To identify the proportion of weight lost as fat-free mass (FFM) by various weight loss interventions.
Medline and Embase were systematically searched for reliable measurements of FFM before and after weight loss of >10 kg and eligible data were pooled. In a fixed effect model of % FFM loss/weight loss (%FFML), linear regression analysis was used to determine the influence of degree of caloric restriction, exercise, magnitude of weight loss, initial body mass index (BMI) and type of surgery.
Data were included from 26 cohorts treated with dietary and behavioral interventions and 29 cohorts of bariatric surgery patients. The degree of caloric restriction was positively associated with %FFML (r 2=0.31, P=0.006) and in three randomized controlled trials exercise was shown to decrease %FFML. Compared with laparoscopic adjustable gastric banding (LAGB), biliopancreatic diversion (BPD) and roux en Y gastric bypass (RYGB) caused greater loge (natural log) %FFML (r 2=0.453, P<0.001). Differences in loge %FFML between surgical procedures were independent of initial BMI and magnitude of weight loss.
The degree of caloric restriction, exercise and rate of weight loss influence the proportion of weight lost as FFM after non-surgical interventions. For surgical interventions, BPD and RYGB result in greater %FFML than LAGB.
The global epidemic of obesity is rapidly becoming a health priority. Currently, close to 60% of the population of the US, the UK and Australia are classified by the World Health Organisation as overweight, and almost a third of these people suffer from obesity and related comorbidities.1, 2, 3 The health costs relating to obesity in the developed world are expected to exceed that of tobacco smoking in the near future,3 prompting much interest in the efficacy, sustainability and safety of weight loss interventions.
The aim of weight loss in obese subjects is loss of fat mass (FM), but inevitably a proportion of weight loss is fat free. Loss of fat-free mass (FFM) may be undesirable if excessive as non-adipose tissues are responsible for the majority of resting metabolic rate, regulation of core body temperature, preservation of skeletal integrity, and maintenance of function and quality of life as the body ages.4 Assessments of FFM loss as a proportion of total weight loss could provide an important measure of the safety of individual weight loss methods. However, the point at which proportional loss of FFM is excessive and potentially hazardous is unclear. If it is assumed that the body composition of an age, gender and ethnically matched never-obese person is desirable, then optimal fat:FFM loss ratio could be based on this. Webster et al.5 took this approach using data from 104 Caucasian women and concluded that FFM should be not more than 22% of total weight loss. This figure may vary with age, gender and ethnicity.6, 7 Thus, it is difficult to set theoretical standards for the percentage of weight lost as FFM (%FFML). As an alternative, examination of the degree to which different weight loss methods achieve preservation of FFM may provide realistic goals for interventions.
Pooled comparisons of weight loss methods are hampered by interventional heterogeneity, but generalizations can be made regarding some fundamental components. Specifically, the balance between the degree of caloric restriction and energy expenditure varies between weight loss interventions, potentially affecting %FFML.8, 9 Bariatric surgery, which provides the greatest sustained weight loss, may arguably be of greatest concern. In addition to significant caloric restriction, bariatric surgical methods can cause malabsorbtion, malnutrition and changes to gastrointestinal hormone levels. In particular, altered levels of the growth hormone secretagogue, ghrelin10, 11 may influence %FFML.
We hypothesized that for weight reduction interventions the degree of caloric restriction, exercise, type of bariatric surgical procedure and magnitude of weight loss may be predictive of the proportion of weight loss consisting of FFM. We reviewed dietary, behavioral and pharmaceutical interventions including low calorie diets (LCD), very low calorie diets (VLCD), LCD+exercise, VLCD+exercise, LCD+orlistat and LCD+sibutramine for their effects on %FFML. In addition, we reviewed %FFML for the three most commonly performed surgical interventions; biliopancreatic diversion (BPD), roux en Y gastric bypass (RYGB) and laparoscopic adjustable gastric banding (LAGB).
Medline (1966 – April 2006) and Embase (1966 – April 2006) were searched using a formal search strategy for studies that assess relative changes in FM and FFM during substantial weight loss. The search strategy is shown in Table 1.
