What is the clinical effectiveness and cost-effectiveness of using drugs in treating obese patients in primary care? A systematic review

Prevalence

The increasing prevalence of obesity in the UK represents a considerable public health problem. The prevalence of obesity [defined as a body mass index (BMI) ≥ 30 kg/m²] in England is reported to have increased between 1993 and 2004 from 13.6% to 24.0% among men and from 16.9% to 24.4% among women. 1 When waist circumference was measured in a UK adult primary care sample in 2005, 38.8% of men and 51.2% of women were classed as abdominally obese (waist circumference > 102 cm and > 88 cm, respectively). 2 It has been estimated that, among young people aged 20 years and under in England, 10% of females and 8% of males are obese. 3 Should increases in the prevalence of obesity continue at the same rate, Zaninotto et al. 1 predicted that the prevalence of obesity in 2012 would be 32% in men and 31% in women, with a likely higher prevalence among adults in manual social classes (43%) than in non-manual social classes (35%). Projections by the UK government’s Foresight programme have postulated that by 2025 40% of Britons may be classed as obese. 3,4

The World Health Organization (WHO) estimated in 2005 that, internationally, there were over 1.6 billion overweight adults, of whom at least 400 million were obese. They also projected that, by 2015, approximately 2.3 billion adults would be overweight and over 700 million would be obese. 5

The estimated prevalence of overweight and obesity among male and female adults in 2010 demonstrated considerable differences by geographical region, with several hotspots of prevalence exceeding 80%, including the USA, Barbados, Dominica, Kuwait and the South Pacific islands. 5

Groups at risk of obesity

A number of population groups are considered at increased risk of obesity. Variation in obesity by ethnic group has been described in a report published by the NHS Information Centre. 6 Data relating to obesity and overweight among ethnic minority groups were drawn from the Health Survey for England (HSE) 2004. The survey applied the definition of overweight and obesity as used in the general population. It was reported that the prevalence of obesity was higher among black Caribbean men and women, black African women, Pakistani women and Irish men than among the general population. Obesity was lower among Chinese, Indian and Bangladeshi men and women than among the general population. Groups at risk of becoming obese also include children with overweight or obese parents7–9 and individuals giving up smoking. 10 A high prevalence of obesity has been observed in adults and children with intellectual disabilities. 11–13 An association also exists between low socioeconomic status in early life and adult obesity. 9 Data from the HSE 2007 indicated that, among women, the age-standardised prevalence of obesity and raised waist circumference increased as the quintile of equivalised household income decreased, but these measures were not related to income in men. 6

Aetiology

Previous work by the Foresight group indicated that the energy imbalance that precedes obesity (whereby energy intake exceeds energy expenditure) is governed by what was described as a ‘complex multifaceted system of determinants’. 14–16 These factors include biological propensity (such as genetic risk and the influence of early life experiences), the generation of an obesogenic external environment (based on, for example, changes in food production and lifestyle, such as increased wealth, increased sedentary lifestyle and increased availability of energy-dense foods), a life course component (whereby the risks of becoming obese may be present at an early stage of life) and a generational dimension (in which parental obesity is known to act as a significant predictor of childhood obesity). 17

Comorbidities associated with obesity

Overweight and obesity are associated with a significant range of comorbidities, including type 2 diabetes (T2DM), hypertension, dyslipidaemia, coronary artery disease, stroke, osteoarthritis, reproductive problems, respiratory and liver conditions and cancers. 18–20 Obstructive sleep apnoea is also associated with obesity, with a potential predisposition among Asian individuals. 21,22 The National Audit Office23 estimated the increased relative risk of the development of comorbidities among obese individuals, which is shown in Table 1.

