Osteoarthritis (OA) is ranked as the 11th highest contributor of disability in the world, with a prevalence of 3·8 and 0·85 % of knee and hip OA, respectively(Reference Cross, Smith and Hoy1). Disability may result from ongoing decline in physical function limiting independence in activities of daily living, particularly in older adults(Reference Manini2). Being overweight or obese is a major risk factor for development and progression of OA(Reference Silverwood, Blagojevic-Bucknall and Jinks3), including worsening of functional limitations(Reference Manini2). Thus begins a cyclic relationship where reduced physical function contributes to reduced activity levels, resulting in further weight gain, which can negatively affect functional status. Therefore, addressing functional impairments and weight management in this population is a significant public health issue.
Best practice guidelines for the management of overweight and obesity in the general population recommend lifestyle modifications primarily to promote moderate weight loss, including a hypoenergetic diet and increased levels of exercise(4). A very low-energy diet (VLED) is utilised in some cases, where diet and exercise strategies have not been successful for weight loss(4). VLED are defined as lowering energy intake to approximately 800 kcal or less per day, with total meal replacements commonly utilised(5). In the long term, lifestyle modifications are considered more sustainable compared with VLED because of associated costs, compliance issues and potential negative effects to body composition(4). Clinical practice guidelines in OA consistently recommend exercise therapy, including aerobic and strength training, and weight management incorporating diet and exercise, for improvements in physical function(Reference Brosseau, Wells and Tugwell6–8). Clinical trials in OA have shown significant improvements in physical function with dietary weight loss alone. The IDEA study by Messier et al. demonstrated significant weight loss and improvements in Western Ontario and McMaster Universities Arthritis Index (WOMAC) physical function that were of a similar magnitude in participants undergoing a dietary weight loss intervention, with or without exercise(Reference Messier, Mihalko and Legault9).
Overall, the body of evidence regarding the effect of weight loss on symptoms of OA is growing rapidly. Clinical trials in OA have investigated the effects of both lifestyle interventions (diet and exercise) and VLED. Previous systematic reviews have demonstrated beneficial effects of weight loss on symptoms of OA, including physical function(Reference Alrushud, Rushton and Kanavaki10–Reference Hall, Castelein and Wittoek12). However, preliminary searches determined that these reviews did not include all relevant dietary weight loss studies, in particular VLED interventions. As such, it is unclear as to which of these dietary strategies is the most effective to induce significant weight loss and improvement in symptoms, namely physical function, in this population. Therefore, the aim of this systematic review is to compare the effects of different dietary interventions for weight loss, with or without exercise, on physical function in overweight and obese adults with OA.
Methods
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines(Reference Moher, Liberati and Tetzlaff13) were utilised, and the systematic review protocol was registered in Prospero: International prospective register of systematic reviews and can be accessed here: http://bit.ly/2yVZCg1.
Inclusion criteria
Included studies were randomised controlled trials (RCT) and uncontrolled clinical trials investigating the effect of dietary weight loss interventions, with or without exercise, on physical function in adults classified as overweight and obese (aged 18 years and over) with OA. To assess the effects of the intervention, primary outcomes included body weight and physical function as either self-reported or performance-based measures. Studies were restricted to the English language, with no date restrictions.
Search strategy
A preliminary search of the Medline database was undertaken to identify keywords and index terms, by analysis of words in the title, abstract, description of articles and Medical Subject Headings. Using the keywords identified in the first search, a second search was completed in Medline, Cochrane, Embase, AMED and CINAHL databases (September 2017 and updated May 2019) to identify potential articles for inclusion. The search strategy included sets of search terms to describe: (i) the population group, for example, overweight, obesity and OA; (ii) intervention components, for example, hypoenergetic diet and weight loss and (iii) outcomes, for example, weight and BMI and has been reproduced (see online supplementary material, Supplementary Table 1). This allowed for identification of dietary weight loss studies and subsequently those that measured different physical function outcomes. A hand search was conducted in which relevant journals and reference lists of retrieved articles were examined to check for additional studies not identified in the database search.
Study selection
Studies identified using the search strategy were screened independently by two reviewers (E.J.W. and S.K.B.) by examining the title, abstract and keywords. Full texts of articles that met the inclusion criteria, or any that were unclear, were retrieved for further reviewing. Full texts were then screened, and articles were classified as included or excluded, and the reasons for exclusion were recorded (i.e. participants, intervention, outcome and study design). Any discrepancies throughout the selection process were resolved by discussion with a third author (P.G.O.).
Quality appraisal
All included articles were assessed for quality and risk of bias using the Quality Criteria Checklist for primary research(14). The tool comprises a checklist of questions within ten domains that address methodological rigour. Articles were given a quality rating of positive (strong quality), negative (weak quality) or neutral (neither exceptionally strong nor exceptionally weak), based on these criteria. Two authors (E.J.W. and S.K.B.) assessed the quality of articles independently, and any discrepancies were resolved by consultation with a third author (P.G.O.).
