate an overestimation of the reported effect sizes for these outcomes. Furthermore, the effects for some parameters of the metabolic syndrome and depression were based on a small number of studies. The CTAM revealed that the quality of the included studies was quite poor, with 21 out of 25 studies not meeting its standards. This questions the reliability of the effect sizes, since low quality studies tend to overestimate effects. Last, BMI change could be seen as a more meaningful outcome than weight change as it takes the height of the patients into account. Alternatively, we chose to report waist circumference and other metabolic parameters alongside body weight. Waist circumference is one of the main risk factors for cardiovascular morbidity [71]–[73]. Abdominal obesity stimulates insulin resistance, which can result in elevated triglyceride concentrations, diabetes and hypertension. All of these present an increased risk of developing cardiovascular diseases [73], [74].

Clinical implications and future research

Lifestyle interventions in general lead to body weight loss and prevent weight gain. However, as we found no evidence for the superiority of a specific intervention model of component –other than using an individual approach- we cannot make recommendations regarding the content of lifestyle interventions. To find out which elements are more effective than others, we should test them all separately, which would be an expensive and time-consuming exercise. Also, an intervention could be more than the sum of its separate elements. It might help if future studies provided clear and detailed depictions of the content of their lifestyle interventions.

Lifestyle interventions might improve other metabolic risk factors than body weight alone, and might also improve depressive symptoms, even though only few studies reported on these outcome measures. We urge researchers investigating effects of lifestyle interventions in people with psychotic disorders to include these measures to further substantiate these findings. Given their positive effects on multiple outcomes, we recommend lifestyle interventions to be listed among other evidence based psychosocial treatments for psychosis and to be included in clinical guidelines.

The CTAM considered most of the included studies to be of poor quality. We would like to underline the importance of high quality research in order to obtain reliable results, as well as to urge researchers to properly describe the design and execution of their studies.