Increased levels of overweight and obesity are linked with increases in mortality, with subjects who have never smoked and with no history of disease but a BMI > 40 kg/m² having a relative risk of death 2.7 times higher for men and 1.9 times higher for women than that among subjects with a lower BMI (between 23.5 and 24.9 kg/m²). 24 Obesity is also associated with psychological stigma. 25 The proportion of chronic disease attributable to obesity has been predicted to undergo a considerable increase by 2050, particularly for T2DM, stroke and coronary heart disease (CHD). 3 It has been suggested that adults in the upper half of the healthy weight category (22.0 kg/m² < BMI < 24.9 kg/m²) are more likely to develop health problems than their leaner counterparts and that adults should attempt to maintain a BMI of between 18.5 kg/m² and 21.9 kg/m² to minimise their risk of disease. 20

Measurements of obesity

Obesity is frequently reported in terms of BMI. BMI is a measurement of body weight relative to height. Based on the WHO criteria, overweight is classed as a BMI of 25–29.9 kg/m², while obesity is defined as a BMI > 30 kg/m². 5 The current National Institute for Health and Clinical Excellence (NICE) clinical guideline for obesity26 states that waist circumference may also be used in addition to BMI in adults with a BMI < 35 kg/m² and may be used to provide additional information on the risk of the development of comorbidities in children. Among adults, waist circumference can be used as an indicator of health risk, with increased risk being identified based on a waist circumference of ≥ 94 cm in men and ≥ 80 cm in women and greatly increased risk with a waist circumference of ≥ 102 cm in men and ≥ 88 cm in women. 27 Other measurements of obesity include body weight, percentage over ideal body weight, waist–hip ratio and skinfold thickness. It is worth noting that a lower cut-off point has been suggested for certain ethnic groups including South Asians. 28

A report suggested that debate surrounded the use of standard BMI cut-offs among some ethnic groups on the basis that variation exists in the association between BMI and body fat according to ethnicity. 6 Dhaliwal and Welborn29 and Kumar et al. 30 proposed that waist–hip ratio be used as a measure of central obesity because of its high precision and no bias across ethnic groups. 29,30

Impact of health problem and significance for the NHS

Most obesity management is undertaken in primary care settings. Hospital admissions for people with obesity-related conditions place a significant burden on the health service, particularly in relation to circulatory diseases, musculoskeletal disorders and endocrine disorders including diabetes. 31

Allender and Rayner32 produced a new estimate of the burden of overweight- and obesity-related disease in the UK. The authors estimated that, when the rates for the burden of overweight-attributable disease were applied to mortality figures for 2003–4, over 203,000 deaths occurred in the UK as a result of diseases associated with overweight and obesity. The authors stated that it was further estimated that approximately 66,737 deaths were directly attributable to overweight and obesity, over half (54%) of these being due to CHD and 31% to stroke.

Management of obesity

The primary aim of the management of obesity is to achieve weight reduction in the interests of health. The Royal College of Physicians33 described the clinical benefits of weight loss (based on a scenario of an individual weighing 100 kg losing 10% of their body weight), estimates of which included decreased blood pressure, a 10% decrease in cholesterol, a > 50% reduction in the risk of developing diabetes, reductions of 30–40% and 40–50% in diabetes-related deaths and obesity-related cancer deaths, respectively, and a 20–25% reduction in total mortality. Non-pharmacological methods for the management of obesity include dietary modification, exercise, structured education and weight management programmes. For obese patients who cannot achieve or maintain a healthy weight by non-pharmacological means, a number of pharmacological interventions exist to aid weight reduction, including sibutramine (Reductil^®, Abbott), orlistat (Xenical^®, Roche; Alli^®, GlaxoSmithKline) and rimonabant (Acomplia^®, Sanofi-Aventis). Drug therapy has been shown to be most effective when combined with dietary modification, physical exercise and behaviour change34 and is recommended for use in the management of obesity in combination with non-pharmacological interventions. Surgical procedures, such as gastric bypass and banding, also play a role in the management of obesity.

Current service cost

The treatment of obesity and its complications is associated with significant health and social care costs, in addition to wider societal financial costs. The House of Commons Select Committee estimated that the direct health-care costs arising from the treatment of obesity and its complications ranged from £991M to £1124M in 2002. This level of expenditure represented approximately 2.3–2.6% of the total NHS spending for the period 2001–2. 35 Allender and Rayner32 estimated the direct cost of overweight and obesity to the NHS to be £3.2B, of which the greatest proportion was attributable to stroke (£983M), followed by CHD (£773M), hypertensive disease (£576M) and diabetes (£533M). The costs arising from overweight and obesity are likely to escalate (from the estimate for 2007 of £4.2B) in the forthcoming years if current increasing trends in the prevalence of obesity continue, with a predicted overall annual total cost to the NHS of overweight and obesity of £9.7B (based on today’s prices) by 2050, representing an increase in the projected percentage of NHS costs (at £70B) from 6.0% in 2007 to 13.9% in 2050. 3