Data extraction and synthesis
Data were extracted against a standardised data extraction template by one author (E.J.W.) and checked by another (S.K.B.) with reference to the full text of the article. Any discrepancies were resolved by discussion with a third author (P.G.O.). The following information was collected from each study: author, year of study and setting; participant details including age, sample size and retention rate; details of the intervention including treatment and control group, length of intervention and follow-up period; study design; and study results including before and after measures of outcomes of interest. Studies were categorised into three groups based on diet intervention type: (i) lifestyle – instruction to consume a hypoenergetic diet using conventional foods, (ii) conventional diet with partial use of meal replacements (DMR) and (iii) VLED – approximately ≤810 kcal/d, with sole use of meal replacements, as per global food standards(5).
Study data from RCT that were comparable in terms of outcome measures and follow-up periods were pooled to complete a meta-analysis. The mean difference was calculated for studies using the same instrument for measurement (i.e. WOMAC), and results reported as mean differences with 95 % CI and displayed as a forest plot. The effect of heterogeneity was quantified by calculating I 2 (range: 0–100 %). If I 2 was >30 %, a random effects model was used to account for heterogeneity between studies(15). Where multiple intervention groups were used, the diet or diet and exercise groups were selected for comparison against controls. Where data were not presented as mean and standard deviation, conversions were performed to obtain these measures(15). All statistical analyses were conducted using the Cochrane Collaboration, Review Manager software (RevMan version 5.3, Copenhagen).
Results
Study selection
The PRISMA flow diagram (Fig. 1) provides an overview of the screening and selection process for articles to be included in the review. The initial database search identified 1024 articles after the removal of duplicates, and two articles were located via hand search. After the first screening, seventy-seven articles were retrieved for full-text review. From this, fifty-eight articles were excluded. Reasons for exclusion are provided in the PRISMA flow diagram (Fig. 1). Nineteen articles were included in the review, with eight included in the meta-analyses. Articles were categorised according to intervention type: lifestyle (n 8), conventional diet and meal replacements (DMR; n 5) and very low-energy diets (VLED; n 6).
PRISMA flow diagram. Overview of the screening and selection process for articles to be included in the review
Study quality
In terms of quality rating (Table 1), nine of the nineteen included studies received a neutral rating due to non-comparable study groups at baseline (n 1)(Reference Somers, Blumenthal and Guilak16) or having no comparison group (n 8)(Reference Aaboe, Bliddal and Messier17–Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24). The remaining ten studies received a positive quality rating(Reference Messier, Mihalko and Legault9,Reference Bliddal, Leeds and Stigsgaard25–Reference Hughes, Tussing-Humphreys and Schiffer33) , of which five were lifestyle studies(Reference Magrans-Courtney, Wilborn and Rasmussen27–Reference Messier, Loeser and Mitchell29,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) , two were DMR studies(Reference Messier, Mihalko and Legault9,Reference Miller, Nicklas and Davis30) and three were VLED studies(Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26,Reference Riecke, Christensen and Christensen31) . These studies all specified clear inclusion criteria, had comparable study group characteristics at baseline and provided detailed information related to intervention components and outcome measures. Patient compliance was measured in eleven studies, via food diaries, exercise logs, pedometers and attendance at nutrition and exercise programmes(Reference Messier, Mihalko and Legault9,Reference de Luis, Izaola and Garcia Alonso20,Reference Martin, Fontaine and Nicklas22–Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Magrans-Courtney, Wilborn and Rasmussen27–Reference Miller, Nicklas and Davis30,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) . Due to the nature of the interventions and study design, blinding of participants was not always possible; however, outcome assessment was blinded in all studies. All studies accounted for all participants from baseline to follow-up. Seven studies did not clearly outline the study limitations or potential biases in the discussion(Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Bliddal, Leeds and Stigsgaard25–Reference Messier, Loeser and Mitchell29,Reference Riecke, Christensen and Christensen31) .
