Potential conflict of interest: Nothing to report.
Accurate assessments of nutritional status are difficult to obtain in patients with cirrhosis. The aim of this study was to devise and validate a global nutritional assessment scheme for use in this patient population. Measures of body mass index (BMI) and mid-arm muscle circumference (MAMC) were combined with details of dietary intake in a semistructured, algorithmic construct to provide a nutritional assessment scheme for use in patients with cirrhosis; evaluated individuals were classified as adequately nourished, moderately malnourished (or suspected to be), or severely malnourished. There was good interobserver agreement in the nutritional categorization of 50 patients with cirrhosis (34 men, 16 women) using this scheme (κ = 0.79) and significant associations with the contributing objective variables—namely, BMI (Spearman's correlation r = −0.78; P < .001) and relative MAMC (r = −0.69; P < .001)—confirming its internal validity. There was a significant association between nutritional categorization in 20 patients with cirrhosis (10 men, 10 women) and estimates of total body protein obtained using a four-component model (r = −0.45; P = .046), confirming the external validity of the scheme. Finally, a significant association was found between poor nutritional status in 116 patients with cirrhosis (65 men, 51 women), followed for 14 to 52 months, and shorter subsequent survival (P = .0005), confirming the scheme's predictive validity. In conclusion, a global assessment scheme has been devised that provides a simple, reproducible, valid, and predictive method of assessing nutritional status in patients with cirrhosis. (HEPATOLOGY 2006;44:823–835.)
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Patients with chronic liver disease are frequently malnourished,1–3 and this has a detrimental effect on outcome.4–6 However, accurate assessments of nutritional status are not easily obtained in patients with cirrhosis,7, 8 making it difficult to identify those at risk for malnutrition and to evaluate the need for—and subsequent efficacy of—nutritional intervention.
The technique of subjective global assessment (SGA)9, 10 uses clinical information collected during history-taking and physical examination to determine nutritional status without recourse to objective measurements. This method of assessment, which has been used successfully to assess nutritional status in general medical and surgical patients, shows good to excellent interobserver reproducibility,9–11 and good convergent validity when compared with measured anthropometric variables.11
Hasse et al.12 adapted the original SGA for use in patients with cirrhosis, but the reproducibility and predictive validity of this modified SGA have only been examined by the original workers themselves.12–14 However, Naveau et al.15 have cautioned against the use of subjective assessment techniques in patients with cirrhosis, especially as a sole means of determining nutritional status, because the results do not accord with those obtained using anthropometry. Nevertheless, no attempts have been made, to date, to devise a global scheme incorporating both subjective and objective variables for use in this patient population.
The aims of the present study were: (1) to assess the reproducibility of the modified SGA12 and to test its external validity against anthropometric measures; (2) to use the modified SGA as a basis for devising an alternative global assessment technique incorporating anthropometric variables, and to test its repeatability and internal validity; (3) to test the external validity of the alternative global assessment scheme against an accurate measure of body composition; and (4) to observe its predictive validity in terms of survival. The study was undertaken in four separate phases over one 52-month period.
SGA, subjective global assessment; BMI, body mass index; MAC, mid-arm circumference; TSF, triceps skinfold; MAMC, mid-arm muscle circumference; RFH GA, Royal Free Hospital Global Assessment; DXA, dual-energy X-ray absorptiometry.
Patients, Methods, and Results
Phase 1: Reproducibility and Validity of the Modified SGA
Patients and Methods.
The study population comprised 26 patients (14 men, 12 women; mean age 50.4 years [range, 34-66]) with cirrhosis of varying etiology and severity (Table 1). The etiology of the liver injury was determined using historical, clinical, laboratory, and histological data, while the functional severity of the liver injury was assessed using the Child-Pugh grading system.16
Table 1. Demographic Details of the Patients with Cirrhosis Included in the Four Phases of the Study
Phase 1 (n = 26)
Phase 2 (n = 50)
Phase 3 (n = 20)
Phase 4 (n = 116)
Severity assessed using the Child-Pugh grading system.16
Patients were assessed independently, using the modified SGA,12 by two observers (A. M. M., C. T. S.) experienced in the nutritional management of patients with chronic liver disease.
