Effects of Lifestyle Interventions That Include a Physical Activity Component in Class II and III Obese Individuals: A Systematic Review and Meta-Analysis

Background In class II and III obese individuals, lifestyle intervention is the first step to achieve weight loss and treat obesity-related comorbidities before considering bariatric surgery. A systematic review, meta-analysis, and meta-regression were performed to assess the impact of lifestyle interventions incorporating a physical activity (PA) component on health outcomes of class II and III obese individuals. Methods An electronic search was conducted in 4 databases (Medline, Scopus, CINAHL and Sportdiscus). Two independent investigators selected original studies assessing the impact of lifestyle interventions with PA components on anthropometric parameters, cardiometabolic risk factors (fat mass, blood pressure, lipid and glucose metabolism), behaviour modification (PA and nutritional changes), and quality of life in adults with body mass index (BMI) ≥ 35 kg/m2. Estimates were pooled using a random-effect model (DerSimonian and Laird method). Heterogeneity between studies was assessed by the Cochran’s chi-square test and quantified through an estimation of the I ². Results Of the 3,170 identified articles, 56 met our eligibility criteria, with a large majority of uncontrolled studies (80%). The meta-analysis based on uncontrolled studies showed significant heterogeneity among all included studies. The pooled mean difference in weight loss was 8.9 kg (95% CI, 10.2–7.7; p < 0.01) and 2.8 kg/m² in BMI loss (95% CI, 3.4–2.2; p < 0.01). Long-term interventions produced superior weight loss (11.3 kg) compared to short-term (7.2 kg) and intermediate-term (8.0 kg) interventions. A significant global effect of lifestyle intervention on fat mass, waist circumference, blood pressure, total cholesterol, LDL-C, triglycerides and fasting insulin was found (p<0.01), without significant effect on HDL-C and fasting blood glucose. Conclusions Lifestyle interventions incorporating a PA component can improve weight and various cardiometabolic risk factors in class II and III obese individuals. However, further high quality trials are needed to confirm this evidence, especially beyond weight loss.


Introduction
Obesity is now recognized as the most prevalent metabolic disease world-wide, reaching epidemic proportions in both developed and developing countries [1]. In North America, the prevalence of class II and III obesity (Body mass index (BMI) 35 kg/m²) has increased rapidly over the last decade [2,3]. Severe obesity is associated with multiple comorbidities such as hypertension, insulin resistance, type 2 diabetes, dyslipidemia, cardiovascular disease, sleep apnea and cancer [4,5], and is often associated with musculoskeletal pain [6,7]. All these comorbidities further lead to impaired health-related quality of life [8,9]. The importance of obesity is also obvious when looking at the considerable resources dedicated to its treatment and care, which account for between 0.7% and 2.8% of a country's total healthcare expenditures [10].
Several strategies are recommended for the treatment of obesity, including dietary therapy, regular physical activity (PA), behavioral therapy (BT), pharmacotherapy, and bariatric surgery as well as combinations of these strategies [11][12][13][14][15]. Although, bariatric surgery remains the most effective treatment to decrease and maintain weight loss, as well as improve comorbidities and mortality [16,17], lifestyle intervention is recommended as the first step to achieve weight loss and to treat obesity-related comorbidities in subjects with severe obesity [12]. In addition, given the limited resources, lifestyle intervention remains an effective option to help more subjects with severe obesity [18] and subjects could also prefer less invasive treatment than bariatric surgery [19].
PA is an important component of lifestyle intervention and should be systematically included in lifestyle management components [12]. PA, self-monitoring, and continued follow-up contacts have been identified as key components of weight control [20]. In addition, several studies showed that PA presents several benefits in individuals with class II and III obesity [21], as well as in class I: improvement of morbidities, cardiovascular diseases mortality and quality of life [21][22][23][24][25]. However, non-surgical obesity programs in Canada include less PA support compared to nutritional support (73 vs. 93%) and have less PA professionals compared to dietitians (43 vs. 74%) [26].
Previously, a review and meta-analysis of lifestyle interventions in obese and overweight individuals concluded that they can significantly reduce body weight and cardiometabolic risk factors in the mid-to long-term [27]. However, no systematic literature review is currently available on the effect of lifestyle interventions (dietary intervention, PA, BT) specifically in class II and III obese individuals. Thus, the present systematic review aims to give an overview of lifestyle interventions that include a PA component (counseling, recommendations, education or exercise training) proposed to more severe obese individuals. We thus carried out a systematic review, meta-analysis and meta-regression on the effects of lifestyle interventions incorporating a PA component among class II and III obese on i) anthropometric parameters; ii) cardiometabolic risk factors; iii) behaviour modification; iv) and quality of life. The secondary objectives were i) to investigate the impact of sex, age, severity of obesity and metabolic disorders on the lifestyle interventions efficiency; ii) to compare lifestyle intervention modalities; and iii) to assess the long-term impact of lifestyle intervention in this population.