Quality appraisal summary: a summary of the ten domains addressing scientific soundness of primary research from ADA quality criteria checklist
Study characteristics
Of the nineteen studies included in the review, eight were lifestyle studies of which five were RCT(Reference Somers, Blumenthal and Guilak16,Reference Magrans-Courtney, Wilborn and Rasmussen27,Reference Messier, Loeser and Miller28,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) , five were DMR studies of which two were RCT(Reference Messier, Mihalko and Legault9,Reference Miller, Nicklas and Davis30) , and six were VLED studies of which two were RCT(Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26) . The remaining ten studies were uncontrolled clinical trials(Reference Aaboe, Bliddal and Messier17–Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Messier, Loeser and Mitchell29,Reference Riecke, Christensen and Christensen31) . A description of study characteristics including study sample and intervention details is provided in Table 2. Overall, participant numbers ranged from 24 to 1383, with an age range of 20–90 years. For the sixteen studies that included both male and female participants, the proportion of female participants ranged from 57 to 89 %. Two lifestyle studies and one VLED study included only female participants(Reference Martin, Fontaine and Nicklas22,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Magrans-Courtney, Wilborn and Rasmussen27) . Seven studies included only obese participants (BMI ≥ 30 kg/m2)(Reference Aaboe, Bliddal and Messier17,Reference Bartels, Christensen and Christensen19–Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Miller, Nicklas and Davis30,Reference Riecke, Christensen and Christensen31) , and the average BMI of all study samples ranged from 31·7 to 40·8 kg/m2. Seventeen studies included participants who had knee OA(Reference Messier, Mihalko and Legault9,Reference Somers, Blumenthal and Guilak16–Reference Bartels, Christensen and Christensen19,Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Martin, Fontaine and Nicklas22,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24–Reference Wang, Miller and Messier36) ; however, one VLED study included patients with hip or knee OA(Reference de Luis, Izaola and Garcia Alonso20), and another lifestyle study included patients with hip OA only(Reference Paans, van den Akker-Scheek and Dilling23). The lifestyle interventions (n 8) provided healthy eating nutrition education, counselling, behavioural therapy and all included instruction for exercise, such as aerobic and resistance training, which were 8 weeks to 18 months in duration(Reference Somers, Blumenthal and Guilak16,Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23,Reference Magrans-Courtney, Wilborn and Rasmussen27–Reference Messier, Loeser and Mitchell29,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) . Seven lifestyle studies provided face-to-face interventions, while O’Brien et al. used a telephone coaching programme(Reference O’Brien, Wiggers and Williams32). Two lifestyle studies prescribed energetic intake that ranged from 1200 to 1600 kcal/d for the intervention groups(Reference Somers, Blumenthal and Guilak16,Reference Magrans-Courtney, Wilborn and Rasmussen27) while moderating energy intake at different stages of the intervention(Reference Magrans-Courtney, Wilborn and Rasmussen27) or based on participant gender(Reference Somers, Blumenthal and Guilak16). Three other interventions aimed for a general reduction in energetic intake(Reference Martin, Fontaine and Nicklas22,Reference Messier, Loeser and Miller28,Reference Messier, Loeser and Mitchell29) . The DMR studies all provided up to two meal replacements per day with meal plans for the third meal using conventional foods, and/or nutrition education and behavioural therapy, that ranged from 12 weeks to 18 months in duration(Reference Messier, Mihalko and Legault9,Reference Atukorala, Makovey and Lawler18,Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Miller, Nicklas and Davis30) . Two DMR studies prescribed energy intake of 1109 and 1035 kcal/d(Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24) . Two other DMR studies defined a reduction in energy intake of 800–1000 kcal/d and also prescribed exercise interventions (aerobic and strength training)(Reference Messier, Mihalko and Legault9,Reference Miller, Nicklas and Davis30) . The VLED studies (n 6) were the most restrictive using only meal replacements to provide a diet of <810 kcal/d. None of these studies included exercise interventions(Reference Aaboe, Bliddal and Messier17,Reference Bartels, Christensen and Christensen19–Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24–Reference Christensen, Astrup and Bliddal26,Reference Riecke, Christensen and Christensen31) . Four VLED studies were 16 weeks(Reference Aaboe, Bliddal and Messier17,Reference Bartels, Christensen and Christensen19,Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Riecke, Christensen and Christensen31) , while one was 8 weeks in length(Reference Christensen, Astrup and Bliddal26); however, one study alternated between a VLED (810 kcal) and hypoenergetic diet (1200 kcal), over 52 weeks(Reference Bliddal, Leeds and Stigsgaard25). Average prescribed energy intake was 711 kcal/d (range: 415–810 kcal/d) in the VLED treatment groups.
Study characteristics: description of study details, including participant and intervention information†
Δ, change; GCM, glucosamine/chondroitin/methlysulphonylmethane; NR, not reported; VLED, very low-energy diet.
† Results reported as mean ± sd or mean ± sem.
*P < 0·05, **P < 0·01, ***P < 0·001.