Gastrointestinal symptoms that might influence nutrient intake, such as anorexia or food-related abdominal pain, were recorded and graded—in relation to their degree, frequency, and duration—as absent, mild, moderate, or severe. Bowel habits and any recent changes in stool frequency, color, or consistency were noted. A history and details of infections, renal dysfunction, hepatic encephalopathy, and gastrointestinal bleeding was sought. Pre-illness weight and weight change in the preceding 6 months were recorded; overall weight change was calculated in kilograms per month. Recent levels of activity and fatigability were assessed; dysfunction, when present, was further characterized by severity and duration.
Appetite was assessed as good, fair, or poor. Recent dietary intake was assessed using an established diet history method17 supplemented, where necessary, by information from relatives, nursing staff, and food record sheets. Details of dietary restrictions and nutritional support were recorded. These data were used to provide an overall impression of the adequacy of the diet in relation to estimated daily requirements, assessed using Schofield's modification of the Harris-Benedict equations.18–20 Intakes were categorized as adequate if they met estimated requirements, inadequate if they failed to meet estimated requirements but exceeded 500 kcal/d, or negligible if they provided fewer than 500 kcal/d.
A subjective evaluation of subcutaneous fat stores (good, fair or poor), and the presence of muscle wasting and fluid retention (none, mild to moderate, severe) was made by visual inspection. Height was measured to the nearest centimeter using a wall mounted or freestanding measure (Seca, Hamburg, Germany). Weight was measured to the nearest 0.1 kg using a weighing (Seca) or seat balance scale (Marsden W/M Group, London, UK).
The observers independently categorized the patients on the basis of the collected data according to the modified SGA scheme12 as: (1) adequately nourished; (2) moderately malnourished (or suspected to be); or (3) severely malnourished.
Independent anthropometric assessments were then undertaken by the two observers. An estimated dry weight was determined by deducting a weight for ascites and/or edema from the measured weight. The deduction was based on clinical assessment, previously documented weights, ascitic volumes removed at paracentesis, and published guidelines.21 BMI was calculated from the estimated dry weight and height
Mid-arm circumference (MAC) and triceps skinfold thickness (TSF) were measured on the nondominant side of the body using Holtain/Tanner-Whitehouse skinfold calipers and a steel tape measure (Holtain Ltd, Crymych, Dyfed, UK).22 Mid-arm muscle circumference (MAMC) was then calculated23:
MAMC and TSF measurements were compared with published standards,24 and expressed both in relation to the 5th percentile and as a percentage of the 50th percentile (relative MAMC) for the appropriate age and gender category.
Intraobserver repeatability for anthropometric assessments was evaluated by remeasurement of variables in 5 patients, by both observers, within one 5-day period.
Statistical analysis was undertaken using the Minitab software package version 10.2 (Minitab Inc.; State College, PA). Repeatability coefficients25 were calculated for anthropometric variables for each of the observers and were expressed as a percentage of the mean of the repeated values. Interobserver agreement in anthropometric measurements was evaluated using the intraclass correlation coefficient (ρ). Comparison of interobserver global categorization of nutritional status was undertaken using the κ statistic. Values for ρ and κ between 0.41 and 0.60 indicate a moderate association; values above 0.60 indicate a substantial relationship.26 Spearman rank correlations were used to examine the relationships between the anthropometric measurements, dietary intake, and the global categorization of nutritional status. Significance was accepted at the 5% probability level.
The intraobserver repeatability coefficients for anthropometric variables ranged from 0% to 8.0% (Table 2). The intraclass correlation coefficient for dietary intake was 0.42 but exceeded 0.8 for all of the anthropometric variables, indicating substantial interobserver agreement (Table 3). The assessment of nutritional status was concordant in 21 (81%) of the 26 patients (κ = 0.67). Where discordance occurred, it was by a single category and without bias. Nutritional categorization was significantly associated with both relative TSF (r = −0.48; P = .013) and relative MAMC (r = −0.53; P = .005), but not with BMI (r = −0.30; P = .14) (Table 3).
Table 2. Interobserver Repeatability Coefficients for Anthropometric Variables Measured in 5 Patients with Cirrhosis on Two Occasions in Phase 1 of the Study
Data are expressed as the repeatability coefficient and the percentage of the mean repeated values in parentheses.