Information sources and study selection
This systematic review followed the guidelines of the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) [28]. The information sources and study selection methods were described elsewhere [21]. Briefly, the research was completed on November 16 th , 2012 across 4 databases (Medline, Scopus, CINAHL and Sportdiscus) using specific keywords and Medical Subject Headings [21]. Two independent reviewers screened all records according to titles and/or abstracts (AB and MMRF) and assessed selected full-text articles for inclusion and exclusion criteria (AB and MA). Disagreements were resolved by a third party (MFL) and reviewers' agreement was calculated using Cohen's kappa coefficient [29].

Eligibility criteria
The following inclusion criteria were applied: i) peer-reviewed original studies; ii) class II and III obese adults (>18 years; more than 75% of the sample with BMI35 kg/m 2 and no normal weight subject); iii) lifestyle interventions, incorporating a PA component (counselling, recommendations, education, or exercise training) and with at least one of these components: BT, diet, nutritional education or recommendations or counselling; and iv) at least one of these outcomes: anthropometric parameters (body weight, waist circumference), cardiometabolic risk factors (% of fat mass, lipid or glucose metabolism, blood pressure), PA or nutritional behaviors (energy expenditure or intake, recommended healthy behaviors), and quality of life.
BT was considered as an approach focusing on modifying the perception of the environment to increase stimuli that promote healthy eating and PA behaviours while decreasing stimuli that make healthy eating and exercise challenging [30]. Studies with interventions including jaw fixation, anti-obesity medication or bariatric surgery were not considered unless they included at least one lifestyle intervention arm and only patients enrolled in the arm of interest were considered. No language restriction was applied. Authors were contacted twice in case of missing or incomplete data for the study selection or when more details on the intervention effects or population were needed. When more than one publication studied the same cohort and had overlapping results, only the most recent was considered. nutritional recommendations or education, BT]. The frequency of contacts was categorized as low (<1 session/month), moderate (1 or 2 sessions/ month), and high (>2 sessions/month) or missing information [15]. Studies were categorized as exercise training only if a professional totally or partially supervised the exercise sessions. Non-supervised exercise programs corresponded to interventions with individualized exercise plans without supervision. PA recommendations category corresponds to interventions that provide only general PA advices or studies that did not provide enough details about the intervention. Means with standard deviation (SD) of baseline, different evaluation time, post-intervention and follow-up outcomes of interest and p-values were reported. Missing baseline characteristic for subgroups or final sample were replaced by the initial whole baseline population characteristics means. Only results from intention-to-treat analyses were extracted, when completers' data were also available. All extracted data were double checked by another reviewer (KBV). Disagreements were resolved by a third party (MFL).

Quality assessment in individual studies
The quality assessment of the included articles was performed (AB) using the Quality Assessment Tool for Quantitative Studies developed by the Effective Public Health Practice Project [31], as reported previously [21]. A second independent investigator (MA) conducted quality control of one third of randomly selected articles (n = 19). We reassigned study design according to the data used in our analysis. For example, if a study randomized subjects in lifestyle intervention with or without jaw fixation, we considered the design as an uncontrolled clinical trial since we only used data of the control group.

