Study selection

Figure 1 shows the PRISMA flowchart for the included studies. A total of 2469 articles were retrieved from the electronic searches. One additional study which meets the inclusion criteria was identified on Google Scholar through backward citation searching.^{Reference Park12} This article was not retrieved through the electronic searches because the article was not indexed in any of the mainstream databases searched. Therefore, in total, 2470 articles were identified for screening. After removing 395 duplicates, 2074 articles remained. Of these, 2026 articles were excluded at the first round of screening. The full texts of the remaining 48 articles were ordered. Of the 47 articles whose full texts are available, 33 studies were excluded. Reasons for exclusion include not assessing the intervention of interest (18/33, 54.5%) or the population of interest (6/33, 18.2%),^{Reference Pezzot-Pearce, LeBow and Pearce21–Reference Galbraith, Rose and Rose26} not an economic evaluation (6/33, 18.2%),^{Reference Böge, Hahne, Bergmann, Wingenfeld, Zierhut and Thomas27–Reference Walburg, de Joode, Brandt, van Tulder, Aiaanse and van Meijel32} protocols (2/33, 6.1%)^{Reference Maurus, Hasan, Schmitt, Roeh, Keeser and Malchow33,Reference Lewis, Chondros, Mihalopoulos, Lee, Gunn and Harvey34} and not in English (1/33, 3.0%).^{Reference Duberg, Möller and Taube35} The authors of the two protocols were contacted to ask for the results of their study but no responses were received. In total, 14 articles reporting 9 studies met the predefined inclusion criteria and were included in the review. The Cohen's kappa score for inter-rater agreement was 0.49, implying a moderate agreement between the investigators.^{Reference Landis and Koch36} A list of included studies is reported in the Supplementary Materials, Section 3.

Fig. 1 PRISMA flowchart for the literature search.

Study description

Table 1 summarises the characteristics of the nine included studies. All included studies were conducted in high-income countries: the UK (4/9, 44.4%),^{Reference Park12,Reference Osborn, Burton, Hunter, Marston, Atkins and Barnes13,Reference Holt, Gossage-Worrall, Hind, Bradburn, McCrone and Morris37,Reference Heslin, Patel, Stahl, Gardner-Sood, Mushore and Smith38} the USA (2/9, 22.2%),^{Reference Meenan, Stumbo, Yarborough, Leo, Yarborough and Green39,Reference Janssen, Jerome, Dalcin, Gennusa, Goldsholl and Frick40} The Netherlands (2/9, 22.2%)^{Reference Looijmans, Jörg, Bruggeman, Schoevers, Corpeleijn and Feenstra41,Reference Deenik, van Lieshout, van Driel, Frederix, Hendriksen and van Harten42} and Belgium (1/9, 11.1%).^{Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43} The settings of the interventions were community mental health services (5/9, 55.6%),^{Reference Holt, Gossage-Worrall, Hind, Bradburn, McCrone and Morris37–Reference Looijmans, Jörg, Bruggeman, Schoevers, Corpeleijn and Feenstra41} primary care (2/9, 22.2%),^{Reference Park12,Reference Osborn, Burton, Hunter, Marston, Atkins and Barnes13} a sheltered housing organisation (1/9, 11.1%)^{Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43} and in-patient care (1/9, 11.1%).^{Reference Deenik, van Lieshout, van Driel, Frederix, Hendriksen and van Harten42}

Table 1 Study characteristics

BMI, body mass index; CBT, cognitive–behavioural therapy; CEA, cost-effectiveness analysis; CUA, cost–utility analysis; HbA_1c, glycated haemoglobin; PTSD, post-traumatic stress disorder; RCT, randomised controlled trial; TAU, treatment as usual.

a. Looijmans et al used one costing perspective (societal) for their economic evaluation, and two costing perspectives for their budget impact analysis (societal and third-party payer).