Table 3. Interobserver Agreement in the Assessment of Nutritional Variables in the Patients with Cirrhosis in Phase 1 of the Study and their Relationship with Nutritional Status Categorized Using the Modified SGA12
Thus, the modified SGA method of nutritional assessment12 was reproducible between observers and was significantly associated with two of the three anthropometric variables measured.
Phase 2: Derivation and Evaluation of an Alternative Global Assessment Scheme
An alternative assessment scheme was devised that incorporated an assessment of dietary intake and two anthropometric measures, BMI and MAMC. These anthropometric variables were selected because they are strongly associated with survival in this patient population.4, 6 The three variables were incorporated into a semistructured, algorithmic construct, which allocates patients to one of three nutritional categories (Fig. 1). Using this scheme, patients with a BMI of 20 or higher, measurements of MAMC on or above the 5th percentile, and an adequate dietary intake would be categorized as adequately nourished; those with a BMI of less than 20, an MAMC below the 5th percentile, and an inadequate or negligible dietary intake would be classified as severely malnourished, while those with a combination of acceptable and depleted anthropometric and dietary intake variables would be considered or suspected to be moderately malnourished. A subjective override was included to allow the assessor to change the nutritional category by a single grading if warranted on clinical grounds. This override could be used, for example, to reclassify an individual who reported profound recent weight loss yet still maintained a BMI above 20 kg/m−2 from adequately nourished to moderately malnourished.
This newly devised, Royal Free Hospital Global Assessment (RFH GA) scheme was then evaluated for interobserver reproducibility and internal validity.
Patients and Methods.
The study population comprised 50 inpatients (34 men, 16 women; mean age 50.1 years [range, 27-70]) with cirrhosis of varying etiology and severity, as previously defined (Table 1).
The patients were assessed by two observers (A. M. M., C. T. S.) working independently. Information was collected on gastrointestinal and other clinical symptoms, functional capacity, dietary intake, and physical status. Height, weight, MAC, and TSF were measured, dry weight was estimated, and BMI, MAMC, and relative MAMC were calculated. The demographic, clinical, and dietary intake data were recorded on a collection form together with the measured and derived anthropometric variables (Appendix). Nutritional status was categorized using the RFH GA algorithm and a final nutritional category determined, with or without application of the subjective override.
The interobserver comparisons and the relationship between global nutritional status and anthropometric measurements were evaluated using κ and ρ statistics. Spearman rank correlations were used to examine the relationships between anthropometric measurements, dietary intake, and global nutritional status.
Measurements of height, weight, estimated dry weight, BMI, MAC, MAMC, and relative MAMC showed excellent interobserver agreement (ρ ≥ 0.90) (Table 4). TSF and relative TSF measurements showed good interobserver agreement (ρ = 0.72 and 0.74, respectively), but the interobserver agreements between estimates of weight change and dietary intake (ρ = 0.26 and 0.40, respectively) were poor (Table 4).
Table 4. Interobserver Agreement in the Assessment of Nutritional Variables in the Patients with Cirrhosis in Phase 2 of the Study and their Relationship with Nutritional Status Categorized Using the RFH GA
The subjective override was employed in 9 (18%) and 8 (16%) of the patients by the two observers, respectively, without overlap. However, there was overall concordance in the categorization of nutritional status in 44 (88%) of the 50 patients (κ = 0.79). Where discordance occurred, it was by a single category. Nutritional categorization was significantly associated with both BMI (r = −0.78; P < .001) and relative MAMC (r = −0.69; P < .001) (Table 4).
Thus, the RFH GA showed good interobserver agreement and internal validation against the contributing objective variables BMI and relative MAMC.
Phase 3: External Validity of the RFH GA Scheme
Accurate body composition data were obtained using a four-component model based on data obtained from densitometry, deuterium dilution, and dual-energy X ray absorptiometry (DXA).27, 28 This model defines the chemically distinct water, fat, protein, and mineral and thus provides direct and accurate measures of the constituents of the fat-free mass.
Patients and Methods.