Statistical analyses
Although some of included studies were controlled studies, for the present meta-analysis there were not enough "true" control groups to pool only randomized controlled trials. For this reason, all study groups were included in a longitudinal meta-analysis. Studies of weak quality were not included in the meta-analysis [32], as well as studies with only follow-up results or missing result data [33][34][35][36][37]. Subgroups analyses were performed according to the intervention lengths: short-term (<6 months), intermediate-term (6-11.9 months) and long-term (12 months) [15]. Studies with variable length of intervention for each subject [38] and no exact length of intervention [39] were excluded from the subgroup analysis. Baseline data and postintervention outcomes were reported as absolute change in mean and the standard error of means (SEM). Associated SEM were calculated using SEM diff. = SEM baseline + SEM final-2r SEM baseline × SEM final. We assumed a moderately conservative coefficient of correlation r = 0.5.
Due to the heterogeneity among included studies, estimates were computed using a random-effect model with the DerSimonian and Laird method [40] that does not assume interventions to be similar and further includes inter-studies heterogeneity (Tau²) in the calculation [41].
Results are presented as mean difference with 95% confidence interval (CI). Heterogeneity between studies was assessed by the Cochran's chi-square test (Q) and its extent was quantified through an estimation of the I². Expressed in percentage, the I 2 statistic describes the proportion of total variance in effect estimates due to the heterogeneity. Thus, homogeneous pooled studies should have an I² close to 0. Conventionally, 25%, 50%, and 75% respectively represent low, medium, and high inconsistency between studies [42,43].
In the presence of heterogeneity, meta-regressions with a random-effect model were performed to test different moderators available in the included studies and known to affect the final estimates such as sample size at inclusion, studies length in months, age of included participants and contact frequencies [44]. Data were analyzed with Open Meta-Analyst [45].

Study selection
The electronic search identified 5014 publications, among which 56 articles were included in the review (Fig. 1). Reviewers had a moderate agreement score concerning screening title/abstracts (kappa coefficient = 0.7) and an excellent score for eligible studies (kappa coefficient = 0.93) [29]. Given that one article [46] provided data from a subpopulation (normal or abnormal glucose tolerance subjects) of a larger study [47], we did not consider this study in the calculation of percentage in the following part (100% = 55 articles), but kept it for the part effects of metabolic disorders on the effectiveness of lifestyle interventions. Table 1 presents the characteristics and intervention modalities.
-PA: 6 individual 1h meetings on cognitive-behavioral methods to foster adherence in PA: goal settings, restructuring unproductive thoughts, addressing cues to exercise, preparedness for occurrences of barriers to exercise and relapse prevention -Nutrition: 6 group sessions of 1h on nutrition: understanding healthy eating. 1) provision of information on consequences, and 2) general encouragement.

(83)
Cognitive-behavioral methods for controlled eating: -Identical intervention except for nutrition: Cognitive-behavioral: additional array of behavior change techniques used in exercise support component. -General dietary recommendation -12 individual and group meetings of 1h based on social-cognitive and educational methods and self-efficacy theory for exercise and nutrition: orientation to exercise apparatus, self-management/self-regulatory methods ((e. g., long-and short-term goal setting, recording incremental progress, cognitive restructuring, stimulus control, and relapse prevention [preparing for barriers and recovering from lapses] and instruction in skills such as cognitive restructuring, stimulus control, and preparedness for occurrences of barriers to exercise. Goalsetting processes and self-regulatory skills, and development of perceived competence (i.e. self-efficacy) were used. Stress management component composed with deep breathing and muscle relaxation and instructions on appropriate prompts for utilization of these methods were given.
Annesi [65]  -Program manual focused on nutrition and exercise strategies Behavioral change support tools (pedometer, food/activity log, coaching call on problem solving and supportive feedback on progress) -Health-Coaching Topics: (1) explanation of the course to help the participant anticipate postsurgery change and long-term success.
(2) information on the types of surgical procedures (risks and benefits of weight loss), and helps to prepare lifestyle changes to ensure long-term success.
(3) support to create a plan with specific tactics including daily breakfast, portion size control, and journaling (self-monitoring).
(4) discussion of recommended levels of PA and benefits, introduces the use of a pedometer, kinds of PA that the participant can initiate and enjoy. (5) impact of common stressors (such as financial problems, conflict with family/ friends, job-related issues, etc.) on weight and weight-related behaviors.
(6) addressing strategies that help the participant in making progress despite challenges: anticipation of risky situations, how to deal with lapses and relapses, how to engage in problem solving, and seeking support to get through tough situations.
Meta-regression. As shown by meta-regression analyses (S2 Table), intervention length was negatively associated with mean weight loss (S4 Fig.) and waist circumference changes, but positively with SBP, total cholesterol, HDL-C, LDL-C, and fasting glucose changes (p 0.01). The frequency of contact was negatively associated with weight, BMI, waist circumference, and SBP changes (p 0.003). Age at inclusion was positively related to fat mass change and negatively with SBP, HDL-C, and fasting glucose changes (p 0.01). Body weight was positively associated with DBP (p = 0.002).