Variation in services and/or uncertainty about best practice

The management of obesity in primary care has been described as being uncoordinated and inconsistent. 36 The Counterweight Project Team (2004) undertook a series of structured interviews with general practitioners (GPs) and practice nurses and analysed patient records from primary care settings across England and Scotland in order to investigate the range of approaches to obesity management utilised by primary care professionals. Although the majority of GPs (83%) and practice nurses (97%) reported that they would raise weight as an issue with obese patients, only 15% of GPs would spend up to 10 minutes in a weight-related consultation compared with 76% of nurses (p < 0.001). BMI was recorded for 64.2% of patients. GPs and practice nurses reported making patient referrals to a dietician (58% vs 59%), exercise referral schemes (50% vs 56%) and commercial weight loss agencies (41% vs 68%). Audit of obese patients’ records showed the use of practice-based diet counselling (20%), dietetic (4%) and obesity centre (1%) referrals and any anti-obesity medication (2%) recorded over 18 months. Patients prescribed anti-obesity medication were more likely to be female (p < 0.01) and more obese (p < 0.01) than, but with a similar prevalence of comorbidities to, patients who were not prescribed medication.

Relevant national guidelines

Healthy Weight, Healthy Lives is a cross-government strategy for England involving a range of programmes across a number of sectors, including schools and food, physical activity, transport and the health service. 17 The strategy is focused on five areas: the healthy growth and development of children; promoting healthier food choices; promoting physical activity; creating incentives for better health; and personalised advice and support. The development of strategies for the management of obesity is also linked to requirements under the national service frameworks for CHD and diabetes.

NICE issued clinical guideline 4326 to provide guidance on the prevention, identification, assessment and management of overweight and obese adults and children. The guidance superseded previous pieces of guidance on orlistat,38 sibutramine39 and surgery for morbid obesity. 40 The clinical guideline recommended that dietary changes and physical exercise should be the first options in the management of obesity before the use of pharmacological interventions is considered. Bariatric surgery was recommended if all of the following criteria were fulfilled: a BMI of ≥ 40 kg/m² (or between 35 kg/m² and 40 kg/m² in the presence of other significant disease that could be improved in the event of weight loss); all appropriate non-surgical measures have been attempted and been unsuccessful; person has or will receive intensive management in a specialist obesity service; patient is generally fit for anaesthesia and surgery; and the patient commits to requirement for long-term follow-up. In addition, bariatric surgery can be considered as a first-line option when appropriate in adults who have a BMI of ≥ 50 kg/m². Surgical intervention was not generally recommended for children or young people. In 2008, NICE guidance was issued relating to the use of rimonabant. 41 However, the marketing authorisation for this drug has since been suspended. 42

The NHS Health Checks Programme was launched in April 2009. Designed to address health inequalities, the vascular risk assessment programme consists of systematic screening of individuals aged 40–74 years of age for cardiovascular and T2DM risk with lifestyle interventions offered to those considered to be at risk. 43

Summary of interventions

Orlistat functions by inhibiting the uptake of dietary fats by the gastrointestinal tract, whereas both sibutramine and rimonabant are centrally acting appetite suppressants. The following sections summarise the product characteristics of each of these interventions using the Summary of Product Characteristics (SPC) for each drug (obtained from the Electronic Medicine Compendium at www.medicines.org.uk; SPC not available for rimonabant) and information from the British National Formulary (BNF).