Body weight and BMI
Lifestyle interventions
Of the eight lifestyle studies, the average baseline weight was 93·1 kg (range: 86·2–108·4 kg), with average weight loss ranging between 0·1 and 8·5 kg for treatment groups (Table 3). Average baseline BMI was 34·0 kg/m2 (range: 31·7–38·0 kg/m2), with only four studies reporting on change in BMI at follow-up (Table 2)(Reference Somers, Blumenthal and Guilak16,Reference Martin, Fontaine and Nicklas22,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) . Of the five lifestyle RCT(Reference Somers, Blumenthal and Guilak16,Reference Magrans-Courtney, Wilborn and Rasmussen27,Reference Messier, Loeser and Miller28,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) , these were of varying duration between 8 weeks and 18 months(Reference Messier, Loeser and Miller28,Reference Hughes, Tussing-Humphreys and Schiffer33) . Messier et al.(Reference Messier, Loeser and Miller28) conducted a long-term (18-month) RCT that compared three intervention groups – dietary weight loss, exercise only and diet plus exercise – with a healthy control group. The diet-only and diet plus exercise groups lost an average of 4·6 and 5·2 kg body weight, respectively, which was statistically significantly lower than the control group (−1·1 kg; P < 0·05)(Reference Messier, Loeser and Miller28). In contrast, two six-month RCT by Somers et al. and O’Brien et al. reported no significant changes in weight or BMI(Reference Somers, Blumenthal and Guilak16,Reference Hughes, Tussing-Humphreys and Schiffer33) . A 10-week lifestyle RCT by Magrans-Courtney et al. compared the effects of a high-carbohydrate (55 % energy intake) or high-protein (55–63 % energy intake) weight loss diet in combination with circuit resistance training, reporting small but non-significant weight loss of 1·6 and 2·4 kg, respectively(Reference Magrans-Courtney, Wilborn and Rasmussen27). After the 2-month intervention, Hughes et al. reported a statistically significant reduction in weight (−2·0 kg) and BMI (−0·7 kg/m2) in the diet and exercise treatment group, compared with the exercise-only control group (P < 0·01)(Reference Hughes, Tussing-Humphreys and Schiffer33).
Weight and physical function outcomes for included studies: reporting of baseline and follow-up results of weight and physical function measures including performance-based and self-reported measures†
Δ, change; WOMAC, Western Ontario and McMaster Universities Arthritis Index (function subscale); NR, not reported; KOOS-FDL: knee osteoarthritis outcome score – function in daily living; KOOS-FSR: knee osteoarthritis outcome score – function in sport and recreation.
† Results reported as mean ± sd or mean ± sem.
*P < 0·05, **P < 0·01, ***P < 0·001.
The three uncontrolled lifestyle studies(Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23,Reference Messier, Loeser and Mitchell29) utilised a combination of diet and exercise via nutrition education and behavioural therapy over 6–8 months(Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23,Reference Messier, Loeser and Mitchell29) , reporting a mean weight loss of 6·6 kg (range: −5·6 to −8·5 kg). Martin et al. also reported a significant reduction in BMI of 1·1 kg/m2 (P < 0·01)(Reference Martin, Fontaine and Nicklas22), while Messier et al.(Reference Messier, Loeser and Mitchell29) and Paans et al.(Reference Paans, van den Akker-Scheek and Dilling23) did not report on change in BMI.
Diet and meal replacement interventions
Five studies utilised partial meal replacements (i.e. up to two per d) with conventional foods. These studies reported an average baseline weight of 95·8 kg (range: 92·3–99·3 kg), with weight loss ranging from 7·3 to 10·6 kg in the treatment groups. Average baseline BMI was 35·5 kg/m2 (range: 33·5–40·8 kg/m2), and change in BMI was reported in three DMR studies(Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Miller, Nicklas and Davis30) . Messier et al.(Reference Messier, Mihalko and Legault9) conducted a second 18-month RCT, comparing a diet-only and diet plus exercise intervention, to an exercise-only control group. The diet-only and diet plus exercise interventions resulted in statistically significant weight loss (−8·9 and −10·6 kg, respectively), compared with controls (−1·8 kg; P < 0·01)(Reference Messier, Mihalko and Legault9). Miller et al. conducted a 6-month intervention, incorporating up to two meal replacements per day, behavioural therapy and exercise (aerobic and strength training) on 3 d/week, compared with a control group provided with general health information(Reference Miller, Nicklas and Davis30). The intervention resulted in a statistically significant weight loss of 8·3 kg (P < 0·01), and a reduction in BMI of 8·1 kg/m2 (significance not reported), compared with the control group(Reference Miller, Nicklas and Davis30).
The three uncontrolled DMR studies were diet-only interventions (i.e. did not include exercise interventions), conducted over 12–18 weeks(Reference Atukorala, Makovey and Lawler18,Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24) . All three studies reported statistically significant weight loss at an average of 7·8 kg (range: −7·3 to −8·3 kg)(Reference Atukorala, Makovey and Lawler18,Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24) , while change in BMI was only reported in two studies (de Luis et al. and López Gómez et al.) with an average reduction of 3·5 kg/m2(Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24) .
