Obesity and its cardiometabolic consequences have become a worldwide preoccupation, given the rapid increase in its prevalence in recent years. The literature on weight loss (WL) management is well documented, dieting (i.e. ‘restrict oneself to small amounts or special kinds of food in order to lose weight’ – Oxford Dictionary) being one of the most popular strategies, leading to significant beneficial effects on patients’ body weight (BW), body composition, and cardiometabolic health(Reference Shalitin, Ashkenazi-Hoffnung and Yackobovitch-Gavan1–Reference Most, Tosti and Redman3). Interestingly, while there is a high prevalence of dieting among individuals with elevated BMI, WL attempts also concern a significant and increasing proportion of normal-weight (NW) individuals(Reference Montani, Schutz and Dulloo4–Reference Julia, Péneau and Andreeva6). For example, among French dieters, for instance, 33 % of men and 57 % of women are NW subjects(Reference Julia, Péneau and Andreeva6). Indeed, the media and social pressure related to leanness, and the widespread notion of an ideal body shape may alter body perception, particularly among young adults(Reference Ordaz, Schaefer and Choquette7–Reference Fayet, Petocz and Samman10). As illustrated in a cohort of European university students, 60 % of women with an average BMI of 22·5 kg/m2 considered themselves as ‘a little too fat’(Reference Mikolajczyk, Maxwell and El Ansari9). Body image distortion is often associated with the desire to lose weight(Reference Yaemsiri, Slining and Agarwal11), and leads individuals to engage in dieting independently of any need for weight management in terms of cardiometabolic health. In contrast, while dieting is associated with health improvements in people with overweight or obesity, it conversely could constitute a potential risk factor for increased adiposity and cardiometabolic disease, particularly in NW people. Indeed, in 2015, Dulloo and collaborators suggested that NW individuals might have a higher risk to experience post-WL ‘fat overshooting’ than people with obesity(Reference Dulloo, Jacquet and Montani12). This phenomenon is well illustrated in the Minnesota Starvation Experiment, in which thirty-two healthy NW males sustained a 24-week semi-starvation, and lost an average of 25 % of their initial BW(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15). During weight recovery, adaptive thermogenesis and hyperphagia induced a fast rebound of fat mass (FM) (i.e. preferential catch-up fat), resulting in FM being fully recovered earlier than fat-free mass (FFM). However, the hyperphagia was found to persist until FFM returned to its initial value, leading hence to higher FM values than those prior to semi-starvation (i.e. fat overshooting)(Reference Dulloo, Jacquet and Girardier14,Reference Dulloo, Miles-Chan and Schutz16) . Importantly, the leaner the participant at baseline, the higher proportion of FFM lost during WL and the greater the risk of fat overshooting during recovery(Reference Dulloo, Jacquet and Montani12,Reference Dulloo, Miles-Chan and Schutz16) .
Interestingly, a recent narrative review suggested some beneficial effects (e.g. increased insulin sensitivity) of interventional dietary-induced WL in NW subjects, still pointing the lack of evidence among healthy lean individuals, since most of the available evidence concerns people suffering from metabolic disorders (such hypertension or insulin resistance)(Reference Magkos17). In addition, some observational studies showed benefits of energetic restriction in healthy NW individuals such as an improvement of lipid and glycemic profiles(Reference Fontana, Meyer and Klein18), or an important decrease in global mortality(Reference Willcox, Willcox and Todoriki19,Reference Willcox, Willcox and Todoriki20) . However, it concerned specific populations (e.g. Okinawans) who sustain a low-energy intake during all their lifespan(Reference Fontana, Meyer and Klein18–Reference Willcox, Willcox and Todoriki20), and it was well established that the ‘obesogenic environment’ of our actual societies, including high food availability, makes it difficult to maintain a low-energy diet on the long term(Reference Greenway21–Reference Hall and Kahan23). Thus, due to the high probability of regaining the weight lost after a diet, individuals engaging in dieting may experience repeated WL and regains during their lifespan, a phenomenon called ‘weight cycling’. Weight cycling is associated with an increased risk of cardiometabolic impairments such as hyperinsulinemia, insulin resistance, hypertension, and hypertriacylglycerolaemia, specifically in NW individuals(Reference Montani, Schutz and Dulloo4). In addition, associated with the fat overshooting theory described above, weight fluctuations are also related to major long-term weight gain (WG)(Reference Montani, Schutz and Dulloo4). However, the long-term effects of dieting on BW and body composition changes in NW healthy people remain unclear. In their prospective observational study, Lowe et al. showed that NW current dieter women gained almost three times more BW than non-dieters after an 8-month period (5 kg v. 1·6 kg, respectively)(Reference Lowe, Annunziato and Markowitz24). In contrast, in a study performed in a large cohort (n 4159, aged 18–29 years), women who declared dieting at baseline gained as much BW as non-dieters after 15 years, while in young men, dieters tended to gain more BW than non-dieters(Reference Korkeila, Rissanen and Kaprio25). However, the majority of longitudinal studies that concern weight evolution in the overall population have relied on self-reported data(Reference Korkeila, Rissanen and Kaprio25–Reference Pietiläinen, Saarni and Kaprio27) and have not considered body composition variations (i.e. FM and FFM).
While there is an increasing concern for weight control in healthy NW individuals leading them to regularly engage in dietary WL strategies, there is a clear need to question the potential effects of dieting on their BW and body composition. Therefore, the aim of the present work was (i) to conduct a systematic overview of the available studies involving dieting and WL, BW, and body composition in NW healthy people, and (ii) to determine the effect of dieting on subsequent and/or long-term BW in NW healthy individuals using a meta-analytic approach. In doing so, the results may help to better decipher the risks and/or benefits to healthy NW individuals engaging in dieting and to promote adapted public health messaging related to diet, based on robust scientific evidence.
Methods
This systematic review and meta-analysis followed the recommendations from the PRISMA statement(Reference Moher, Liberati and Tetzlaff28), and the protocol was prospectively registered in the International Prospective Register of Systematic Reviews database (registration number: CRD42021281442).