Current usage in the NHS

The NICE clinical guideline for obesity26 recommended that dietary changes and physical exercise should be the first options in the management of obesity. The use of pharmacological interventions for weight loss was not generally recommended for children younger than 12 years but the guideline stated that such measures may be used in exceptional circumstances (e.g. the presence of severe comorbidities). In children aged ≥ 12 years, treatment with orlistat or sibutramine was recommended only in the presence of severe physical or psychological comorbidities. In adults, it was recommended that orlistat be prescribed as part of an overall obesity management plan in patients with a BMI of ≥ 28.0 kg/m² (with associated risk factors) or ≥ 30.0 kg/m². It was recommended that orlistat therapy should be continued beyond 3 months only if the patient had lost at least 5% of his or her initial body weight since commencing therapy. Treatment with orlistat beyond 12 months should be made after discussing the potential benefits and limitations with the patient. The guideline also recommended that sibutramine be prescribed as part of a weight reduction plan in patients meeting one of the following criteria: a BMI of ≥ 27.0 kg/m² (with associated risk factors) or a BMI of ≥ 30.0 kg/m², with careful monitoring of weight loss and adverse events. Therapy with sibutramine was to be continued beyond 3 months only if the patient had lost at least 5% of initial body weight while taking the drug. Treatment with sibutramine was not recommended beyond the licensed duration of 12 months. Co-prescribing of pharmacological interventions for weight reduction was not recommended. As noted above, the marketing authorisations for rimonabant and sibutramine have been suspended following reviews by the European Medicines Agency. 42,44

Anticipated costs associated with intervention

Using the latest data available,47 Figure 1 shows the number of items prescribed annually from 1999 to 2008 in the treatment of obesity in England. There was a substantial increase in prescribing rates for both orlistat and sibutramine following publication of guidance from NICE in March 200138 and October 2001,39 respectively. After a period of relatively steady use, prescription numbers started to increase again after the publication of revised guidance in 2004. 48

FIGURE 1.

Annual number of prescription items for the treatment of obesity.

Table 2 shows the number of items and associated net ingredient cost (NIC) of drugs for the treatment of obesity prescribed in primary care. Rimonabant became available on prescription in July 2006; thus, the figure for that year reflects just 6 months of data. In 2008, there were 1.28 million prescription items for the treatment of obesity. Overall, the total number of prescription items in 2008 was ten times the number in 1999, and the current trend is an increase of around 14% per year.

The total NIC for drugs for the treatment of obesity increased from £4.9M in 1999 to £51.6M in 2007, but fell in 2008 to £44.8M. Correspondingly, the NIC per item increased from £38 in 1999 to £42 in 2007 and then fell to £35 in 2008.

Following the withdrawal of rimonabant in 2008 and the suspension of sibutramine prescribing in 2010, it is reasonable to expect that the uptake of orlistat will increase as patients switch treatments and the alternatives available for new patients decrease. As orlistat was already the main treatment, accounting for two-thirds of all prescriptions, and the NICs per item for sibutramine (£30) and rimonabant (£50) are similar to that for orlistat (£35), it is not expected that this change in prescribing patterns will affect the observed trend in total costs. If total prescribing rates for orlistat increase at 14% (5–20%) per annum, the total annual net cost for prescriptions directly related to obesity treatment is estimated to be approximately £57.8M in 2010 and over £109M in 2015.

Literature search

The following electronic databases were searched. Searches were carried out in January 2009. Examples of the search strategies used are given in Appendix 1. Where completed trials were yet to be published, we the contacted the principal investigator.

• MEDLINE

• MEDLINE In-Process & Other Non-Indexed Citations

• Cumulative Index to Nursing and Allied Health Literature (CINAHL)

• Health Technology Assessment Database

• EMBASE

• Cochrane Database of Systematic Reviews

• Database of Abstracts of Reviews of Effects (DARE)

• NHS Economic Evaluation Database (NHS EED)

• Cochrane Controlled Trials Register

• Web of Science Proceedings

• Science Citation Index

• Current Controlled Trials

• BIOSIS.

Management of references

The results of the literature search were imported into EndNote reference managing software version X5 (Thomson Reuters, CA, USA). Duplicates were removed. Where multiple papers reported data from the same study, these were grouped together and only included once in the analysis.

Inclusion and exclusion of studies

Studies were excluded if they were not reported in English.

Study design

Only studies using a RCT design were included. Both parallel and crossover designs were included. Non-randomised studies were excluded.

Patient population

Studies with adults who were overweight or obese or at high cardiovascular disease (CVD) risk were included. High CVD risk was defined as having one or more of the following conditions: hypertension, T2DM, gestational diabetes, polycystic ovary syndrome, high cholesterol, metabolic syndrome, angina, coronary artery disease and non-alcoholic steatohepatitis. Studies were excluded if they included those with mental illness, for example binge eating, or if they included children or adolescents.