Very low-energy diet interventions
Of the six VLED studies, the average baseline weight was 99·4 kg (range: 95·7–104·1 kg), with average weight loss ranging between 10·9 and 13·6 kg. The average baseline BMI was 36·4 kg/m2 (range: 35·2–37·5 kg/m2), with only two studies reporting on change in BMI at follow-up(Reference Aaboe, Bliddal and Messier17,Reference Riecke, Christensen and Christensen31) . Of the four uncontrolled studies(Reference Aaboe, Bliddal and Messier17,Reference Bartels, Christensen and Christensen19,Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Riecke, Christensen and Christensen31) , these ranged between 12 and 16 weeks in duration. The CAROT study, as reported by Riecke et al., compared a VLED (415 kcal/d) with a low-energy diet (LED; 810 kcal/d) for 8 weeks, after which participants were transitioned to a diet of 1200 kcal/d using usual foods, for the final 8 weeks of the 16-week study(Reference Riecke, Christensen and Christensen31). Both the VLED and LED dietary regimens demonstrated similar reductions in weight (−13·3 and −12·2 kg, respectively) and BMI (−4·8 and −4·0 kg/m2, respectively), although there was no significant difference between the two groups(Reference Riecke, Christensen and Christensen31). Due to the small difference of weight loss between dietary regimens in the CAROT study, the groups were collapsed for secondary data analysis, and average weight loss was reported as 13 kg (P < 0·01)(Reference Aaboe, Bliddal and Messier17,Reference Bartels, Christensen and Christensen19,Reference Henriksen, Christensen and Danneskiold-Samsoe21) , and a change in BMI of −5·1 kg/m2 (P < 0·01)(Reference Aaboe, Bliddal and Messier17).
The only two VLED RCT reported similar weight loss results(Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26) . Over the intervention period of 52 weeks, Bliddal et al. provided an alternating treatment of 810 kcal VLED (weeks 1–8 and 32–36) provided as a nutritional supplement dissolved in water (taken six times/d), and nutrition education to achieve a hypoenergetic diet of 1200 kcal/d (weeks 8–32 and 36–52)(Reference Bliddal, Leeds and Stigsgaard25). The control group attended five nutrition education sessions (baseline and weeks 8, 32, 36 and 52), with instruction to consume a hypoenergetic diet using usual foods (1200 kcal/d)(Reference Bliddal, Leeds and Stigsgaard25). The short-term (8-week) RCT by Christensen et al. compared a continuous VLED intervention (810 kcal/d) using a nutritional powder dissolved in water, with a healthy control group, who received nutrition education at baseline, and instruction to consume a hypoenergetic diet (1200 kcal/d)(Reference Christensen, Astrup and Bliddal26). Both Bliddal et al. and Christensen et al. reported a statistically significant weight loss of 11 kg (P < 0·01) in the treatment groups, compared with 4 kg in the control groups, despite the varying durations of 8 and 52 weeks of the intervention period(Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26) . Change in BMI was not reported in either of these studies(Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26) .
Performance-based physical function measures
Table 3 provides a summary of physical function outcomes including before and after treatment outcomes for performance-based and self-report physical function measures. A wide range of tests were used to measure performance-based physical function in the included studies; however, the most commonly used measures were 6-min walk distance (n 6), walk speed (n 3) and stair-climb time (n 3).
Of the lifestyle studies, six assessed one or more of the performance-based measures of interest(Reference Somers, Blumenthal and Guilak16,Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23,Reference Messier, Loeser and Miller28,Reference Messier, Loeser and Mitchell29,Reference Hughes, Tussing-Humphreys and Schiffer33) . Six-minute walk distance was measured in five lifestyle studies (two RCT(Reference Messier, Loeser and Miller28,Reference Hughes, Tussing-Humphreys and Schiffer33) and three uncontrolled studies(Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23,Reference Messier, Loeser and Mitchell29) ), of which the average increase in walk distance was 56·2 m (range: 9·7–125·8 m) in the treatment groups. Two RCT reported a statistically significant increase in 6-min walk distance in the diet and exercise groups, compared with controls(Reference Messier, Loeser and Miller28,Reference Hughes, Tussing-Humphreys and Schiffer33) . Two lifestyle studies measured stair-climb time reporting an average reduction in time by 1·6 s (range: −1·1 to −2·5 s) in the treatment groups(Reference Messier, Loeser and Miller28,Reference Messier, Loeser and Mitchell29) . Walk speed was measured in two lifestyle studies(Reference Somers, Blumenthal and Guilak16,Reference Paans, van den Akker-Scheek and Dilling23) . Paans et al. measured walk speed via the 20 m walk test, with the treatment group showing a significant increase in speed of 1·2 s (P < 0·05)(Reference Paans, van den Akker-Scheek and Dilling23), while Somers et al. measured gait speed at normal and fast paces, however reported only small non-significant changes in the weight loss group(Reference Somers, Blumenthal and Guilak16).
Miller et al.(Reference Miller, Nicklas and Davis30) and Messier et al.(Reference Messier, Mihalko and Legault9) were the only two DMR studies to measure performance-based physical function. Both RCT measured 6-min walk distance, reporting an average increase of 56·3 m (range: 26·0–72·8 m) which was statistically significantly greater than the control groups(Reference Messier, Mihalko and Legault9,Reference Miller, Nicklas and Davis30) . Miller et al. also reported a significant reduction in stair-climb time of 1·5 s in the treatment group, compared with the control group (P < 0·01)(Reference Miller, Nicklas and Davis30). Messier et al. reported statistically significant increase in walk speed of 0·1 m/s in both the diet-only and diet plus exercise group, compared with controls(Reference Messier, Mihalko and Legault9). Aaboe et al., the only VLED study to measure performance-based physical function, also reported a statistically significant increase in walk speed of 0·1 m/s (P = 0·02) following the 16-week intervention(Reference Aaboe, Bliddal and Messier17).