The study population comprised 20 outpatients (10 men, 10 women; mean age 48.7 years [range, 26-67]) with alcohol-related cirrhosis of varying severity, as previously defined (Table 1). They had been abstinent from alcohol for a mean period of 24.4 months (range, 3-76). The patients' nutritional status was assessed and categorized by one observer (A. M. M.) using the RFH GA scheme. Body density was measured using a standard underwater weighing procedure.29, 30 Total body mineral was measured using a Hologic QDR-1000/W DXA scanner and the Enhanced Whole Body V5.61 analysis program (Hologic Inc., Waltham, MA). Total body water was measured by deuterium dilution after oral dosing of deuterium (0.40 g 2H2O per kilogram body weight; Sigma Chemical Company, Poole, UK). Deuterium enrichment in saliva was measured using Fourier transform infrared spectrometry (ATI Mattson Genesis Series FTIR; ATI, Cambridge, UK). Fat mass, fat-free mass, and total body protein were calculated using the four-component model.27, 28
Spearman rank correlations were used to examine the relationships between fat mass, fat-free mass, and total body protein and nutritional status categorized using the RFH GA scheme.
Approval for this phase of the study was provided by the Ethical Practices Sub-Committee of the Royal Free Hospital, Hampstead, NHS Trust. All subjects provided written informed consent to participate.
Using the RFH GA scheme, 7 men and 9 women were categorized as adequately nourished, 3 men were categorized as moderately malnourished, and 1 woman was categorized as severely malnourished. Nutritional categorization was significantly associated with total body protein in the group as a whole (r = −0.45; P = .046) and in men (r = −0.80; P < .006), but not in women (r = −0.52; P = .12). Nutritional categorization was also significantly associated with fat-free mass in men (r = −0.72; P < .018). The relationship between nutritional status and fat mass was not significant either in the group as a whole or when separated by gender.
Poor nutritional status, defined using the RFH GA scheme, was thus significantly correlated with depletion of total body protein stores and fat-free mass, particularly in men, confirming its external validity.
Data were systematically collected for phases 1, 2, and 3 of the study over a 20-month period. No drift was found in agreement between observers in the early stages compared with the later stages of each phase.
Phase 4: Predictive Validity of the RFH GA Scheme
Patients and Methods.
The study population comprised 116 consecutively referred patients (65 men, 51 women; mean age 50.1 years [range, 26-76]) with cirrhosis of varying etiology and severity, as previously defined (Table 1).
The patients' nutritional status was assessed and categorized by one observer (A. M. M.) using the RFH GA scheme. Patients were followed for 14-52 months after initial assessment or until death if it occurred earlier; none of the patients assessed underwent liver transplantation.
The number of patients who died and the time intervals between nutritional assessment and death were examined, in the three nutritional categories, using the Kaplan-Meier technique, and differences in the survival curves were compared using a log rank test. The analysis was repeated after stratifying the patients using the Child-Pugh grading scale.16 The individual and combined prognostic values of BMI, MAMC, and dietary intake were examined using the Cox proportional hazards model with the Stata statistical package version 9.1 (StataCorp, TX). Data were analyzed for the group as a whole and separately by gender.
A total of 39 patients died a median of 13.4 months (range, 1-43) after assessment (Table 5). There were significant differences in survival in relation to nutritional status (χ2 15.04; df = 2; P = .0005) (Fig. 2). Cumulative survival was significantly worse in the patients who were moderately malnourished (hazard ratio 3.02 [95% CI 1.37-6.67]; P = .006) or severely malnourished (hazard ratio 5.56 [95% CI 2.21-14.0]; P < .001) compared with their adequately nourished counterparts (Table 5). Patients with decompensated liver disease were more likely to be malnourished than those with compensated disease (χ2 34.5; df = 4; P < .0001) (Fig. 3). However, the predictive validity of the nutritional status remained significant even after adjustment for the degree of hepatic impairment (χ2 6.18; df=2; P = .046).
Table 5. Details of Deaths and Median (Range) Survival Times in the Patients with Cirrhosis in Phase 4 of the Study by Nutritional Status Classified Using the RFH GA Scheme and Gender
Nutritional Category, n (%)
Total (n = 116)
Men (n = 65)
Women (n = 51)
Dead, n (%)
Hazard Ratio (95% CI)
Dead, n (%)
Hazard Ratio (95% CI)
Dead, n (%)
Hazard Ratio (95% CI)
Adequately nourished 50 (43%)
Moderately malnourished (or suspected to be) 48 (41%)
(P = .006)
(P = .006)
(P = .52)
Severely malnourished 18 (16%)
(P < .001)
(P = .001)
(P = .05)
Total 116 (100%)
Twenty-four (37%) men died after a median of 7 months (range, 1-40); 15 (29%) women died after a median of 19 months (range, 1-43) (Table 5). The relationship between nutritional status and survival was significant in men but not in women (Fig. 4), although the interaction between nutritional category and gender was not significant (χ2 2.79; df = 2; P = .25).