There are a number of methods for measuring changes in FFM and FM that are currently in use, but many of these lack adequate validation. Only studies employing methods that have shown a high level of agreement with under water weighing (UWW) were included. These methods included dual energy X-ray absorptiometry (DEXA), deuterium-oxide dilutometry for total body water (TBW) and air displacement plethysmography (ADP). All have been shown to agree with UWW to within 5%.12, 13, 14, 15, 16
The method of bioelectrical impedance (BIA) has received much attention as a convenient alternative to the more cumbersome methods listed above. However, BIA has not been validated in obese subjects or with weight loss and studies using it were excluded from the analysis.17, 18
Estimates of total body potassium (TBK), although indicative of the aqueous volume of lean tissue, were also excluded owing to concern about assumptions of constant TBK/FFM ratio.19 Finally, although multislice magnetic resonance imaging (MRI) is sensitive to changes in tissue volumes20 and FFM,21 it is incompatible with techniques that measure whole body fat and fat free as two compartments. Therefore, studies using MRI were excluded from the pooled analysis owing to the lack of direct quantitative comparison with UWW, DEXA or TBW. Such studies were considered separately.
All abstracts from the final search set were screened and studies that fulfilled the following criteria were retrieved.
Studies of children, adolescents and animals were excluded. The reference lists of each retrieved publication were checked for any relevant studies not electronically indexed, and manual searches of Obesity Research, International Journal of Obesity and Obesity Surgery were conducted covering the years 1990–2005.
Analysis of observational and case–control studies
The heterogeneity of weight loss methods and lack of randomized controlled trials (RCTs) limited statistical analysis. Interventions could be placed broadly into groups and some general comparisons between the groups could be assessed. Surgical and medical weight loss interventions could not generally be compared because the range of initial body mass index (BMI) (28–38 for medical weight loss methods and 40–60 for surgical weight loss methods), time to follow-up, and the magnitude of weight loss were substantially different.
Changes in weight and FFM were recorded from each study and the percentage of weight loss consisting of FFM (FFM loss × 100/weight loss=%FFML) was calculated. Initial BMI, time to follow-up (weeks), number of patients, gender, methods of FM determination and details of the intervention were recorded. In the absence of multiple RCTs comparing weight loss interventions, statistical comparisons of %FFML in controlled and observational studies were made using a fixed effect model without weighting by the number of patients (n).
Where more than two cohort time points were available for a particular intervention, the descriptive statistics are shown as median and inter-quartile ranges (IQR). Differences between interventions were examined using the Mann–Whitney U-test. Where data were normally distributed, the mean and s.d. was calculated and the Student's t-test was used to make comparisons of pooled means.
The %FFML in the pooled studies of dietary and behavioral weight loss interventions was normally distributed. Linear regression was used to assess the degree of caloric restriction, exercise, weight loss and gender as predictors of the percentage of weight lost as FFM. For the surgical studies, there was a right skew and natural log (loge) transformed data were used for analysis. Linear regression analysis was used to determine whether the type of surgery (LAGB or BPD), the degree of weight loss, initial BMI and gender were predictive of the loge of the percentage of weight loss consisting of FFM.
The mean %FFML was calculated for all male subjects and all female subjects on dietary and behavioral weight loss interventions. These values were used as cutoffs to compare groups of study cohorts divided by categories of research interest, for example, LCDs compared with VLCDs. Where studies reported a mean of male subjects and female subjects, the cutoff was adjusted in proportion to the ratio of female subjects to male subjects in the study. χ2 analysis was used to look for significant differences in the number of cohorts with above average %FFML.
Analysis of randomized studies
Very few RCTs that compare interventions in terms of %FFML were found (Tables 2 and 4). Three RCTs were retrieved showing the effects of aerobic exercise and resistance exercise39, 40, 41 on the composition of weight loss achieved using LCDs. Pooled means were calculable from four randomized trials with three treatment arms that used multislice MRI to measure changes in FFM with weight loss (Table 3).
A total of 958 citations from Medline and 1653 from Embase were screened. From these, 16 studies of medical weight loss interventions with 26 cohort time points, and 17 studies of surgical weight loss interventions with 29 cohort time points fulfilled the selection criteria (Tables 2 and 4). Three RCTs were retrieved that used multislice MRI to measure changes in FFM. These studies were omitted from the pooled analysis and analyzed separately.