Self-reported functional status
The majority of studies measured physical function using self-report measures of interest, with the exception of one lifestyle study(Reference Messier, Loeser and Mitchell29) and one VLED study(Reference Aaboe, Bliddal and Messier17). The two most common tools used were the Knee Injury and Osteoarthritis Outcome Score (KOOS) and Western Ontario and McMaster Universities Arthritis Index (WOMAC) as summarised in Table 3.
Knee injury and osteoarthritis outcome score (KOOS)
The KOOS includes two physical function subscales, one that measures function in daily living (KOOS-FDL) and another that measures function in sport and recreation (KOOS-FSR). Each subscale is scored between 0 and 100, with higher scores indicating better function. Three uncontrolled studies measured change in physical function with KOOS, including one DMR study by Atukorala et al.(Reference Atukorala, Makovey and Lawler18) and two VLED studies by Bartels et al. and Riecke et al.(Reference Bartels, Christensen and Christensen19,Reference Riecke, Christensen and Christensen31) Following the 18-week diet-only intervention, Atukorala et al. reported a statistically significant improvement in KOOS function in daily living (KOOS-FDL) and KOOS function in sport and recreation (KOOS-FSR) score by 13·3 and 12·6, respectively (P < 0·01)(Reference Atukorala, Makovey and Lawler18). The VLED studies reported an average increase in KOOS-FDL by 11·5 (range: 11·0–12·5)(Reference Bartels, Christensen and Christensen19,Reference Riecke, Christensen and Christensen31) , while only Riecke et al. reported KOOS-FSR, with an average increase of 8·6 for the VLED and LED treatment groups(Reference Riecke, Christensen and Christensen31).
Western Ontario and McMaster universities arthritis index
The most commonly used physical function measure was the WOMAC function subscale which was reported in seven lifestyle studies(Reference Somers, Blumenthal and Guilak16,Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23,Reference Magrans-Courtney, Wilborn and Rasmussen27,Reference Messier, Loeser and Miller28,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) , four DMR studies(Reference Messier, Mihalko and Legault9,Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Miller, Nicklas and Davis30) and three VLED studies(Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26) . Scoring of the WOMAC was reported differently between studies, with two reporting on a scale of 0–1700(Reference Christensen, Astrup and Bliddal26,Reference Magrans-Courtney, Wilborn and Rasmussen27) , obtained using a visual analogue scale method, five reporting a scaled score of 0–100(Reference Somers, Blumenthal and Guilak16,Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Paans, van den Akker-Scheek and Dilling23–Reference Bliddal, Leeds and Stigsgaard25) , and six reporting on a scale of 0–68(Reference Messier, Mihalko and Legault9,Reference de Luis, Izaola and Garcia Alonso20,Reference Martin, Fontaine and Nicklas22,Reference Messier, Loeser and Miller28,Reference Miller, Nicklas and Davis30,Reference O’Brien, Wiggers and Williams32) , obtained using a Likert scale method. A lower score represents better function in all studies but one, by Paans et al.(Reference Paans, van den Akker-Scheek and Dilling23) For the uncontrolled lifestyle and DMR studies, an average improvement by 26·8(Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23) and 17·6 %, respectively, in WOMAC function score was reported(Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24) . The only uncontrolled VLED study that measured WOMAC function reports a significant improvement of 30·2 % (P < 0·01)(Reference Henriksen, Christensen and Danneskiold-Samsoe21).
Eight RCT were included in the meta-analysis that measured change in WOMAC physical function (thirteen independent comparisons)(Reference Messier, Mihalko and Legault9,Reference Somers, Blumenthal and Guilak16,Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26,Reference Messier, Loeser and Miller28,Reference Miller, Nicklas and Davis30,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) , which were categorised based on the intervention type. In the lifestyle studies (n 4)(Reference Somers, Blumenthal and Guilak16,Reference Messier, Loeser and Miller28,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) , the pooled effects analysis did not support an improvement in WOMAC physical function (effect size (ES) −5·3, 95 % CI −15·4, 4·7; P = 0·30) (Fig. 2). This result was following a statistically significant group mean difference in weight loss of 1·6 % (95 % CI −2·2, −1·0; P < 0·01). In contrast, the DMR RCT(Reference Messier, Mihalko and Legault9,Reference Miller, Nicklas and Davis30) showed a statistically significant mean difference between the treatment and control groups in WOMAC physical function of 12·4 % (95 % CI −21·5, −3·3; P < 0·01), based on 563 randomised participants (Fig. 3). This change followed a significant mean difference in weight loss of 8·5 % (95 % CI −10·2, −6·8; P < 0·01) between treatment and control groups. Likewise, the VLED RCT(Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26) demonstrated a significant mean difference between the control and intervention groups in WOMAC physical function of 12·5 % (95 % CI −24·5, −0·6; P = 0·04) favouring the VLED intervention group (Fig. 4). This result is based on 136 randomised participants in the VLED studies, following a statistically significant group mean difference in weight loss of 7·1 % (95 % CI −8·5, −5·8; P < 0·01).