Literature search strategy and article selection
The following electronic bibliographic databases were searched from April 2021, and the last run was performed in June 2022: PubMed-Medline, Embase, and Google Scholar. The search terms were a combination of medical subject headings (MeSH terms) and text words (title and abstract) which were adapted for use in each database. The main keywords selected were ‘weight gain’, ‘weight change’, ‘weight loss’, ‘body weight’, ‘body composition’, ‘normal-weight’, ‘healthy’, ‘non-obese’, ‘diet/ing/ietary’, and ‘caloric restriction’. Limits were set to include all papers in healthy human adults published after 1945. The search was followed by a careful selection of eligible studies according to the criteria outlined below. References from narrative or systematic reviews and eligible publications were also screened to find additional records.
Inclusion and exclusion criteria
The inclusion criteria were as follows: (1) healthy (without pathology or co-morbidity) participants with initial BMI between 18·5 and 24·9 kg/m2 (for the group mean); (2) adult women and men (data available from 18 to 55 years); (3) subjects who dieted in the past or dieted (i.e. restrict food to lose weight) during the study; (4) any study that mentioned weight changes (or the terms ‘weigh maintenance’ or ‘weight (re)gain’) related to diet and WL, which included (i) dietary interventional studies with a follow-up after WL, (ii) longitudinal observational studies (over a few months to several years) with weight changes in dieters, and (iii) cross-sectional observational studies that compared BW and/or body composition between dieters and non-dieters; and (5) publications in English or French language. Studies involving animal models, athletes or trained individuals were not included nor were interventional protocols involving an imposed overfeeding (i.e. where subjects must ingest daily food intake higher than their daily requirement) before and/or after WL.
Data extraction and synthesis
Search results were exported to a spreadsheet, and duplicates were removed. The first selection was performed based on titles only and then based on abstracts. Finally, remaining full texts were screened to assess eligibility according to the inclusion and exclusion criteria above. The selection process was performed independently by two reviewers, and any discrepancies were collectively discussed amongst these reviewers until a consensus was reached. The third reviewer was consulted when necessary. The flow diagram presented in Fig. 1 illustrates this selection process. Then, the qualitative data were extracted using a spreadsheet containing the following items chosen collectively by the authors: (1) reference; (2) population characteristics (‘n’ sample size, age, BMI, unique sample or specific sample analysed in others studies); (3) design of the study (randomised and/or controlled or not, longitudinal or cross-sectional, interventional or observational); (4) total duration of the study, number of measurement point and duration between each measurement point for longitudinal studies; (5) details of the intervention protocol for the interventional studies; (6) type of measurements or parameters assessed with details on the methods used; (7) details on subjects’ food intake and/or diet; (8) main results for BW, weight changes and/or body composition; and (9) additional results linked to dietary or weight changes. When results were available for several groups, only those of control groups (called ‘non-dieters’) and the groups that matched the inclusion criteria were extracted. In addition, when BW and/or body composition data were not detailed in the paper, authors were contacted by email to obtain raw data and/or more information. When the extracted results were expressed as mean ± sem, sem was converted into sd. The methods and main results of the selected interventional studies are displayed in Table 1. Table 2 details the methodological characteristics of the selected observational studies, while Table 3 presents its detailed main results.
Flow diagram of the identification, screening, eligibility, and inclusion of the studies in the systematic review.
Characteristics of the populations, description of the protocols, measurements and parameters, and main results linked to body weight changes of the interventional longitudinal studies selected in the systematic review
m, males; NR, not reported; WL, weight loss; CHO, carbohydrates; WG, weight gain; EI, energy intake; BW, body weight; FM, fat mass; FFM, fat-free mass; REE, resting energy expenditure; TEE, total energy expenditure; f, females; T3, triiodothyronine; T4, thyroxine; NAS, no apparent statistical between pre-WL and WL or WG measures; NS, no significant; NW, normal-weight group; OW: overweight group; PA, physical activity; Italic, values collected from graphs; Δ, change from pre-WL measure.
Results are expressed in mean and standard deviation.
* P < 0·05,
** P < 0·01,
*** P < 0·001, significantly different from pre-WL measure.
† Significantly different from pre-WL but P-value not reported.
‡ Statistical unclear.
§ Statistical performed from raw data (repeated-measures ANOVA).
Characteristics of the populations, time points assessments, measurements and parameters assessed in the observational studies
FM, fat mass; HF, high-body fat group; f, females; LF, low-body fat group; BW, body weight; FFM, fat-free mass; DXA, dual-energy X-ray absorptiometry; EAT-26, 26-item Eating Attitudes Test; REE, resting energy expenditure; PAEE, physical activity energy expenditure; TEE, total energy expenditure; FFA, fat-free acids; OGTT, oral glucose tolerance test; HOMA, homeostatic model assessment of insulin resistance; NR, not reported; m, males; NAS, no apparent statistical; WL, weight loss; EDI, Eating Disorder Inventory; TFEQ, Three-Factors Eating Questionnaire; DEBQ, Dutch Eating Behavior Questionnaire; WG, weight gain; DSM-IV, Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition; WC, weight-cyclers; NC, non-cyclers.
Results are expressed in mean and standard deviation.
* P < 0·05,
** P < 0·001, dieters v. non-dieters or HF v. LF or WC v. NC.
† No information about the nature of data (sd or sem).
Main results of the observational studies selected in the systematic review, classified as three subgroups: results relative to dietary habits and weight loss (food intake, dietary and WL), main results on body weight and body composition, and additional results in link with body weight changes and/or diet
WL, weight loss; BW, body weight; HF, high body-fat group; LF, low-body-fat group; REE, resting energy expenditure; FM, fat mass; EAT-26, 26-item Eating Attitudes Test; TEE, total energy expenditure; OGTT, oral glucose tolerance test; ; FFA, fat free acids; NAS, no apparent statistical test; CHO, carbohydrates; WG, weight gain; m: male; f, female; OW, overweight; PA, physical activity; NW, normal weight; WS, weight suppression; NR, not reported; TFEQ, Three-Factors Eating Questionnaire; DEBQ, Dutch Eating Behavior Questionnaire; IWL: intentional weight loss; PAEE, physical activity energy expenditure; WC, weight-cyclers; NC, non-cyclers.
Results are expressed in mean and standard deviation.
* P < 0·05.
** P < 0·01.
*** P < 0·001.
**** P < 0·1 (trend).
† Significant but P-value not reported, dieters v. non-dieters or HF v. LF or WC v. NC.