Regarding the population, all included studies recruited people with a mixed diagnosis of mental disorders except Park,^{Reference Park12} which recruited only people with schizophrenia and comorbid type 2 diabetes mellitus. People with schizophrenia were included in all nine studies, whereas people with bipolar disorder were included in six studies.^{Reference Osborn, Burton, Hunter, Marston, Atkins and Barnes13,Reference Heslin, Patel, Stahl, Gardner-Sood, Mushore and Smith38–Reference Looijmans, Jörg, Bruggeman, Schoevers, Corpeleijn and Feenstra41,Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43} Regarding the intervention assessed, all included studies compared the cost-effectiveness of usual care alone with MBLIs (that included a PAI component) with or without usual care. Two studies (2/9, 22.2%)^{Reference Holt, Gossage-Worrall, Hind, Bradburn, McCrone and Morris37,Reference Heslin, Patel, Stahl, Gardner-Sood, Mushore and Smith38} compared the intervention plus usual care with usual care only. Five studies assessed group-based interventions (5/9, 55.6%) and the other four assessed individual-based interventions (4/9, 44.4). The PAI component within the multicomponent lifestyle intervention varied greatly, including group or individual aerobic exercises, low to medium intensity exercises, walking, running, yoga, biking and indoor team sport to achieve and maintain weight loss. Various time horizons were used by the included studies: 6 months (1/9, 11.1%),^{Reference Park12} 12 months (3/9, 33.3%),^{Reference Osborn, Burton, Hunter, Marston, Atkins and Barnes13,Reference Holt, Gossage-Worrall, Hind, Bradburn, McCrone and Morris37,Reference Looijmans, Jörg, Bruggeman, Schoevers, Corpeleijn and Feenstra41} 15 months (1/9, 11.1%),^{Reference Heslin, Patel, Stahl, Gardner-Sood, Mushore and Smith38} 18 months (2/9, 22.2%),^{Reference Janssen, Jerome, Dalcin, Gennusa, Goldsholl and Frick40,Reference Deenik, van Lieshout, van Driel, Frederix, Hendriksen and van Harten42} 24 months (1/9, 11.1%)^{Reference Meenan, Stumbo, Yarborough, Leo, Yarborough and Green39} and 20 years (1/9, 11.1%).^{Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43}

The types of economic evaluation adopted by included studies were cost–utility analysis (CUA) (8/9, 88.9%), and cost-effectiveness analysis (CEA) excluding CUA (1/9, 11.1%).^{Reference Janssen, Jerome, Dalcin, Gennusa, Goldsholl and Frick40} CEA is one of the most commonly used types of economic evaluation in healthcare. Within a CEA, effectiveness outcomes are valued using a single disease-specific measure. For example, Janssen et al^{Reference Janssen, Jerome, Dalcin, Gennusa, Goldsholl and Frick40} – the only included study that adopted a CEA excluding CUA framework – used per kilogram loss of weight as the single disease-specific measure. CUA is generally considered to be a special case of CEA; the two analyses differ in how outcomes are measured, but the analytical approach is the same and the results are presented in the same way. Within a CUA, effectiveness is measured in terms of generic ‘healthy years’. Of the eight CUA studies included, all used QALYs as the effectiveness outcome. Seven studies (7/9, 77.8%) were pure trial-based economic evaluations, one study was a pure model-based economic evaluation^{Reference Park12} and one study was based on both trial and model.^{Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43} In terms of costing perspective, five studies used a healthcare perspective only (5/9, 55.6%); three studies (3/9, 33.3%) used two perspectives: a healthcare and social care perspective and a societal perspective; ^{Reference Holt, Gossage-Worrall, Hind, Bradburn, McCrone and Morris37–Reference Meenan, Stumbo, Yarborough, Leo, Yarborough and Green39} and one study used one perspective (societal) for their cost-effectiveness and two perspectives for their budget impact analysis (societal and third-party payer).^{Reference Looijmans, Jörg, Bruggeman, Schoevers, Corpeleijn and Feenstra41} Two studies reported that they had conducted subgroup analysis,^{Reference Holt, Gossage-Worrall, Hind, Bradburn, McCrone and Morris37,Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43} but only one of them reported results of subgroup analysis.^{Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43} None of the included studies reported on their consideration of patient subgroups, such as potential subgroups considered, the rationale for conducting or not conducting subgroup analysis, or any limitations.

Quality assessment

For the sake of simplicity, only the results of the 10-point version of the Drummond checklist are summarised here. The results of the full 35-point version are reported in the Supplementary Material, Section 4.

The results of the 10-point Drummond checklist are reported in Table 2 and graphically represented in Fig. 2. None of the included studies met all ten criteria outlined in the checklist. Criteria where studies performed poorly were the use of a time horizon not long enough to reflect all important differences in costs and outcomes (8/9, 88.9%), failure to test all important parameters whose values were uncertain in sensitivity analysis (8/9, 88.9%), failure to discount costs or consequences for differential timing despite adopting a time horizon of over 1 year (5/9, 55.5%), failure to capture all important costs or consequences for each intervention under assessment (3/9, 33.3%) and not performing an incremental analysis of costs and consequences of alternatives (1/9, 11.1%).