Comparison of lifestyle intervention to control for % change in Western Ontario and McMaster Universities Arthritis Index (WOMAC) physical function
Comparison of diet and meal replacement (DMR) intervention to control for % change in Western Ontario and McMaster Universities Arthritis Index (WOMAC) physical function
Comparison of very low energy diet (VLED) to control for % change Western Ontario and McMaster Universities Arthritis Index (WOMAC) physical function
Discussion
The present systematic review provides a comprehensive summary of studies that have assessed the effect of weight loss, comparing different dietary interventions of varying dietary restriction (lifestyle, DMR and VLED), on physical function in adults with OA. Several measures of physical function were defined including participant self-report via questionnaires (e.g. WOMAC) or objective performance-based measures (e.g. 6-min walk test). Of the nineteen included studies most reported statistically significant weight loss and improvements in physical function (n 15). Overall, there were 4178 participants (1832 from RCT and 2346 from uncontrolled studies), the majority of whom had knee OA and only forty-six had hip OA, from two uncontrolled studies(Reference de Luis, Izaola and Garcia Alonso20,Reference Paans, van den Akker-Scheek and Dilling23) . At baseline, the average BMI was 34·0 kg/m2 (range: 31·7–38·0) in the lifestyle studies, 35·5 kg/m2 (range: 33·5–40·8) in the DMR studies and 36·4 kg/m2 (range: 35·2–37·5) in the VLED studies, placing the majority of participants in the obese class I and class II categories. Of the included studies, nine (47 %) were RCT, the majority of which were lifestyle studies (n 5), and the remaining ten (53 %) studies were uncontrolled trials.
The average weight loss was variable between the lifestyle, DMR and VLED studies. The lifestyle studies incorporated nutrition education, behavioural therapy and exercise, over a period of 8 weeks to 18 months, demonstrating an average weight loss of 4 % (range: 0–9 %) in the treatment groups(Reference Somers, Blumenthal and Guilak16,Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23,Reference Magrans-Courtney, Wilborn and Rasmussen27–Reference Messier, Loeser and Mitchell29,Reference O’Brien, Wiggers and Williams32,Reference Hughes, Tussing-Humphreys and Schiffer33) . In comparison, the DMR studies which incorporated partial use of meal replacements and nutrition education, with or without exercise, over 12 weeks to 18 months, reported an average weight loss of 9 % (range: 8–11 %)(Reference Messier, Mihalko and Legault9,Reference Atukorala, Makovey and Lawler18,Reference de Luis, Izaola and Garcia Alonso20,Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Miller, Nicklas and Davis30) . As expected, the VLED intervention studies demonstrated the largest average weight loss of 12 % (range: 11–13 %) for the treatment groups(Reference Aaboe, Bliddal and Messier17,Reference Bartels, Christensen and Christensen19,Reference Henriksen, Christensen and Danneskiold-Samsoe21,Reference Bliddal, Leeds and Stigsgaard25,Reference Christensen, Astrup and Bliddal26,Reference Riecke, Christensen and Christensen31) . The greater energy restriction of the VLED intervention (energy intake range: 415–810 kcal/d) explains the greater weight loss compared with the lifestyle and DMR studies. Previous reports of weight loss strategies are in agreement with that hypoenergetic diets incorporating partial meal replacements provides greater long-term weight loss (up to 12 months) than a conventional diet control(Reference Astbury, Piernas and Hartmann-Boyce37,38) . While the use of VLED may provide greater weight loss, clinical guidelines report that these are ineffective in the long-term, are nutritionally inadequate and should only be used for up to 12 weeks(39,Reference Raynor and Champagne40) .
The WOMAC function subscale was the most commonly reported physical function measure (n 14). In the lifestyle studies, average change in WOMAC function was 23 % (range: −5–41 %) in the treatment groups. While the DMR and VLED studies reported an average change in WOMAC function of 28 % (range: 17–42 %) and 29 % (range: 19–37 %) in the treatment groups, respectively. Meta-analysis did not demonstrate a statistically significant change in WOMAC function in the lifestyle studies (n 4). While a significantly greater improvement in WOMAC physical function by 12·4 and 12·5 % was detected in the DMR and VLED treatment groups compared with controls, with similar differences in weight changes (DMR: 8·5 % v. VLED: 7·1 %). It should be noted that only two studies were included in the VLED and DMR groups, respectively, which may limit the reproducibility of results. Previous studies have demonstrated that weight loss of ≥10 % (and up to 20 %) is conducive to better improvements in OA symptoms(Reference Atukorala, Makovey and Lawler18,Reference Messier, Resnik and Beavers41) . While the authors reported no safety concerns, weight loss of this magnitude should be carefully monitored, particularly in older adults at risk of loss of lean and bone mass, which can lead to risk of falls, fractures and disability(Reference Padilla Colon, Molina-Vicenty and Frontera-Rodriguez42).