‡ Statistical performed from raw data (t tests).
Quality assessment
The quality assessment of the included studies was performed independently by two reviewers using the ‘Quality Assessment Tool for Quantitative Studies’ developed by the Effective Public Health Practice Project(Reference Thomas, Ciliska and Dobbins29). Six criteria were assessed: selection bias, study design, confounders, blinding, data collection method, and withdrawals dropouts. Each criterion was qualified as ‘strong’, ‘moderate’, or ‘weak’. The overall methodological quality was then rated as strong if there were no weak rating criteria, moderate if there was one weak rating, and weak if there were two or more weak ratings(Reference Armijo-Olivo, Stiles and Hagen30). Any discrepancies were collectively discussed among the two reviewers until a consensus was reached, and the third reviewer was consulted when necessary. The quality assessment described above was presented individually for each study and collectively represented using a summary bar plot (in %) for each criterion and the overall quality.
Meta-analysis procedure
Given that interventional studies presented a high heterogeneity in terms of WL protocol, a meta-analysis was performed from the quantitative data for the observational studies only. Publications were selected with more restrictive criteria, as follows: (i) comparison between groups considered like ‘dieters’ and ‘non-dieters’; (ii) data on BW in kg in dieter and non-dieter groups; (iii) data in women-only considering the little number of available data in NW men; and (iv) data were estimated from graphs, as accurately as possible, if numerical values were not available in tables or text.
Statistical analysis
Statistical analyses were performed with Stata software (version 15, StataCorp). The meta-analysis considered between and within study variability. To address the non-independence of data due to the study effect, random effects model was preferred to the usual statistical tests to assess the standardised mean difference and their 95 % CI. More precisely, Hedge’s g standardised mean difference was estimated to assess the BW difference between dieters and non-dieters. Means and standard deviation were compiled when available or were estimated when median and interquartile range were reported. When sd was not available, an estimation according to available sd, for other studies, was calculated. Hedge’s g is considered to be a conservative estimate, which is useful for studies with small sample sizes, and the results may be interpreted as reflecting a small (g = 0·2–0·5), medium (g = 0·5–0·8) or large effect (g > 0·8). Heterogeneity in the study results was assessed by forest plots and the I² statistic which is the most common metric for measuring the magnitude of between-study heterogeneity. I² values range between 0 and 100 % and are typically considered low for < 25 %, moderate for 25–50 %, and high for > 50 %(Reference Higgins, Thompson and Deeks31). Publication bias was assessed by funnel plot, CI and with the Egger’s regression test as a formal statistical test. Two-sided type I error was fixed at 5 %.
Results
Selection of the studies
The search yielded a total of 1320 publications identified from database searching and 167 records identified from additional sources, making a total of 1344 scientific articles after the removal of duplicates. Based on title and abstract screening, 1269 publications were excluded. A total of seventy-five records were fully read to assess the eligibility according to the inclusion and exclusion criteria. At this step of the selection process, fifty-seven articles were removed, as detailed in Fig. 1. Finally, a total of eighteen scientific papers remained in the qualitative analysis of the systematic review, published from 1990 to 2019.
Quality assessment
The quality of selection bias minimisation, study design, and blinding of participants and personnel was judged to be moderate for the majority of the studies. The quality of confounder reduction was considered as strong for 61 % (n 11) and weak for 39 % (n 7) of the studies. Data collection methods were judged strong for 56 % (n 10) of the articles, while withdrawals and dropout were considered as strong for 33 % (n 6). Finally, the global quality assessment was defined as strong for six studies(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15,Reference Lowe, Annunziato and Markowitz24,Reference Korkeila, Rissanen and Kaprio25,Reference van Strien, Herman and Verheijden32) , moderate for nine studies(Reference Pietiläinen, Saarni and Kaprio27,Reference Kärkkäinen, Mustelin and Raevuori33–Reference Siggaard, Raben and Astrup40) , and weak for three studies(Reference Colditz, Willett and Stampfer26,Reference Anastasiou, Yannakoulia and Pirogianni41,Reference Palascha, van Kleef and van Trijp42) . Individual quality assessment of the selected publications is detailed in Fig. 2.
Individual and collective quality assessment of the studies included in the systematic review.
Designs of the studies
Two main study designs were identified among the eighteen publications selected in the systematic review. Height studies(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15,Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37–Reference Siggaard, Raben and Astrup40) followed an interventional longitudinal protocol that involved a WL induced by diet from 3 weeks(Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37) to 24 weeks(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15) duration in NW subjects, followed by a period of free-living or imposed recovery diet during 14 d(Reference Kajioka, Tsuzuku and Shimokata38) to 40 weeks(Reference Siggaard, Raben and Astrup40) (Table 1). Only one(Reference Siggaard, Raben and Astrup40) of these studies included a control group without WL intervention, but results were thus not extracted as the mean BMI for this cohort was 27 kg/m² (overweight subjects). Two studies(Reference Roberts, Fuss and Heyman34,Reference Moriguti, Das and Saltzman39) compared healthy young NW subjects with overweight and/or old ones (results were not extracted for these groups). Three studies(Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37,Reference Moriguti, Das and Saltzman39) imposed an individualised energy deficit of about 3 to 4 MJ/d (deficit of 800–1000 kcal/d). In Siggaard et al.’s study(Reference Siggaard, Raben and Astrup40), subjects received guidelines to set up an ad libitum high-carbohydrate and low-fat diet guarantying a minimum of 7 to 8 MJ/d (1670–1910 kcal/d). Fat was restricted to 20–25 % of daily energy intake, but subjects presented also an 11 % decrease in total energy intake(Reference Siggaard, Raben and Astrup40). Finally, the other studies imposed a restricted standard daily food intake of 6·6 MJ/d (1570 kcal/d) in male subjects(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15) and below 5·0 MJ/d (1200 kcal/d) in female subjects(Reference Kajioka, Tsuzuku and Shimokata38). During the first 12-week recovery period (prior to ad libitum access to food), the Minnesota cohort was divided into four groups with an imposed recovery diet whose energy values varied from 9·95 MJ/d (2377 kcal/d) to 14·19 MJ/d (3389 kcal/d)(Reference Dulloo, Jacquet and Girardier14). In the other studies, diet interventions were immediately followed by a period of ad libitum energy intake. One study repeated, in the same participants, two 30-d energy-restricted dietary periods, separated by a free-living period of 14 d(Reference Kajioka, Tsuzuku and Shimokata38). The ten remaining publications of the systematic review were observational studies, including two cross-sectional(Reference Anastasiou, Yannakoulia and Pirogianni41,Reference Palascha, van Kleef and van Trijp42) , and eight longitudinal studies that examined weight changes in the long term (8 months to 15 years) in participants representing the overall general population, with two to five measurement time points(Reference Lowe, Annunziato and Markowitz24–Reference Pietiläinen, Saarni and Kaprio27,Reference van Strien, Herman and Verheijden32,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35,Reference Van Wye, Dubin and Blair36) (Table 2).