Fig. 2 Quality assessment of included studies. 0049CER, incremental cost-effectiveness ratio; N/A, not applicable.

Table 2 Summary of study results

BMI, body mass index; EQ-5D-3 L, 3-level five-dimensional EuroQoL; PSA, probabilistic sensitivity analysis; QALY, quality-adjusted life year; SF-36, 36 item Short-Form Health Survey; WTP, willingness to pay.

Findings of included studies

The main findings reported by the included studies are summarised in Table 2. In summary, compared with usual care:

1. (a) five studies found that PAI was more effective and more expensive;^{Reference Park12,Reference Meenan, Stumbo, Yarborough, Leo, Yarborough and Green39–Reference Looijmans, Jörg, Bruggeman, Schoevers, Corpeleijn and Feenstra41,Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43} for these five studies, whether PAI was considered to be cost-effective or not depended on the local willingness-to-pay (WTP) threshold for an additional unit of effectiveness (e.g. QALY);

2. (b) one study found that PAI was more effective and equally expensive, and therefore was cost-effective;^{Reference Deenik, van Lieshout, van Driel, Frederix, Hendriksen and van Harten42}

3. (c) one study found that PAI was equally effective and equally expensive, and therefore as cost-effective as usual care;^{Reference Heslin, Patel, Stahl, Gardner-Sood, Mushore and Smith38}

4. (d) one study found that PAI was equally effective and more expensive, and therefore not cost-effective;^{Reference Holt, Gossage-Worrall, Hind, Bradburn, McCrone and Morris37}

5. (e) one study found that PAI was less effective and less expensive.^{Reference Osborn, Burton, Hunter, Marston, Atkins and Barnes13}

The details of each of the above studies are as follows.

PAI more effective and more expensive than usual care (n = 5 studies)

Janssen et al conducted a trial-based economic evaluation to compare the cost-effectiveness of an 18-month group intervention including group exercise sessions and moderate-intensity aerobic exercise with no intervention in the USA.^{Reference Janssen, Jerome, Dalcin, Gennusa, Goldsholl and Frick40} This economic evaluation was based on a randomised controlled trial (RCT) which recruited 291 participants with SMI and overweight or obesity from community psychiatric rehabilitation out-patient programmes;^{Reference Daumit, Dickerson, Wang, Dalcin, Jerome and Anderson44} 58.1% of participants had schizophrenia or a schizoaffective disorder, 22.0% had bipolar disorder and 12.0% had major depression. They found that participants in the intervention group lost, on average, 3.2 kg more than the control group over 18 months. The ICER of the intervention was $501 per kilogram lost, which was higher than many commercial weight-loss plans or lifestyle interventions. One-way sensitivity analyses (which can be used to assess the sensitivity of the results to variations in a specific input parameter or assumption) were conducted for a series of parameters related to the cost of providing the intervention. The results of sensitivity analyses showed that varying the number of staff members per contact session changed the overall cost the most. Probabilistic or bootstrapping sensitivity analysis (which can be used to quantify the level of confidence in the results, in relation to the joint uncertainty of multiple parameters simultaneously) was not conducted. The findings of Janssen et al^{Reference Janssen, Jerome, Dalcin, Gennusa, Goldsholl and Frick40} are similar to the findings of the trial-based economic evaluation conducted by Meenan et al in the USA, who reported that participants with SMI lost an average of 4.4 kg at the end of the 6-month intensive group-based lifestyle intervention period and maintained a 2.6 kg loss at 12 months.^{Reference Meenan, Stumbo, Yarborough, Leo, Yarborough and Green39} The economic evaluation of Meenan et al was based on an RCT which recruited 200 individuals 18 years or older who had been taking antipsychotic medications for at least 30 days prior to enrolment and who had a BMI ≥27.^{Reference Yarborough, Leo, Stumbo, Perrin and Green45} The recruited participants had a mixed diagnosis of SMI, including affective psychosis (38%), bipolar disorder (32%), schizophrenia spectrum disorders (29%) and post-traumatic stress disorder (PTSD) (2%). Meenan et al reported that the ICER of the intervention ranged from $1940 to $2527 per kilogram lost depending on which costs and which patient group were included (e.g. completers versus intention to treat). One-way sensitivity analyses were conducted for certain measures and assumptions, the results of which showed that the single largest individual effects on estimated ICERs resulted from lowering the overheads rate and reducing salaries by 25%. Probabilistic or bootstrapping sensitivity analysis was not conducted. Neither Janssen et al^{Reference Janssen, Jerome, Dalcin, Gennusa, Goldsholl and Frick40} nor Meenan et al^{Reference Yarborough, Leo, Stumbo, Perrin and Green45} reported a WTP threshold; therefore, no conclusion can be drawn regarding the cost-effectiveness of the interventions. Verhaeghe et al used a combination of trial and Markov models to compare the 20-year cost-effectiveness of group-based exercise plus usual care with usual care alone for people with SMI in Belgium.^{Reference Verhaeghe, de Smedt, de Maeseneer, Maes, van Heeringen and Annemans43} The within-trial analysis was based on a cluster preference RCT which recruited 284 individuals with SMI living in sheltered housing.^{Reference Verhaeghe, Clays, Vereecken, de Maeseneer, Maes and van Heeringen46} The diagnoses of the recruited participants included schizophrenia (37.9%), mood disorder (24.5%), substance misuse (15.9%), personality disorder (14.4%) and other (7.2%). An age- and gender-dependent Markov model was then used to project the 20-year cost-effectiveness of the intervention, assuming a repeated yearly implementation of the intervention. Based on the results of deterministic analysis, they reported that the 20-year ICER of the intervention was €27 096 per QALY for men and €40 139 per QALY for women, results that are very similar to the average ICER based on the probabilistic sensitivity analysis (PSA) results (€27 624 per QALY for men and €36 571 per QALY for women). At a cost-effectiveness threshold of €30 000 per QALY, the intervention was deemed to be cost-effective for men but not for women. Comprehensive one-way sensitivity analyses were conducted to test a range of parameters/assumptions; the results showed that the model is most sensitive to the intervention effect and intervention cost in both men and women. The probability of the intervention being cost-effective was not reported.