Interventions

Studies were included if they compared orlistat, sibutramine or rimonabant with lifestyle and/or exercise advice (standard care), placebo or metformin. The anti-obesity treatments had to be given at the recommended dose: orlistat 120 mg three times daily (maximum 360 mg daily), sibutramine 10 mg, 15 mg or 10 mg increasing to 15 mg once daily, rimonabant 20 mg once daily. Studies were also included if they gave orlistat and sibutramine in combination or if orlistat, sibutramine or rimonabant were given in combination with other active interventions. We excluded trials with a treatment period of < 12 weeks. We also included head-to-head comparisons of the pharmacological agents.

Outcome measures

Trials had to include one or more of the following outcome measures, measured at 3, 6 or 12 months:

• weight change from baseline

• BMI change from baseline

• number losing ≥ 5% body weight

• number losing ≥ 10% body weight.

Assessing relevancy of included studies

The titles and abstracts of all studies identified by the electronic searches were screened for inclusion by two reviewers, who each assessed half of the identified articles. The full texts of all studies found to be potentially relevant were sought and were assessed for inclusion by two independent reviewers, using the inclusion criteria outlined above. Disagreements were discussed with the project steering committee.

Data extraction

Data were extracted using a standard form by one reviewer. See Appendix 2 for the data extraction form.

The data extracted included author, year of publication, country, population included (diabetic, with comorbidities, obese but otherwise healthy, or other), trial design, treatment length, follow-up length, study quality, interventions used, level of lifestyle/exercise advice, whether or not a wash-in period was used, if so how long and whether or not it used an active intervention, and baseline characteristics by group.

Lifestyle and/or exercise advice was categorised using the following criteria:

• standard – one visit with general dietary/exercise advice given or patients given a lifestyle leaflet

• enhanced – more than one visit or patient given more than just advice.

Because of the poor reporting of the lifestyle components of the interventions we assumed that standard advice was given if lifestyle and/or exercise advice was not mentioned, as this is standard care for overweight and obese patients. We also assumed that if diet (or exercise) advice had been given this also included advice on exercise (or diet) and therefore did not extract data on diet and exercise separately.

Outcome data were extracted in a number of formats depending on how the data were presented: data could be presented either at the arm level or as trial-level differences. For the continuous outcomes (weight and BMI change) the following data were extracted:

• arm based (data given for each intervention):

  1. – mean weight at baseline and follow-up with standard deviation (SD), standard error or significance levels and confidence intervals (CIs)

  2. – mean change from baseline with SD, standard error or significance levels and Cis

  3. – mean change from baseline adjusted for baseline value (ANCOVA) with SD, standard error or significance levels and CIs.

• trial based (data given as difference between interventions)

  1. – mean difference between interventions at follow-up with SD, standard error or significance levels and CIs.

For the binary outcomes (5% and 10% weight loss) the number achieving the target was extracted for each intervention. For all outcomes the number of participants included was also extracted.

Where possible, data from intention-to-treat analyses were extracted. If data were presented by subgroup only (e.g. data for those with and without hypertension given separately) then these were meta-analysed to give the results for the entire study population. Data were extracted only from either the manuscript text or tables (no attempt was made to extract from figures). Where data were incomplete the corresponding author of the study was contacted.

Quality assessment

All studies were assessed for quality. The quality tool used was based on that developed by Jadad et al. 49 with the addition of a score for allocation concealment, as suggested in Schulz et al. 47 The tool is described in Table 3.

Pair-wise meta-analysis

Studies were pooled using random-effects models for each treatment comparison for which data were available for each of the outcomes outlined in Outcome measures at the time points 3, 6 and 12 months. Random-effects models were used as studies were expected to be heterogeneous. Heterogeneity was assessed using the I² and χ² statistics. Forest plots were constructed for all comparisons. All analysis was carried out in Stata version 11.0 (StatCorp LP, College Station, TX, USA).