Characteristics of the participants
All the studies were performed in NW healthy individuals with a mean BMI between 19·5(Reference Anastasiou, Yannakoulia and Pirogianni41) and 24·3 kg/m²(Reference Korkeila, Rissanen and Kaprio25). Among the eight interventional studies, three presented analyses using the same young male population (Minnesota study cohort, n 32, 25 ± 4 years)(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15). Three other studies enrolled only (or near, n 17 males and n 1 female for one study(Reference Siggaard, Raben and Astrup40)) males(Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37,Reference Siggaard, Raben and Astrup40) , one study was performed in females(Reference Kajioka, Tsuzuku and Shimokata38), and the last interventional study was performed in both male and female subjects(Reference Moriguti, Das and Saltzman39). The majority of the observational studies involved the sampling from the general population, aged 16 years (but results extracted from 18 years)(Reference Pietiläinen, Saarni and Kaprio27) to 54 years(Reference Colditz, Willett and Stampfer26) at baseline, including large cohort studies such as the Finnish Twin cohort(Reference Korkeila, Rissanen and Kaprio25,Reference Pietiläinen, Saarni and Kaprio27,Reference Kärkkäinen, Mustelin and Raevuori33) and the Nurses’ Health Study cohort(Reference Colditz, Willett and Stampfer26). Three observational studies were performed in women only(Reference Lowe, Annunziato and Markowitz24,Reference Colditz, Willett and Stampfer26,Reference Anastasiou, Yannakoulia and Pirogianni41) , and only female participants were selected in one study since the males were overweight at baseline (BMI ≥ 25 kg/m²)(Reference Van Wye, Dubin and Blair36). In addition, one study selected two specific groups of women in function of their FM: the ‘high-body-fat’ group having a FM > 30 % and the ‘low-body-fat’ group having a FM < 30 %(Reference Anastasiou, Yannakoulia and Pirogianni41). The remaining observational studies compared groups according to their dieting frequency(Reference Lowe, Annunziato and Markowitz24,Reference Korkeila, Rissanen and Kaprio25,Reference Sares-Jäske, Knekt and Männistö35) or their weight changes(Reference Colditz, Willett and Stampfer26,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Van Wye, Dubin and Blair36) , resulting in groups of ‘dieters’ and ‘non-dieters’, or ‘weight-cyclers’ and ‘non-cyclers’. For the study from van Strien et al. (2014)(Reference van Strien, Herman and Verheijden32), NW dieter and non-dieter groups were identified based on the raw data that were available. Participants were defined as ‘dieters’ when they self-reported dieting ‘sometimes’, ‘often’, or ‘always’, and as ‘non-dieters’ when they reported having ‘never’ or ‘rarely’ done. Due to the low number of participants and the dropout rate during the long-term follow-up in the dieter groups (50 % of dropouts in males and 38 % in females), we extracted data at T0 only (Table 2).
Parameters assessed and methodological approaches
BW and body composition were assessed in all the interventional studies before WL (‘pre-WL’), immediately after the WL phase (‘WL’), and after a recovery diet period, termed ‘weight (re)gain’ (‘WG’) (except one study which assessed body composition only at pre-WL and WL(Reference Heyman, Young and Fuss37)). FM and FFM were measured by hydrodensitometry(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15,Reference Moriguti, Das and Saltzman39) , plethysmography(Reference Kajioka, Tsuzuku and Shimokata38), doubly labelled water(Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37) , or bioimpedance(Reference Siggaard, Raben and Astrup40). Resting energy expenditure (REE) was evaluated by indirect calorimetry in six of the eight interventional studies(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15,Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37,Reference Kajioka, Tsuzuku and Shimokata38) . Three studies analysed the ad libitum food intake during the recovery period(Reference Dulloo and Jacquet13,Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37) . In addition, one interventional and two observational studies performed biochemical analyses (e.g. total cholesterol, insulin, and thyroid hormones)(Reference Sares-Jäske, Knekt and Männistö35,Reference Kajioka, Tsuzuku and Shimokata38,Reference Anastasiou, Yannakoulia and Pirogianni41) (Tables 1 and 2). Over half of the observational studies essentially used questionnaire measurements, with self-reported BW, height, and weight fluctuations(Reference Korkeila, Rissanen and Kaprio25–Reference Pietiläinen, Saarni and Kaprio27,Reference van Strien, Herman and Verheijden32,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Palascha, van Kleef and van Trijp42) . In addition, body composition was assessed in only one study by dual-energy X-ray absorptiometry(Reference Anastasiou, Yannakoulia and Pirogianni41), and waist circumference was obtained in three studies(Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35,Reference Anastasiou, Yannakoulia and Pirogianni41) . Dieting status and food intake were measured by different techniques: 24-h or 1-week self-reported food record(Reference Colditz, Willett and Stampfer26,Reference Anastasiou, Yannakoulia and Pirogianni41) , 25-item Weight Cycling Questionnaire(Reference Anastasiou, Yannakoulia and Pirogianni41), FFQ(Reference Colditz, Willett and Stampfer26,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35) , Restraint Scale(Reference Lowe, Annunziato and Markowitz24,Reference van Strien, Herman and Verheijden32) , or self-reported dichotomous questions/no validated questionnaires(Reference Korkeila, Rissanen and Kaprio25,Reference Pietiläinen, Saarni and Kaprio27,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35,Reference Van Wye, Dubin and Blair36,Reference Palascha, van Kleef and van Trijp42) . In longitudinal studies, dieting status was evaluated once during the follow-up period, essentially at baseline (T0)(Reference Lowe, Annunziato and Markowitz24,Reference Korkeila, Rissanen and Kaprio25,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35,Reference Van Wye, Dubin and Blair36) , but also at the end of study(Reference Pietiläinen, Saarni and Kaprio27) (Table 2). Van Strien and collaborators(Reference van Strien, Herman and Verheijden32) measured it at each time point (but data were only extracted at T0 as explained above). One study did not assess the dieting status of subjects but measured food intake in the middle of the longitudinal follow-up(Reference Colditz, Willett and Stampfer26). Finally, several observational studies conducted energy expenditure assessments, including total energy expenditure and REE in one study(Reference Anastasiou, Yannakoulia and Pirogianni41), and self-reported physical activity energy expenditure in six studies(Reference Korkeila, Rissanen and Kaprio25,Reference Pietiläinen, Saarni and Kaprio27,Reference van Strien, Herman and Verheijden32,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35,Reference Anastasiou, Yannakoulia and Pirogianni41) . Finally, aerobic capacity was evaluated in two observational studies by a maximal or submaximal cardiorespiratory test(Reference Van Wye, Dubin and Blair36,Reference Anastasiou, Yannakoulia and Pirogianni41) .