The findings of our review are in agreement with earlier studies including systematic reviews that have reported improvements in physical function following weight loss(Reference Alrushud, Rushton and Kanavaki10–Reference Hall, Castelein and Wittoek12,Reference Christensen, Bartels and Astrup43) . Two of the more recent reviews looked at the effect of dietary weight loss, with or without exercise treatment, on physical function in adults with knee OA classified as overweight and obese(Reference Chu, Lim and Ng11,Reference Hall, Castelein and Wittoek12) . Pooled effects demonstrated statistically significant improvements in self-report physical function and disability following dietary weight loss(Reference Chu, Lim and Ng11,Reference Hall, Castelein and Wittoek12) . However, Hall et al. reported that these effects were not sustained in the long term (i.e. longer than 12 months)(Reference Hall, Castelein and Wittoek12). Our review demonstrates a more comprehensive summary of a larger sample of both controlled and uncontrolled dietary weight loss studies, which have not been included in previous reviews(Reference Somers, Blumenthal and Guilak16–Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24,Reference Magrans-Courtney, Wilborn and Rasmussen27,Reference Riecke, Christensen and Christensen31) . Further, we investigated the dietary modalities by which weight loss is induced, providing a comparison of three dietary weight loss interventions (lifestyle, DMR and VLED). Our findings show the change in physical function is similar between DMR and VLED weight loss interventions, and both DMR and VLED interventions are more effective than lifestyle programmes to induce significant change in physical function.
Quality of the evidence
The quality assessment highlighted limitations in study design of several of the included studies. A validated quality appraisal tool(14) was used to assess quality and risk of bias of included studies. Of the included studies, half (n 9) were of high-quality rating and low risk of bias, indicating a moderately positive quality of the review overall. The remaining nine studies received a neutral quality rating, as they were without a comparative study group (n 8)(Reference Aaboe, Bliddal and Messier17–Reference Lopez-Gomez, Izaola-Jauregui and Torres-Torres24) or had non-comparable groups at baseline (n 1)(Reference Somers, Blumenthal and Guilak16) potentially resulting in selection bias. The three CAROT sub-studies(Reference Aaboe, Bliddal and Messier17,Reference Bartels, Christensen and Christensen19,Reference Henriksen, Christensen and Danneskiold-Samsoe21) may be prone to a greater degree of bias, as research questions were not decided a priori. None of the included studies received a poor-quality rating. To be considered a high-quality study, participant retention rate is defined at 80 % or more(14) as this influences the interpretation of results. Four of the included studies did not reach this target, with retention rates of 63(Reference Martin, Fontaine and Nicklas22,Reference Bliddal, Leeds and Stigsgaard25) , 71(Reference Paans, van den Akker-Scheek and Dilling23) and 70 %(Reference Somers, Blumenthal and Guilak16), of which two were uncontrolled studies(Reference Martin, Fontaine and Nicklas22,Reference Paans, van den Akker-Scheek and Dilling23) . While RCT provide the highest level of evidence, poor retention rates need to be considered when interpreting results. As with many dietary intervention studies, there was considerable variability between studies in terms of methodology, including intervention components, length of follow-up and outcome measures, which may have affected the comparability of findings of included studies and interpretation of the results.
Clinical implications
This review is the first to provide a comprehensive summary and comparison of the different dietary interventions commonly used for weight loss and change in physical function in adults with OA classified as overweight and obese. Overall, the findings of the included studies indicate that significant weight loss is generally associated with an improvement in physical function, demonstrated by objective and subjective measures. While the VLED interventions demonstrated the highest mean weight loss of 12 %, followed by the DMR (9 % WL) and lifestyle interventions (4 % WL), both DMR and VLED groups showed similar improvements in physical function. This suggests that the less restrictive dietary intervention with partial use of meal replacements results in weight loss of clinical significance between 5 and 10 %(4) and the improvements in physical function are comparable to VLED interventions that are wholly reliant on the use of meal replacements. Therefore, DMR interventions may be the most optimal treatment for improved physical function and long-term weight management in adults with OA.
Acknowledgements
Acknowledgements: The authors thank Debbie Booth, Faculty Librarian, Health and Medicine, for her assistance with the database search strategy. Financial support: This work was supported by the Australian Government Research Training Program Scholarship, of which first author E.J.W. is a recipient. Conflict of interest: None. Authorship: E.J.W. and S.K.B. completed screening, data extraction and synthesis, and assessment of study quality. All authors were responsible for the study design and completed final reviewing and editing of the manuscript. Ethics of human subject participation: Not applicable.
Supplementary material
For supplementary material accompanying this paper visit https://doi.org/10.1017/S1368980020002529