Mixed-treatment comparison

Mixed-treatment comparison methods were used to compare all treatments under investigation for obesity and their comparators within a single model, which allowed us to make both direct and indirect comparisons (where no head-to-head trials are available). 50 Initially, all outcome measures and time points (as described for the pair-wise meta-analyses above) were checked to make sure they formed closed networks. Placebo was used as the reference category throughout. We used a logistic regression model for the binary outcomes and a linear regression model for the continuous outcomes. In all cases a burn-in of 10,000 simulations was discarded and the results are presented based on a further 40,000 simulations. Convergence was checked visually using the history plots. The goodness of fit was checked using the residual deviance. Vague priors were used for all parameters. For each treatment the percentage of times that treatment gained the highest rank across all of the simulations was also calculated. All MTC analysis was conducted using a Bayesian Markov chain Monte Carlo method using the Bayesian software WinBUGS version 1.4.3. (MRC Biostatistics Unit, Cambridge, UK). 51

We compared the results of the pair-wise meta-analysis with the MTC and defined these as inconsistent when the MTC estimate did not fall within the 95% CI from the pair-wise meta-analysis.

Covariate analysis

For the 12-month weight change outcome we also considered exploring the effect of two covariates on the treatment effect: the proportion of participants with T2DM and the level of lifestyle advice given. Each covariate was modelled separately. The treatment covariate interactions were modelled as a separate regression coefficient for each treatment.

Sensitivity analysis

We carried out a sensitivity analysis for the 12-month weight change outcome according to the following variables: intention to treat (excluding those in which intention to treat had not been used or it was not clear which method had been used) and wash-in (excluding those studies that had used a wash-in phase). A wash-in is defined as a pretrial practice whereby all eligible patients are given an intervention (either active or placebo) for a period to test compliance; only those who comply are then randomised into the trial.

Publication bias

Publication bias was assessed visually using contour-enhanced funnel plots for all comparisons that contained five or more studies.

Study selection

Figure 2 shows the flow of studies. The electronic searches identified 3183 potentially relevant articles. After removing duplicates and those that were not eligible after reading the title and abstract, 161 full texts were assessed. Of these, 67 were excluded (see Appendix 3, Reference list). Overall, 94 studies were included in the meta-analysis. Orlistat was assessed in 54 studies, 44 studies included sibutramine and five studies assessed rimonabant.

FIGURE 2.

Flow diagram.

Study characteristics

The majority of trials were carried out in North America and Europe from 1995 to 2008 (Table 4). Overall, 24,808 individuals were included. The mean trial size was 264, ranging from 14 to 3277, and 55 trials (58.5%) included ≥ 100 participants. Two crossover trials were included, with all other trials having a parallel design. The mean length of intervention was 8.3 months (range 3–48 months). A total of 45 studies (47.9%) used enhanced lifestyle advice, with the remainder giving standard advice (49, 52.1%). A wash-in period was used in 44 (46.8%) of the studies.

Risk of bias within studies

The results of the bias assessment are given in Table 4. Overall, the quality of the studies included is generally low. All included studies were randomised but in only 40% of studies was the randomisation procedure described fully and adequately. In total, 22 (23.4%) studies concealed allocation. The majority of studies were double blind: 50% mentioned double blinding and 20% described their blinding adequately. Participant flow was not described in 17 (18.1%) studies.

Results of individual studies

Individual study results are given in Table 5. In total, 83 trials included data on weight change, 41 on BMI change and 45 and 36 on 5% and 10% body weight loss respectively. A total of 33 trials measured outcome at 3 months, 38 at 6 months and 35 at 1 year.

At baseline, participants had an average age of 45.5 years (SD 6.97 years), 25.7% were male, 33.2% were diabetic and the mean BMI was 34.92 kg/m² (SD 2.58 kg/m²).

Results of the evidence synthesis

Mixed-treatment comparison meta-analysis

Table 6 shows a summary of the results of the MTC meta-analysis, with placebo as the reference group. Figure 3 shows the network diagrams for each of the outcomes considered. Figure 16 in Appendix 3 shows the percentage best plots for each treatment at each time point (this is the percentage of simulations in which each treatment option came out as having the largest treatment effect).