Main results from the interventional studies systematically reviewed
WL interventions resulted in a significant decrease of BW in all studies included(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15,Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37–Reference Siggaard, Raben and Astrup40) (Table 1). FM (% and/or kg) and FFM (kg) significantly decreased (P < 0·001) in two cohorts after 30 d, 12 weeks, and 24 weeks of semi-starvation(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15,Reference Kajioka, Tsuzuku and Shimokata38) , with a reduction of REE also observed(Reference Dulloo and Jacquet13,Reference Kajioka, Tsuzuku and Shimokata38) . In the publication of Heyman et al. (1992), a decrease in absolute and relative (to FFM) REE was highlighted during and after WL (P < 0·01)(Reference Heyman, Young and Fuss37). In the analyses of Dulloo and collaborators, most of the subjects of the Minnesota study did not regain their initial BW nor FM (kg), and none had regained FFM, after 12 weeks of imposed recovery diet(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15). However, following a subsequent 8 weeks of ad libitum energy intake, BW and FM (kg and %), but not FFM, were increased to higher values than those at baseline(Reference Dulloo, Jacquet and Girardier14). In one study(Reference Siggaard, Raben and Astrup40), BW was not recovered after 40 weeks of a free-living period, but participants were encouraged to maintain the low-fat diet during this period. In three other studies, BW was restored as early as day 10 of ad libitum food intake(Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37,Reference Kajioka, Tsuzuku and Shimokata38) , the regain being showed more rapid than the WL (+257 ± 120 g/d v. –97 ± 79 g/d, P < 0·01) in one publication(Reference Heyman, Young and Fuss37). In Moriguti et al.’s cohort, BW was restored after 6 weeks, and NW subjects regained significantly more than the group of overweight subjects(Reference Moriguti, Das and Saltzman39). In Kajioka’s study, after the first recovery period (14 d), BW overshot baseline value while FM did not change (Table 1). Yet, after the second dieting phase and 106 d of free-living, BW and FM were similar to pre-WL values(Reference Kajioka, Tsuzuku and Shimokata38). In addition, hyperphagia was observed in three studies relative to pre-WL energy intake(Reference Dulloo and Jacquet13,Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37) . Finally, after several weeks of recovery diet (12 to 15 weeks), REE remained at lower values than at baseline(Reference Dulloo and Jacquet13,Reference Kajioka, Tsuzuku and Shimokata38) . Results of biochemical parameters assessed in one study are detailed in Table 1. Finally, in Kajioka’s study, blood pressure (both systolic and diastolic) was significantly reduced after the second WL phase but was significantly elevated compared with pre-WL after 15 weeks of free-living conditions(Reference Kajioka, Tsuzuku and Shimokata38).
Main results from the observational studies systematically reviewed and meta-analysed
Studies conducted in Greece and the USA observed that high-body-fat and weight-cycler groups had significantly higher BW and BMI and dieted more often compared with low-body-fat(Reference Anastasiou, Yannakoulia and Pirogianni41) and non-cycler(Reference Van Wye, Dubin and Blair36) groups, respectively. In another such study conducted in Finland, BMI at baseline was higher (P < 0·001) in male and female dieters compared with non-dieters(Reference Korkeila, Rissanen and Kaprio25), while a further study showed no difference in BMI and BW between these groups(Reference van Strien, Herman and Verheijden32) (Table 3). Forty per cent of the studies reviewed observed a significant increase, or increasing trend, in BMI or BW in NW individuals engaging in diet/WL compared with those who did not diet and/or lose weight(Reference Korkeila, Rissanen and Kaprio25,Reference van Strien, Herman and Verheijden32,Reference Sares-Jäske, Knekt and Männistö35,Reference Van Wye, Dubin and Blair36) , while two others studies showed a positive association between WG and dieting(Reference Kärkkäinen, Mustelin and Raevuori33,Reference Palascha, van Kleef and van Trijp42) . Moreover, major WG was associated with a lower initial BW or BMI(Reference Korkeila, Rissanen and Kaprio25,Reference Pietiläinen, Saarni and Kaprio27,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35) , irregular eating(Reference Kärkkäinen, Mustelin and Raevuori33), dieting(Reference Lowe, Annunziato and Markowitz24,Reference Korkeila, Rissanen and Kaprio25,Reference Pietiläinen, Saarni and Kaprio27,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Palascha, van Kleef and van Trijp42) , smoking(Reference Kärkkäinen, Mustelin and Raevuori33), and restrained eating(Reference Palascha, van Kleef and van Trijp42). Finally, five studies were selected for inclusion in the meta-analysis according to our selection criteria(Reference Lowe, Annunziato and Markowitz24,Reference Colditz, Willett and Stampfer26,Reference van Strien, Herman and Verheijden32,Reference Van Wye, Dubin and Blair36,Reference Anastasiou, Yannakoulia and Pirogianni41) . In the study of Anastasiou and collaborators, the high-body-fat group was considered analogous to ‘dieters’ and low-body-fat group to ‘non-dieters’, given that the high-body-fat group had a higher number of WL achieved per year and a higher total energy expenditure, while it had a reduced daily food intake and a higher score for the 26-item Eating Attitudes Test dieting subscale compared with the low-body-fat group(Reference Anastasiou, Yannakoulia and Pirogianni41). For another study(Reference Colditz, Willett and Stampfer26), we defined the ‘dieters’ as those who consumed less than 5·0 MJ/d (1205 kcal/d) (similar value to a semi-starvation), while ‘non-dieters’ were those who ingested more than 7·7 MJ/d (1844 kcal/d). Overall, the result of the meta-analysis showed that dieters had, on average, a higher BW compared with non-dieters (standardised mean difference: 0·512 [95 % CI: 0·069, 0·954]; P = 0·023), as illustrated in Fig. 3(a). The standard differences in means ranged from –0·086 to 1·838 (n 5), and there was a high heterogeneity between studies (I² = 87·6 %; Q = 40·42; df = 5; P < 0·001). The results of the Egger’s test highlighted a publication bias (2·813 [95 % CI: 1·192, 4·434]; P = 0·009), and the funnel plot is illustrated in Fig. 3(b).