In the population as a whole, patients with a BMI below 20 kg/m−2 had a poorer survival than those with values of 20 or higher (hazard ratio 2.19 [95% CI 1.1-4.33]; P = .03). MAMC below the 5th percentile and an inadequate dietary intake were both significantly predictive of death when individually compared with values of MAMC equal to the 5th percentile or above (hazard ratio 2.30 [95% CI 1.22-4.32]; P = .01) and to an adequate dietary intake (hazard ratio 3.73 [95% CI 1.87-7.43]; P < .001). The combined effects of BMI, MAMC, and dietary intake were strongly predictive of death (χ2 19.76; df = 4; P = .0006). The global score did not contribute any additional predictive value (χ2 0.53; df = 2; P = .76), indicating that the subjective override did not influence the predictive ability of the RFH GA scheme.
Although the difficulties of assessing nutritional status in patients with cirrhosis are widely recognized,31–33 little work has been undertaken to evaluate the reliability and validity of the various assessment methods available. Single objective assessment variables, such as percentage ideal body weight or serum albumin cannot be used in this patient population because of the innate confounding effects of fluid retention and alterations in protein metabolism.31–33 Equally, many of the composite objective methods currently available, such as the prognostic nutrition index,34 include variables such as serum transferrin concentrations and delayed cutaneous hypersensitivity, which have no predictive value in patients with chronic liver disease.31, 33, 35, 36 Clearly, a composite method of assessment that includes appropriate variables,31, 33, 35 and which is easily applied, is needed to optimize nutritional management in this patient population.
The SGA method, as originally devised,10, 11 lacks numerical structure to facilitate the monitoring of changes in response to disease progression and nutritional therapy. However, it is more useful than objective measures alone for identifying individuals at nutritional risk because of its ability to encompass the multitude of factors that influence nutritional status. Indeed, several workers have used the original SGA methodology to assess nutritional status in patients with chronic liver disease with some success in categorizing risk groups.36–38
Hasse et al.12 modified the original SGA for use in liver transplant candidates and have used this technique to assess nutritional status in a large number of patients—although, to date, these data have only been published in abstract or summary form.14, 39 These authors12 assessed the reliability of the modified SGA in 20 individuals, 12 of whom had cirrhosis, and reported an interobserver agreement of 80% and a Cronbach coefficient α-test, examining internal validity, of 0.71. However, no further evaluations of this modified SGA have been undertaken until now.
The results from phase 1 of this study show that the modified SGA12 is reproducible between observers and that the global categories of nutritional status are significantly associated with anthropometric variables. However, the merit of only using subjective measures of assessment in patients with liver disease has been questioned by Naveau et al.,15 who reported that a subjective clinical evaluation of nutritional status in 260 patients with alcohol-related cirrhosis failed to identify “severe malnutrition,” as defined anthropometrically, in 28% of patients.
In phase 2 of this study, the combination of specific objective and subjective variables in the semistructured framework of the RFH GA scheme increased the agreement of nutritional categorization between observers and improved the strength of the relationships between categorization and anthropometric measurements, particularly with BMI and relative MAMC—although this improvement was anticipated, in part, because of the contribution of these anthropometric variables to the algorithm.
The significant correlation between the categorization of nutritional status and total body protein—and, in men, fat-free mass— in phase 3 of this study confirms that the RFH GA scheme provides a valid assessment of nutritional status. The four-component model27, 28 used to evaluate body composition is considered to be a “gold standard,” because the assumptions used in the calculations are robust even in extremes of hydration and abnormalities of mineralization. However, the practicalities of evaluating body density by underwater weighing limited the number of patients included in the study to those who were physically robust enough to participate. Nevertheless, even within this relatively well population, the majority of whom were considered to be adequately nourished, the RFH GA scheme still identified those individuals who had measurably depleted total body protein, confirming its validity in this respect.