Dietary and behavioral weight loss interventions
Median (IQR) %FFML for LCDs, VLCDs and VLCDs with exercise was 14.0 (10), 23.4 (8) and 22.5 (11)% of weight loss, respectively. There were only two studies of LCD+exercise, precluding calculation of median and IQR. Using the mean %FFML for each cohort (n=20), linear regression analysis was used to look for factors associated with greater loss of FFM. The %FFML was greater when VLCDs were used (r2=0.31; P=0.006). No additional variance was explained by reported exercise, gender, initial BMI or the magnitude of weight loss (Figure 1). There was a tendency for higher mean %FFML in cohorts of men (27±7%) when compared with women (20±8%, P=0.08).
RCT evidence regarding exercise
Three randomized studies from the one research group consistently show that during 16 weeks of weight loss on a LCD, supervised exercise reduced %FFML.39, 40, 41 Using LCDs that restricted caloric intake by 1000 kcal/day, 27.8±6.4% of weight loss was fat free in 42 subjects (Table 3). This was significantly reduced to 13±4.1% in 41 subjects who engaged in aerobic exercise to between 50 and 85% of maximal heart rate for 15 to 60 min 5 days/week. For the 44 subjects who completed 30 min (3 days/week) resistance training designed to increase strength by 30–45%, the proportion of weight loss that was fat free was 16.6±3.7%.39, 40, 41 Aerobic exercise provided better retention of FFM, but all groups with an exercise program as part of the intervention achieved <22% of %FFML. These four RCTs also show a significant effect of gender, with women losing less %FFML than men in all treatment arms (P<0.001) (Table 3).
Only two studies of sibutramine+LCD met the inclusion criteria and both indicated potentially problematic %FFML of 31 and 38% (Table 2).33, 34 However, neither of these studies used a placebo control, precluding conclusive assessment of sibutramine. Only one study of orlistat+LCD met the criteria for inclusion in this review. In this placebo controlled RCT, orlistat caused a marginal increase in weight loss, but made no change to the proportion of weight loss consisting of FFM compared to a LCD alone (Table 2).25
Surgically induced weight loss
When compared to LAGB, both BPD (P<0.001) and RYGB (P=0.002) caused greater %FFML. The median (IQR) %FFML was 25.6 (11), 31.3 (12.2) and 17.5 (3.7)% following BPD, RYGB and LAGB, respectively.
Linear regression analysis revealed that diversionary surgery resulted in greater loge %FFML than restrictive LAGB surgery (r2=0.453, P<0.001). This difference was independent of initial BMI and the magnitude of weight loss. Using Spearmans non-parametric correlation, a trend appears between the loge %FFML after BPD and the magnitude of weight loss (r=0.53, P=0.077; Figure 2). No such relationship was evident in the pooled studies of LAGB and was not calculable among studies of RYGB (Figure 2).
Frequencies of %FFML exceeding the gender weighted average
The pooled mean %FFML was 27% for men and 20% for women using dietary and behavioral weight loss interventions. Where studies reported a mean of both female subjects and male subjects the average %FFML was weighted according to the ratio of female subjects to male subjects in the study. The number of cohorts for each intervention for whom %FFML was below or above the gender weighted average was recorded (Table 5). All cohorts using VLCDs with no specified exercise experienced above average %FFML, whereas only 8% of LCD cohorts had above average %FFML (Table 5). Non-exercising cohorts on VLCDs were more likely to report above average %FFML when compared with LCDs (P<0.001). Both BPD and RYGB were more likely to result in above average loss of FFM when compared to LAGB (P<0.001) (Table 5).
In this review, all identified studies in which a reliable two compartment model of body composition was applied before and after weight loss >10 kg have been pooled. Despite the relevance of FFM to obesity research, body composition appears in only a few studies of weight loss and the majority of these employ only two compartment models that lack the accuracy of three and four compartment models. Moreover, the quality of the available data is limited by the lack of RCTs and the heterogeneity of interventions and their outcomes. It is also notable that the same cohorts measured at several time points and the limited number of centers measuring body composition may undermine generalizations and reduce broad applicability of the findings.