Body weight comparison between dieters and non-dieters. (a) Forest plot of the meta-analysis. Data were obtained from observational studies and are expressed in standardised mean difference and their 95 %CI (lower and upper limits). (b) Funnel plot from the Egger’s test. SMD, standardised mean difference.
Discussion
While rising rates of overweight and obesity have led to an increasing pre-occupation around weight control and WL in the recent years(Reference Santos, Sniehotta and Marques43), dieting also became quite common among NW individuals(Reference Montani, Schutz and Dulloo4–Reference Julia, Péneau and Andreeva6,Reference de Ridder, Adriaanse and Evers44) . Social pressure relative to thinness, particularly among women, can drive them to lose weight, despite the absence of medical necessity contrary to people suffering from overweight or obesity(Reference Ordaz, Schaefer and Choquette7–Reference Mikolajczyk, Maxwell and El Ansari9). Although frequent, the effects of WL in healthy NW individuals remain poorly investigated. While a sustainable energetic restriction seemed to show some beneficial effects in NW populations(Reference Fontana, Meyer and Klein18–Reference Willcox, Willcox and Todoriki20), the high availability of fatty, sweet and ultra-processed food in the occidental diet leads to overconsumption and it became difficult to maintain low-energy intake on the long term(Reference Greenway21–Reference Hall and Kahan23). Consequently, the probability of weight regain after a WL induced by diet is high, and few studies and reviews suggested negative impacts of dieting, in particular an increased risk factor for later obesity(Reference Montani, Schutz and Dulloo4,Reference Dulloo, Jacquet and Montani12,Reference Pietiläinen, Saarni and Kaprio27) . In this context, the present work undertook a systematic analysis of the available studies questioning diet-induced WL and subsequent or long-term BW changes in healthy NW individuals. After a detailed and strict selection process, eighteen publications met the inclusion criteria for review. Importantly, our systematic approach clearly highlighted a high level of methodological heterogeneity between studies, which was certainly the first and maybe main result of the present work. Indeed, while eight of the included papers were interventional, ten were observational; six of the eight interventional studies being conducted among males only (or near, 94·4 % of men in one study(Reference Siggaard, Raben and Astrup40))(Reference Dulloo and Jacquet13–Reference Dulloo, Jacquet and Girardier15,Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37,Reference Siggaard, Raben and Astrup40) , and four of the ten observational studies involved females only(Reference Lowe, Annunziato and Markowitz24,Reference Colditz, Willett and Stampfer26,Reference Van Wye, Dubin and Blair36,Reference Anastasiou, Yannakoulia and Pirogianni41) . As detailed in Table 1, the duration, the nature of diets (e.g. proportion of macronutrients and magnitude of energy deficit), and even the methods used to assess WL itself were highly heterogeneous, leading to a complex comparison of study results.
Observational studies
In observational studies, the dieting status of the participants was assessed using different self-reported methods, which considerably limits the reliability of results. In addition, information about the practice of dieting or food intake was available at only one time point for 88 % of the longitudinal studies, and thus there was no indication of the persistence or interruption of the diet during the follow-up (Table 2). Importantly, the term ‘diet’ was poorly defined in all the observational studies, and the characteristics to determine the dieting status of subjects were various. While we defined diet as restricting energy intake or certain kinds of food in order to lose weight, dieting can be also described as maintaining, gaining or losing weight by the regulation of food intake(Reference Montani, Schutz and Dulloo4,Reference de Ridder, Adriaanse and Evers44) . Therefore, if ‘dieting’ includes the control of food intake to maintain a stable weight, it appears crucial to ‘diet’ in an obesogenic environment to prevent WG, without any WL induced by restricted food. Regarding the lack of consensus on the definition of dieting and the subjectivity and heterogeneity of methods to assess this status, the results have to be interpreted with caution. Finally, concerning the BW assessment, data were self-reported in 60 % of the included observational studies, with only two measuring body composition(Reference Van Wye, Dubin and Blair36,Reference Anastasiou, Yannakoulia and Pirogianni41) . While there was a large heterogeneity in these observational studies, all of them seem to show a negative effect of dieting on BW, body composition and/or long-term BW changes. Despite the absence of statistical analysis, Colditz and collaborators observed that individuals who previously lost more than 5 kg gained about fourtimes as much weight over an 8-year period than those with lower weight variations (≤ 3 kg)(Reference Colditz, Willett and Stampfer26). Moreover, 80 % of these observational studies showed a relationship between dieting and long-term weight change, with a major risk of WG related to diet(Reference Lowe, Annunziato and Markowitz24,Reference Korkeila, Rissanen and Kaprio25,Reference Pietiläinen, Saarni and Kaprio27,Reference van Strien, Herman and Verheijden32,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35,Reference Van Wye, Dubin and Blair36,Reference Palascha, van Kleef and van Trijp42) . Finally, three studies found a significantly higher BMI or BW in dieters than in non-dieters(Reference Korkeila, Rissanen and Kaprio25,Reference Van Wye, Dubin and Blair36,Reference Anastasiou, Yannakoulia and Pirogianni41) , that was confirmed by the results of our meta-analysis (Fig. 3). However, this result is questionable: is higher BW in dieters the cause or the consequence of dieting? Indeed, as suggested by Heberbrand & Hinney(Reference Hebebrand and Hinney45), initial BW and the tendency to easily gain weight are in part imputable to genetics as a risk factor for obesity. Consequently, people with higher initial BW, while still being within the NW range, are more likely to engage in diet compared with those with a lower BW, all the more if dieting can be defined as control of food intake to maintain weight or limit WG without necessary WL (in link with the definition of diet discussed above). Although our systematic and meta-analytic approaches both suggest that regular NW dieters might be at risk for later extra weight regain compared with non-dieters, it remains to be clearly determined (using a controlled interventional study design) whether WL per se is the cause of this subsequent higher BW. Importantly, Pietiläinen and collaborators(Reference Pietiläinen, Saarni and Kaprio27) questioned the role of genetics in long-term WG in twins. They suggested an important effect of genetic predisposition illustrating by a higher BW in dizygotic twins trying to lose weight, but they also highlighted an effect of dieting per se, with larger long-term WG in dieters compared with non-dieters in monozygotic twin pairs(Reference Pietiläinen, Saarni and Kaprio27). However, objective measures of WL and weight regain are lacking in these observational studies, and interventional studies are therefore necessary to better understand the consequences of dieting on later weight changes.