Effects of lifestyle interventions on behaviors
Fifteen studies of high and moderate quality assessed the effect of interventions on PA level and 8 on nutritional behavior changes. No meta-analysis was carried out given the heterogeneity of outcome assessment tools (questionnaire, interview, diary, pedometer) and reported data units (kcal/week, steps/day, categorical data, minutes, METs, fruits/vegetables consumed per day). Nine uncontrolled studies and one high quality RCT found significant positive impacts of 6 to 12-month lifestyle interventions on PA level [38,55,58,[63][64][65]69,71,72,76], and five uncontrolled studies on nutritional behaviors [47,58,65,71,76]. However, one RCT showed that the increase in steps/day after 4.5 months of intervention was not significantly different between the walking promotion group meetings and standard support groups (p = 0.46) [48]. In addition, a second RCT found no significant difference in PA level and eating behaviors between the preoperative medically supervised weight management program and the usual care groups [50]. Another study found no significant change in exercise level (KJ/day) and a compensatory increase in energy intake after 4 (+1004 kJ/day) and 7 (+836 kJ/day) months of intervention compared to the data obtained after the first month of intervention [80].

Effects of lifestyle interventions on quality of life
Only two studies of moderate quality assessed quality of life. The first study showed that a 2year weight loss program improved significantly 3 dimensions of quality of life: physical function (+12.8%), body pain (+7.2%), and general health scores (+11.6%) as measured with the Short Form-36 Health Survey in 30 class II and III obese individuals with obstructive sleep apnea [59]. The second study reported significant improvements after 1-month of interdisciplinary rehabilitation program in several important quality of life aspects like: sleep, dietary behavior, resistance to fatigue, mobility, activity, mood, emotions, participation, and self-control (Sat-P questionnaire) among 59 class II and III obese participants with sleep-disturbance related symptoms and disabilities [62].

Effects of sex, age, severity of obesity and metabolic disorders on the effectiveness of lifestyle interventions
Eleven articles provided comparison between females and males. Three studies showed that males lost higher amount of their initial body weight than females after a same short-length lifestyle interventions [53,85,86] without differences in blood pressure change [86]. However, over the 12-month period of lifestyle intervention, 2 studies reported better weight loss in females than males [53,69]. Other studies found no significant sex difference in body weight, waist circumference, and fat mass changes after interventions lasting from 3.75 to 12 months [39,54,79,81,[87][88][89].
Five studies have looked at the effect of age on the effectiveness of lifestyle intervention. Two studies did not identify any association between age and weight loss [39,88] and two others found no significant age difference between the regain and the weight loss groups [68,69]. Another study suggested that older age predicted greater systolic blood pressure improvement after 6-month of lifestyle intervention [66].
Eleven studies were interested in the impact of the severity of obesity on the effectiveness of lifestyle interventions. Seven studies found that subjects with higher initial BMI lost significantly more weight after interventions ranging from 2 to 61.5 months [35,39,55,68,74,79,87]. In contrast, other studies found no difference or association between baseline BMI class and weight loss. [64,69,70,88]. In addition, Unick et al. [64] showed also similar improvements in LDL-C, triglycerides, blood pressure, fasting glucose, and HbA1c at 12 months between class II and III obese individuals. However, class III obese subjects had smaller increase in HDL-C compared with class II (1.8±6.0 vs. 3.3±7.2; p<0.01).
Two studied provided results in normal or abnormal glucose tolerance subjects [46], and glucose-impaired and unimpaired subgroups [76]. Unfortunately, no statistical analyses were performed to compare changes between groups.
Martins et al. [52] found that a residential intermittent program and a commercial weight loss camp resulted in greater weight loss compared to a hospital outpatient program (22±13 vs. 18±12 kg vs. 7±10 kg; p < 0.001).
Three studies compared different supervised exercise modalities combined with diet and BT throughout 12 weeks of intervention. No significant weight loss difference was found between individualized compared to non-individualized training groups [54], and endurance exercise compared to strength and endurance exercise groups [57]. However, endurance and strength exercise training led to greater weight loss (−5.4 vs. −4.0 kg; p < 0.05) compared to non-individualized endurance and resistance exercise training. No significant weight loss difference was found between endurance training and these two groups in the study of Sartorio et al. [56].