In this systematic review, we have identified two important deficiencies in the literature. Firstly, there is little quality data concerning pharmaceuticals for weight loss, despite the lack of interventions that are suitable for long term weight management and the potential for nutritional concerns that may lead to excessive loss of FFM.8, 59 Only one study of 16 subjects treated with orlistat met the inclusion criteria.25 The effects of sibutramine on FFM loss are also understudied, with only two studies including only 27 subjects meeting the inclusion criteria.34, 33 It is of note that the %FFML in both these small studies was higher than could be expected for the associated LCD, raising significant concern about the components of weight loss with this medication. Although these two studies lacked placebo controls, they may allude to biological effects of synthetically elevated resting metabolic rate induced by sibutramine.
Secondly, although RYGB is the most common bariatric surgical intervention, especially in the US,60 reliable measurements of FFM loss in only two studies with a total of 49 subjects could be found.51, 50 Given the potential for adverse consequences of this intervention,61 it is very poorly studied. Although only a small proportion of published results of BPD and LAGB include FFM loss as an outcome, the data concerning these interventions are much more robust than for RYGB.
Comparison of LCDs and VLCDs gives clear evidence that the degree of caloric restriction affects %FFML. Assuming that the difference between these diets is purely the degree of caloric restriction, the increased initial rate of weight loss achieved using VLCDs compared to LCDs may be the cause of greater FFM loss on these diets, at least in the short term (<17 weeks). VLCDs provide quite rapid weight loss and substantial loss of FFM, but ultimately the longer-term body composition in the post weight loss, weight stable state may be more important. Studies of the longer-term body composition effects of weight loss induced by VLCDs are clearly needed.
The only credible evidence that exercise preserves lean tissue during significant weight loss came from three RCTs conducted by Janssen et al.39, 40 and Rice et al.41 (Table 5). Singly and cumulatively these studies show significant decreases in the loss of FFM attributable to rigorously supervised resistance and aerobic exercise regimens (Table 4).40, 41 It is possible that many of the dietary studies in this review did not report concurrent exercise interventions or advice regarding exercise. Poor follow-up of this information may have confounded our comparisons regarding exercise.
Substantial and sustained weight loss is typical of surgical weight loss interventions. Aspects of surgery, including postoperative convalescence, rapidity of weight loss, malabsorption of macronutrients and micronutrients, malnutrition, and neuro-hormonal changes to gastrointestinal physiology may all play a role in the %FFML after bariatric surgery. It is clear that the loss of FFM is generally favorable after non-diversionary LAGB surgery, indicating its safety and analogy to simple sustained caloric restriction. In contrast, BPD is a major gastrointestinal diversion inducing significant nutritional risk. There is also concern that with BPD greater weight loss may be accompanied by greater %FFML. Future additional studies of BPD may allow this to be shown with statistical certainty. Although the BPD procedure provides the best weight loss of all weight loss interventions, numerous collateral effects have been reported and its safety in regard to nutrition and body composition is potentially problematic.
Non-nutritional factors may influence body composition following BPD and RYGB surgery. With some knowledge regarding the possible mechanism of action of diversionary bariatric surgery some surgeons are targeting areas of hormonal importance, particularly ghrelin production.62 Alterations in ghrelin levels are subject to the surgical technique with BPD, RYGB and more recently sleeve gastrectomy. Reduced postoperative levels of this important growth hormone secretagogue may result in excessive loss of FFM.62, 63
To date, quantitative definition of ‘excessive loss of FFM’ is lacking. An obvious way to set standards of %FFML is to use normative data to calculate appropriate fat and fat-free composition of weight loss. The study of Fernandez et al.6 indicates that gender and ethnicity play important roles in the composition of body weight over a range of BMI (15–50 kg/m2). Using their regression models, it is possible to calculate the FFM loss required to achieve normal body composition at lower BMI, this may give some indication of optimal %FFML.
It appears that LCD with exercise and LAGB programs provide safe weight loss against which other therapies could be judged. In order to better establish guidelines on the optimal composition of weight loss and to ensure safe weight loss, assessment of changes in fat and FFM should become a standard measured outcome of all novel medical and surgical weight loss interventions.
Knowledge regarding the effect of significant weight loss on two compartment body composition is limited. Suitable validated methods are available to examine this important safety indicator for weight loss interventions. DEXA and ADP provide fast accurate measures with little inconvenience. Unfortunately, poorly validated methods of assessing body composition are commonly used. We demonstrate that within the literature there is a broad range of %FFML with some weight loss methods producing potentially problematic losses. Some important interventional factors influencing the proportion of FFM loss are the degree of caloric restriction, exercise and the type of bariatric surgery.