Interventional studies
The classic Minnesota study remains, to date, the only energy restriction study to show a higher BW and FM after recovering from 24 weeks of drastic (semi-starvation) WL relative to baseline in NW healthy subjects(Reference Dulloo, Jacquet and Girardier14). However, weight recovery after 8 weeks of energy deficit (and about 14 % WL) induced by intense physical activity in young men undergoing the US Army Ranger training has also been reported to be accompanied by higher BW and FM values than at baseline; reviewed in reference(Reference Dulloo, Jacquet and Girardier14). In their work on dieting and later WG, Kajioka et al. (Reference Kajioka, Tsuzuku and Shimokata38) only observed a lower FFM in NW individuals, without any change in BW or FM, after two repeated cycles of WL and weight regain. Importantly, both this study and the Minnesota study observed a lower REE after WL and weight regain compared with pre-WL(Reference Dulloo and Jacquet13,Reference Kajioka, Tsuzuku and Shimokata38) , highlighting an adaptive reduction of thermogenesis. This phenomenon is well documented in the literature in response to WL both in NW people(Reference Dulloo, Jacquet and Montani46–Reference Koehler, De Souza and Williams49) and people with obesity(Reference Dulloo, Jacquet and Montani46,Reference Dulloo and Schutz50–Reference Casanova, Beaulieu and Finlayson52) . While a decrease in FM and FFM could partly explain the decrease in REE, cellular adaptations are made by the organism to promote energy conservation. In fact, a reduction of REE relative to FFM was observed after WL, both in male(Reference Heyman, Young and Fuss37) and female(Reference Koehler, De Souza and Williams49) NW subjects. Although some studies reported that this ‘thrifty metabolism’ in REE seems to be attenuated or to disappear after a period of weight maintenance in individuals with obesity(Reference Leibel, Rosenbaum and Hirsch47,Reference Nunes, Casanova and Francisco51) , there is evidence that adaptive thermogenesis in the compartments of REE and/or in non-REE may persist for several months to a year during the weight-reduced state(Reference Rosenbaum, Hirsch and Gallagher53–Reference Fothergill, Guo and Howard57). Importantly, the suppression of adaptive thermogenesis has been shown to persist in NW individuals, with a reduction of 15 to 20 % in REE over 15 weeks after weight recovery(Reference Dulloo and Jacquet13,Reference Kajioka, Tsuzuku and Shimokata38) . In addition, Kajioka et al. (Reference Kajioka, Tsuzuku and Shimokata38) observed a decrease of triiodothyronine and thyroxine blood concentrations after WL intervention and recovery, both being implicated in energy metabolism and adaptive thermogenesis(Reference Mullur, Liu and Brent58). Semi-starvation and refeeding experimentations in rats have also highlighted cellular adaptations leading to an energy economy in skeletal muscles that persists during refeeding, including decreased muscle thyroid hormone levels(Reference Calonne, Isacco and Miles-Chan59).
Beyond compensatory adaptations in energy expenditure, an increase in energy intake above baseline has been observed during weight recovery in NW individuals, illustrating post-WL hyperphagia(Reference Dulloo, Jacquet and Girardier14,Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37) . This increase in energy intake was shown in NW people after acute energy deficits induced by diet(Reference Thivel, Metz and Julian60) and in subjects with overweight or obesity after dietary WL intervention(Reference Blundell, Gibbons and Caudwell61). Overall, hyperphagia and adaptive thermogenesis are compensatory adaptations that aim to resist WL during energy deficit and to facilitate weight regain. However, these modifications seem to persist after weight regain in NW individuals(Reference Dulloo, Jacquet and Girardier14,Reference Roberts, Fuss and Heyman34,Reference Heyman, Young and Fuss37) and could explain the increase in BW above baseline values during post-dieting recovery. Indeed, in the Minnesota study, weight recovery was driven by persistent slowed metabolic rate and concomitant hyperphagia, resulting in more rapid restoration of FM than FFM. The hyperphagic state persisted beyond complete recovery of FM, until FFM has returned to baseline values(Reference Dulloo, Jacquet and Girardier14,Reference Dulloo, Jacquet and Girardier15,Reference Keys, Brožek and Henschel62) , thereby resulting in an increase in FM beyond pre-WL values (i.e. fat overshooting)(Reference Dulloo, Jacquet and Girardier14,Reference Dulloo, Miles-Chan and Schutz16) , and consequently a higher BW than baseline.