In addition to its reproducibility and validity against accurate measures of body composition, the RFH GA scheme—and the individual contributing variables BMI, MAMC, and dietary intake—provide useful predictive measures of clinical outcome, as shown in phase 4 of the study. Nutritional status and the degree of hepatic dysfunction are significantly associated.1, 3, 4, 6, 31, 40 Nevertheless, nutritional status, defined using the RFH GA scheme, was a significant predictor of survival even when controls were exercised for the confounding effects of hepatic decompensation.
Although the relationship between nutritional status and clinical outcome in patients with cirrhosis has been widely investigated, in both transplanted5, 37–39, 41 and nontransplanted populations,4, 6, 7, 42–44 many of the studies reporting a significant relationship between impaired nutritional status and survival are based on use of inappropriate “nutritional” variables such as serum albumin concentrations, creatinine-height index, and bioelectrical impedance analysis.
Few data are available from studies using more “acceptable” nutritional assessment methods. Hasse et al.14 reported reduced 1- and 3-year patient and graft survival rates in liver transplant recipients classified, using the modified SGA, as severely malnourished. Merli et al.4 showed that impaired nutritional status, assessed on the basis of nutritional history and physical examination, or on the basis of muscle depletion, defined as a mid-arm muscle area below the 5th percentile, was significantly associated with poorer survival in individuals with Child-Pugh grade A and B cirrhosis but not in more severely decompensated patients. However, nutritional status, as defined in their study, was not an independent predictor of survival when allowance was made for the confounding effects of hepatic decompensation. Alberino et al.6 found that malnutrition, defined using percentile thresholds for MAMC and/or TSF, was an independent predictor of survival, but they did not adjust for the degree of hepatic dysfunction. However, they did find that inclusion of these anthropometric variables increased the prognostic accuracy of the Child-Pugh score, confirming the findings of Abad-Lacruz et al.,44 who reported that inclusion of the MAMC improved the predictive power of this scoring system.
Very little attention has been paid to potential gender differences in the prevalence of malnutrition and outcome in patients with cirrhosis and to their interrelationship. Merli et al.4 reported significantly higher cumulative survival rates in women with cirrhosis compared with men and identified gender as an independent prognostic variable. However, they did not investigate the effect of nutritional status on outcome by gender. Alberino et al.6 reported no significant difference in overall survival rates in men and women with cirrhosis followed over a 2-year period nor in men and women with severe malnutrition defined as an MAMC and /or TSF below the 5th percentile. However, they did not control for the potential confounding effects of hepatic dysfunction.
The prevalence of malnutrition in the patients included in phase 4 of the present study was similar in men and women and, although no statistically significant interaction between nutritional status and gender was identified, gender differences in survival in relationship to nutritional status were observed. The prevalence of anthropometric abnormalities in patients with cirrhosis varies between men and women.3, 4, 6, 45 reflecting, at least in part, intrinsic gender differences in body composition. In patients with cirrhosis, malnutrition is characterized by preferential loss of fat mass in women and muscle mass in men.3, 4, 6, 45, 46 Thus, it is tempting to speculate that malnutrition may have a less detrimental effect on survival in women with cirrhosis because their muscle mass is relatively preserved. Alternatively, the absence of a significant gender interaction in the present study may reflect the fact that very few women died during the follow-up period. Clearly, however, this is a finding worthy of study in a larger, independent data set.
The techniques employed in the RFH GA scheme are simple to perform, and the equipment is easily obtained. The accuracy of the BMI calculated from estimated values of dry weight21 may be questioned. However, excellent interobserver agreement was observed for both estimated dry weight and BMI. Similarly, the validity of the subjective assessment of dietary intake in relation to estimated energy requirements may also be questioned. However, the evaluation of dietary intake is still considered useful in the context of nutrition assessment and has an established role in other published nutrition screening tools.47–49 More importantly, dietary intake was found to be an independent predictor of survival in the present study, confirming the findings of others.3, 50
In conclusion, a global nutritional assessment method, the RFH GA scheme, has been developed for use in patients with cirrhosis. It is reproducible, valid against a gold standard method for assessing body composition, and a significant predictor of survival. Further validation studies should now be undertaken by other workers in the field. 1
The authors thank Marinos Elia, Nigel Fuller, and Graham Jennings of the MRC Dunn Clinical Nutrition Centre, Cambridge, for their expertise and practical assistance in performing the body composition assessments in phase 3.