Looijmans et al in The Netherlands conducted an economic evaluation and budget impact analysis for a web tool that can be used to generate a personalised lifestyle plan (including physical exercise) for people with SMI and monitor their progress.^{Reference Looijmans, Jörg, Bruggeman, Schoevers, Corpeleijn and Feenstra41} Their analysis was based on a pragmatic RCT which recruited 244 people with SMI managed by community mental health centres, whose annual physical screening showed one or more increased metabolic risk factors: waist circumference >88 cm (females) or >102 cm (males), BMI > 25 kg/m², fasting glucose levels >5.6 mmol/L or glycated haemoglobin (HbA_1c) >5.7% or >39 mmol/mol.^{Reference Looijmans, Jörg, Bruggeman, Schoevers and Corpeleijn47} They found that the intervention did not result in higher QALYs but reduced participants’ waist circumference by −1.03 cm (95% CI −3.42 to 1.35 cm; non-significant) after 12 months. The intervention was not considered to be cost-effective at 12 months. The results were robust for the number of participants per trained coach and the use of complete data only (one-way/scenario sensitivity analysis was only conducted for these two changes). The results of probabilistic sensitivity analysis showed that the probability of the intervention being cost-effective in terms of QALYs is around 40% for the whole range of thresholds explored (€0–300 000 per QALY). Park conducted a model-based economic evaluation using a decision tree for middle-aged adults with chronic schizophrenia and comorbid type 2 diabetes in primary care in the UK.^{Reference Park12} The intervention assessed was a group-based lifestyle intervention that focused on psychosocial education on diabetes, healthy eating habits and physical activity. Park found that the intervention resulted in reduced BMI (from 33.6 to 32.9 kg/m², P < 0.001) after 6 months.^{Reference Park12} The 6-month ICER of the intervention was £81.23 per extra unit of BMI loss and less than £700 per QALY gained. One-way and two-way sensitivity analyses were conducted for several parameters/assumptions related to the cost of the intervention and usual care; the results showed that when the cost of the intervention increased from £173 to £399, the intervention was no longer cost-effective. The results of PSA showed that the intervention had a 52.8% probability of being cost-effective, at a WTP of £20 000 per QALY gained. Park concluded that the intervention was cost-effective in the short term, but refresher sessions might help prolong the beneficial effects accrued beyond 6 months.