Importance of body weight status
In the Minnesota study, initial FM (%) was negatively associated with individuals’ P-ratio, an index of body energy partitioning(Reference Dulloo, Jacquet and Girardier15), suggesting that leaner subjects are more susceptible to mobilise energy from protein (FFM) during WL and to experience a FM rebound during weight regain(Reference Dulloo, Jacquet and Montani12,Reference Dulloo, Jacquet and Girardier14) . The recent narrative review of Magkos (2022) supports this major loss of lean tissue in response to diet in NW people compared with subjects with obesity(Reference Magkos17). In addition, Roberts et al. observed a negative correlation between initial FM and hyperphagia during recovery(Reference Roberts, Fuss and Heyman34), corroborating the hypothesis that leaner subjects were at higher risk of weight regain. Besides the dieting effect, several other studies of the systematic review highlighted the potential influence of weight status on BW evolution, whereby NW dieters are at higher risk for long-term WG than dieters suffering from overweight or obesity(Reference Korkeila, Rissanen and Kaprio25,Reference Pietiläinen, Saarni and Kaprio27,Reference Sares-Jäske, Knekt and Männistö35,Reference Moriguti, Das and Saltzman39) . According to the theory of fat overshooting, habitual dieters, and specifically NW individuals, could be subject to catch-up fat resulting in higher BW than pre-WL(Reference Dulloo, Jacquet and Montani12,Reference Dulloo, Jacquet and Girardier15) . Consequently, if NW dieters try again to lose weight by repeating WL and weight regains (i.e. weight cycling), the cumulative amount of fat overshoot over several weight cycles could ultimately result in obesity, as recently illustrated using a mathematical model of weight cycling that integrates the relationship between post-dieting fat overshooting with initial adiposity(Reference Jacquet, Schutz and Montani63). Independently of the gain in BW, weight cycling induced by diet seems to be associated with increased cardiometabolic risk, particularly in NW people(Reference Montani, Schutz and Dulloo4), and even in athletes despite their high physical activity levels(Reference Miles-Chan and Isacco64,Reference Lakicevic, Mani and Paoli65) . In that context, it appears important to prevent these potential risks associated with diet and weight cycling in initially healthy NW people.
Which role for physical activity?
Although physical activity and/or fitness levels were assessed in 70 % of the included observational studies(Reference Korkeila, Rissanen and Kaprio25,Reference Pietiläinen, Saarni and Kaprio27,Reference van Strien, Herman and Verheijden32,Reference Kärkkäinen, Mustelin and Raevuori33,Reference Sares-Jäske, Knekt and Männistö35,Reference Van Wye, Dubin and Blair36,Reference Anastasiou, Yannakoulia and Pirogianni41) , there was limited information about the relationship between WG, diet, and physical activity level. Nevertheless, physical activity has been suggested to limit the compensatory adaptations in response to energy deficits induced by diet in NW people(Reference Thivel, Metz and Julian60). Physical activity may therefore serve as a preventive strategy against fat overshooting thanks to the maintenance of FFM (or in minimising its loss) during WL, as demonstrated in people with obesity(Reference Catenacci, Grunwald and Ingebrigtsen66,Reference Verheggen, Maessen and Green67) and suggested in athletes(Reference Miles-Chan and Isacco64). However, this potential protector effect may depend on the degree of energy deficit and nature of physical activity performed. Indeed, while participants of the Minnesota study were asked to walk 35·4 km each week(Reference Keys, Brožek and Henschel62), they were also confronted with a semi-starvation diet, which might have surpassed the preventive effect of physical activity on FFM and other compensatory mechanisms.
Limitations
Several limitations must be considered in the present study. First, we use three important and relevant databases (i.e. PudMed-Medline, Embase, and Google Scholar) for our literature search strategy, but missing to include other databases, such as Web of Sciences or Scopus, might represent an important limitation of the present work that needs to be considered. Further, the systematic approach revealed an important heterogeneity between studies in terms of methodology, which somewhat restrict our analyses and conclusions. Then, in the observational studies, dieting status was not clearly defined by authors and not ascertained by objective measurements of WL and food intake in individuals reporting to diet, and our results must therefore be considered with precaution. Concerning the interventional studies, sample sizes remain modest and the majority relied on very-low energetic diets (semi-starvation). However, Julia and collaborators reported in the general population a high prevalence of consumption of non-individualised commercial diets (and low prevalence of diets prescribed by professionals), which are generally low- to very-low-energy diets(Reference Julia, Péneau and Andreeva6). Finally, the interventional studies were not controlled by a group of NW subjects without WL intervention. These limitations highlight the need to conduct studies to standardise methodological approaches with investigator-verified, objective measurements of BW and body composition throughout WL – WG cycles in NW individuals.
Conclusion
A significant proportion of NW individuals are engaged in dieting to control their weight, notably because of body image distortion and social pressure. The present systematic review seems to indicate that both interventional and observational studies, separately, suggest potential negative effects of dieting on BW and body composition in this population. Despite the observed high methodological heterogeneity and the significant publication bias, the meta-analysis of observational studies underlines a higher BW in dieters than in non-dieters. However, the results do not allow to confirm that higher BW and higher WG observed in dieters is the consequence or the origin of the practice of dieting. Similarly, the present systematic analysis of the included interventional studies suggests that negative energy balance induced by dieting generates compensatory adaptations, which may persist during refeeding and even after BW recovery in NW people, but leading to increased FM and BW relative to pre-WL only in one study. Importantly, the present systematic and meta-analytic approaches clearly add to the existing literature that the available evidences suffer from high methodological heterogeneity (both for interventional and observational studies), deeply limiting the formulation of strong conclusions and calling for further well-designed studies. While dieting could be a controversial risk factor for later obesity in NW individuals, there is a clear need to better understand the physiological adaptations occurring during WL and weight regain in this population via randomised controlled interventional studies.
Acknowledgements
This research received no specific grant from any funding agency, commercial or not-for-profit sectors.
D. T., L. I., J. M-C. and A. D.: Formulating the research question; D. T., L. I., Y. B. and M. D.: Designing the study; L. P., S. B., D. T. and L. I.: Carrying out the study; B. P., L. P. and S. B.: Analysing the data; L. P., L. I. and A.D.: Interpreting the findings; L. P., D. T. and J. M-C.: Writing the article. D. T., L. I., A. D., J. M-C., L. P. and S. B.: Conceptualisation; L. P., S. B. and L. I.: Data curation; A. D., L. I., D. T. and M. D.: Investigation; B. P.: Formal analysis; D. T., A. D., J. M-C., L. I., Y. B. and M. D.: Methodology and Project administration; L. P. and S. B.: Writing; D. T., L. I., J. M-C., A. D., B. P., Y. B. and M. D.: Review and Editing.
There are no conflicts of interest.