Effects of lifestyle interventions on the long-term (follow-up)
Observational follow-up without intervention was performed only in 4 studies [33,54,62,70].
The first study showed no overall significant additional weight loss (Table 1) during the 18month follow-up period [70]. Nevertheless, 33.6% of subjects continued to lose weight by more than two BMI-points, 29.1% regained weight by more than two BMI-points and 37.3% maintained stable weight [70]. The second study also found an overall weight loss maintenance after the 6-month follow-up period, with 46% of subjects reducing body weight (1 to 5%), 51% regaining and 3% maintaining weight loss (< 1% change) [62].
Lafortuna et al. [54] reported that, after 6 months of follow-up, the individualized 3-week lifestyle group had a higher level of PA compared to the non-individualized group (p<0.05), displaying a trend for further decrease in body weight [54]. Another study provided results from the "Biggest Loser" telecast, where subjects were initially housed together until voted off by their peers every 6-11 days until all were home at 3 months. At 7-months follow-up, the intervention resulted in major reductions in body weight (−39%), body fat (−66%), serum insulin level (−52%), glucose (−21%), and HbA1c (−11%) [33].
Quality of life improvement observed during the first month of interdisciplinary rehabilitation turned down to baseline at 6 months of follow-up. However, the item scores dealing with sleep efficiency, problem solving and social interactions were still maintained at the end of the follow-up period [62].

Summary of evidences
From the 56 studies included in this review, the majority used uncontrolled design, and most of them were performed after 2010 mainly in women. The analytical part of this present review underlined that most lifestyle interventions containing a PA component are efficacious in class II and III obese individuals. In fact, significant effects were found for body weight, BMI, fat mass, waist circumference, blood pressure, total cholesterol, LDL-C, triglycerides, and fasting insulin.
Waist circumference, which is associated with visceral adipose tissue content [97] showed a pooled mean decrease of −6.9 cm after lifestyle interventions in class II and III obese individuals. This change is promising and could contribute in part to the improvement shown in the other health factors (blood pressure, total cholesterol, LDL-C, triglycerides, fasting glucose and fasting insulin), given the role of visceral adipose tissue in the metabolic alterations [98].
In accordance with other studies in overweight and obese subjects [99,100], anthropometric outcomes (weight, BMI, fat mass and waist circumference) decrease over time after lifestyle intervention in class II and III obese individuals. For example, the subgroup analysis showed that the higher decrease in weight was found with long-term studies (−11.3 kg) compared to shortterm and intermediate-term studies with −7.2 kg, and −8.0 kg respectively.
In contrast, a systematic review in overweight and obese subjects found no relationship between the length of the intervention and the percentage of weight loss [101]. In addition, our meta-regression results showed no linear association between intervention length and BMI, and fat mass in contrast with weight and waist circumference. The number of included studies and the absence of intervention diversity in the category intermediate length interventions (7 studies but only 3 tested interventions) could explain this discrepancy. Furthermore, because of incomplete reporting of anthropometric parameters (weight, BMI, fat mass and waist circumference), not all studies have been included in each meta-analysis explaining varying effects. Thus, it is difficult to conclude on the impact of the intervention length on anthropometrics parameters, probably because other factors can impact the results, as shown in our meta-regressions for contact frequency and age. In addition, the number of other potential predictors of weight loss is large (comorbid conditions, individuals' obesity history, socioeconomic factors like sex, employment, income, education and social status, individual's quality of life, psychological factors) [102].
Regarding LDL-C, our meta-regression results showed significant linear decrease over time with short-length interventions displaying larger decreases than longer studies. Therefore, time could be a significant moderator as identified by the meta-regression. Nonetheless, the result from the meta-regression should be tempered because for this studied outcome, there was no significant result for short-term studies. This absence of significant result suggests a high degree of variability and heterogeneity in short-term studies and the presence of confounding variables like the use of medications in long-term studies. This reasoning is justified by the large confidence interval as the I².
Concerning the lipid profile, although an overall effect was found, few studies were available and only one study with an intermediate-length was considered. For triglycerides, although not significant, a trend was found for body weight variation, supporting that improvements in triglycerides levels were more likely due to weight variation rather than study length as it is the case for total cholesterol. In fact, for this outcome long-term studies were associated with no significant change. Again, confounding effects of medication use can influence results.
No significant global effect was found for HDL-C in the meta-analysis, since we observed a significant decrease in short-term studies and a significant increase in long-term studies. In contrast with systolic blood pressure, fasting glucose, total cholesterol, and LDL-C, HDL-C tends to improve over time in the meta-regression, as anthropometric outcomes, with longlength studies having larger increases. It is possible that the different baseline comorbidities, medications or PA intensity between studies may explain this discrepancy. Indeed, a systematic review performed in obese adult concluded that the changes in cardiometabolic risk factors are more likely in subjects with abnormal baseline levels [103]. They also stated that "weight loss, irrespective of patient characteristics and intervention, does not uniformly improve cardiovascular risk factors. However, there is a lack of data to correlate weight loss with effect on markers of cardiovascular risk, as there may be weight loss thresholds" [103]. Weight loss or more exactly visceral fat loss is important, however behaviour changes (physical activity and diet) are also necessary to maintain and improve further cardiometabolic alterations on long-term [93,94,104,105]. In addition, medications have little impact on HDL-C level, thus allowing to better capture the effect of lifestyle modification.