PAI more effective and equally expensive compared with usual care (n = 1 study)

Deenik et al assessed the cost-effectiveness of a holistic lifestyle intervention tailored to individual patients with SMI in a psychiatric hospital in The Netherlands.^{Reference Deenik, van Lieshout, van Driel, Frederix, Hendriksen and van Harten42} The intervention focused on decreasing sedentary behaviour, increasing physical activity and improving dietary habits. Resource use data before and after the implementation of the intervention were retrospectively retrieved from an 18-month cohort study.^{Reference Deenik, Tenback, Tak, Rutters, Hendriksen and van Harten48} In total, 114 individuals with schizophrenia spectrum and other psychotic disorders, who had been in-patients for at least 1 year were recruited. The results of adjusted linear regression models (adjusted for baseline costs and baseline differences) showed a non-significant decrease of €736.28 per patient per quarter year (95% CI –2145.18 to 672.62) in favour of the intervention. The results of PSA showed statistically non-significant cost savings against health improvements for all health-related outcomes (e.g. weight, blood pressure and quality of life) in the intervention group compared with usual care. Deenik et al concluded that PAI did not increase healthcare costs while improving health outcomes.

PAI equally effective and equally expensive compared with usual care (n = 1 study)

Heslin et al conducted a trial-based economic evaluation in the UK to compare the cost-effectiveness of a 15-month integrated health promotion intervention (IMPaCT) plus treatment as usual (standard community mental health team care) with treatment as usual alone in people with established psychosis.^{Reference Heslin, Patel, Stahl, Gardner-Sood, Mushore and Smith38} IMPaCT aimed to improve health and reduce substance use. It targeted exercise, diet, tobacco smoking, alcohol use, cannabis use, other illegal substances and diabetes (where applicable) and included cognitive–behavioural therapy (CBT) and usual care. The economic evaluation was based on a pragmatic multi-centre phase III cluster RCT which recruited 406 participants with psychotic disorder (ICD-10 diagnosis: F20–29, F31.2, F31.5), who were 18–65 years old and under the care of community mental health teams. They found that IMPaCT showed no difference in cost or health benefit compared with usual care. Results of bootstrapping showed that the probability of the health promotion intervention being cost-effective did not exceed 0.4 for WTP thresholds ranging from £0 to £50 000 per QALY. Comprehensive determinist sensitivity analyses were conducted and the conclusion was robust for all parameters/assumptions tested.

PAI was equally effective and more expensive compared with usual care (n = 1 study)

Holt et al conducted a trial-based economic evaluation in the UK to compare the cost-effectiveness of usual care with a 12-month intervention (weekly group sessions on psychoeducation and behavioural change targeting diet and physical activity).^{Reference Holt, Gossage-Worrall, Hind, Bradburn, McCrone and Morris37} This economic evaluation was based on an RCT that recruited 412 patients with schizophrenia, schizoaffective disorder or first-episode psychosis, who had been prescribed an antipsychotic for ≥1 month, were able and willing to participate in a group education programme and had a BMI ≥25 kg/m² (≥23 kg/m² for South Asian and Chinese backgrounds) or had expressed concern about their weight. They found that compared with usual care, the intervention was more expensive but did not show improvement in QALY or weight loss over the 12-month intervention period. A deterministic sensitivity analysis was conducted to test missing data mechanisms and a subgroup analysis was conducted for participants with different duration of SMI, but the results of these analyses were not reported. PSA was not conducted.

PAI was less effective and less expensive than usual care (n = 1 study)

Osborn et al conducted a trial-based economic evaluation in the UK to compare the cost-effectiveness of usual care with a primary care intervention that aimed to decrease total cholesterol concentrations and cardiovascular disease risk for people with SMI (bipolar disorder, schizophrenia or schizoaffective disorder and other psychoses).^{Reference Osborn, Burton, Hunter, Marston, Atkins and Barnes13} This economic evaluation was based on an RCT that recruited 327 participants aged 30–75 years with SMI, who had raised cholesterol concentrations (≥5.0 mmol/L) or a total HDL cholesterol ratio of ≥4.0 mmol/L and one or more modifiable cardiovascular disease risk factors. They reported that at 12 months, compared with usual care, the intervention was associated with a lower cost (mean difference –£895, 95% CI –1631 to –160) and lower QALY gains (−0.011, 95% CI −0.034 to 0.011). The results of bootstrapping showed that the probability of the intervention being cost-effective was 98% at a £0 per QALY WTP threshold and 89% at a £20 000 per QALY WTP threshold. No results of deterministic sensitivity analyses were reported.

GRADE evaluation

The GRADE table for the eight trial-based economic evaluations are reported in the Supplementary Materials, Section 5. All eight studies were rated as moderate except Deenik et al.^{Reference Deenik, van Lieshout, van Driel, Frederix, Hendriksen and van Harten42} Deenik et al was rated as low because it was based on an observational study.