Limitations
If results from the present review and meta-analysis are interesting, some limitations should be discussed. In fact, in the meta-analysis part, a high degree of heterogeneity was detected through the included outcomes. This heterogeneity can be explained by methodological aspects that were not considered in the meta-regression. A second limitation is the fact that we did not use control groups because of the limited number of controlled studies in the literature. This absence of control groups has probably overestimated the results. Third, some papers with relevant data may have been excluded because of missing BMI data, or lack of response from authors to our queries. Fourth, the lack of specific lifestyle key words in our research strategy may have introduced a selection bias. Finally, it cannot be excluded that publication bias could affect our findings.

Implications for research
Given the small number of high quality studies (n = 4), additional high quality RCT are necessary to improve the current evidence-based knowledge on the beneficial effects of lifestyle interventions including a PA component in class II and III individuals. To improve clinical interpretation, authors have to provide all data on BMI and % of weight loss.
Studies should also consider outcomes beyond weight loss, such as body composition, metabolic risk factors and quality of life. Indeed, as recommended recently by the European Association for the Study of Obesity, obesity management should more focus on ameliorating or maintaining fat-free mass and decreasing fat mass, manage co-morbidities, and improving quality of life and well-being rather than focus on body weight loss [106].
The assessment of health behaviors (nutrition and PA level) is important to reveal subjects' compliance and to better understand the implication of each lifestyle change in the results of the intervention. Studies on the effect of lifestyle intervention on health behaviors in class II and III subjects are scarce (27%; n = 15) and equivocal, probably due to the different designs, assessment tools (self-reported vs. objective method) and intervention modalities. Thus, future studies have to report subjects' adherence to the intervention and behavior changes to improve data quality.
Weight loss maintenance after lifestyle interventions seems to follow different patterns according to each subject [62,70]. Additional studies are needed to follow over the long-term the effects of lifestyle interventions on weight loss and other outcomes (body composition, quality of life. . .) in class II and III obese subjects, since currently only two studies provided this data [62,70]. Insufficient and equivocal results prevent us to summarize the effect of sex, age and obesity severity on the effectiveness of lifestyle interventions. Comparison between gender, age categories, obesity class, metabolic status, responders and non-responders could be another avenue of research to develop. Indeed, predictors of success have to be studied to help health professionals to identify non-responders and better adapt their intervention, since large weight loss difference can be observed between subjects [107].
Only 7 studies compared different intervention modalities. Thus, future researches should investigate the impact of specific intervention modalities (deliverance, length, contact frequency) to better understand optimal intervention. In addition, the report of effective intervention modalities, health professional types and subjects' characteristics (age, sex, ethnic group, and comorbidities) has to be improved in the studies to be replicated in clinical practice and improve knowledge transfer.
Finally, studies should also include assessment of implementation outcomes (compliance, adverse outcomes and satisfaction) and cost-effectiveness analyses to help health professionals, healthcare managers and policy makers to support lifestyle intervention implementation.

Conclusions
Lifestyle intervention is effective to improve health in Class II and III obese individuals. Although bariatric surgery is more effective than lifestyle interventions for the treatment of severe obesity and its comorbidities, some individuals have striking response to lifestyle interventions and the number of surgeries performed is insufficient to treat all severely obese individuals. Therefore, lifestyle programs in the hospital and/or primary-care settings should be developed and supported.
Supporting Information S1 PRISMA Checklist. The PRISMA checklist. (PDF) S1 Table. Quality assessment results categorized by design in overall studies included. UIS = uncontrolled interventional studies; CCT = controlled clinical trial; RCT = randomized controlled trial. (PDF) S2 Table. Impacts of length, sample size, age of participants, intensity of contact and weight change on the efficacy of lifestyle intervention in class II and III obese individuals. Note: Meta-regression is interpreted as an analysis of regression; the sign gives the direction of the relation (see S3