TABLE 6 - Mixed-treatment comparison model results unadjusted for covariates

Outcome	Treatment	3 months			6 months			12 months
		OR	95% CrI	% best ranking	OR	95% CrI	% best ranking	OR	95% CrI	% best ranking
5% weight loss	Placebo	Reference		0	Reference		0	Reference		0
	Orlistat	3.86	0.06 to 15.11	2.5	2.95	1.62 to 4.97	0.1	2.89	2.22 to 3.72	2.0
	Sibutramine 10 mg	5.87	1.46 to 17.65	16.0	4.24	2.39 to 6.84	2.2	3.25	1.56 to 6.22	16.1
	Sibutramine 15 mg	9.95	3.10 to 32.71	81.4	6.90	3.49 to 12.99	21.5	4.06	2.51 to 6.29	47.3
	Rimonabant	–	–	–	–	–	–	3.78	2.39 to 5.79	34.5
	Orlistat and sibutramine	–	–	–	16.99	2.45 to 62.01	76.1	–	–	–
	Standard care	0.90	0.05 to 4.02	0.01	0.39	0.10 to 1.04	0	1.01	0.42 to 2.06	0
10% weight loss	Placebo	Reference		4.2	Reference		0	Reference		0
	Orlistat	–	–	–	3.10	1.44 to 6.14	0	2.43	1.72 to 3.39	0.3
	Sibutramine 10 mg	16.41	0.34 to 93.23	55.1	6.57	3.28 to 12.97	1.3	3.38	1.39 to 7.13	10.7
	Sibutramine 15 mg	14.34	0.28 to 77.43	40.7	18.83	6.70 to 48.1	54.3	5.02	2.63 to 9.12	57.9
	Orlistat and sibutramine	–	–	–	22.96	2.82 to 88.08	43.3	–	–	–
	Rimonabant	–	–	–	–	–	–	4.25	2.35 to 7.31	31.2
	Standard care	–	–	–	2.76	0.21 to 12.1	1.0	–	–	–
Outcome	Treament	Mean difference	95% CrI	% best ranking	Mean difference	95% CrI	% best ranking	Mean difference	95% CrI	% best ranking
Weight change	Placebo	Reference		0	Reference		0	Reference		0
	Orlistat	–2.65	–4.00 to –1.31	0	–3.08	–4.20 to –2.03	0	–4.12	–5.07 to 3.15	0.2
	Metformin	–4.63	–7.46 to –1.68	0	–3.15	–6.51 to 0.29	0.4	–	–	–
	Sibutramine 10mg	–4.88	–6.40 to –3.43	0	–5.08	–6.55 to –3.62	1.3	–5.42	–7.36 to –3.42	16.6
	Sibutramine 15mg	–5.37	–6.59 to –4.10	0	–6.11	–8.11 to –4.23	5.3	–6.35	–8.06 to –4.63	78.2
	Rimonabant	–11.23	–17.17 to –5.15	62.6	–	–	–	–4.55	–6.20 to –2.92	5.0
	Orlistat and sibutramine	–10.18	–13.82 to –6.59	37.7	–9.67	–14.32 to –5.04	93.1	–	–	–
	Standard care	–1.36	–3.23 to 0.48	0	–1.95	–3.83 to –0.11	0	–2.89	–4.90 to –0.84	0
BMI change	Placebo	Reference		0	Reference		0.1	Reference	–	0
	Orlistat	–1.56	–2.54 to –0.58	0	–0.59	–2.60 to 1.39	7.5	–1.43	–2.67 to –0.18	1.8
	Metformin	–3.50	–5.02 to –1.87	0.6	0.18	–4.05 to 4.37	13.2	–	–	–
	Sibutramine 10 mg	–2.43	–3.33 to –1.54	0.1	–0.95	–2.89 to 1.02	16.4	–2.27	–5.08 to 0.59	33.0
	Sibutramine 15 mg	–2.25	–2.97 to –1.54	0	–1.81	–4.25 to 0.61	60.6	–2.91	–5.45 to –0.62	62.4
	Rimonabant	–6.24	–9.07 to –3.33	95.5	–	–	–	–	–	–
	Orlistat and sibutramine	–3.16	–5.22 to –1.11	3.7	–	–	–	–	–	–
	Standard care	0.50	–0.51 to 1.48	0	0.79	–2.04 to 3.72	1.3	–1.02	–3.11 to 1.12	2.8

FIGURE 3.

Network diagrams, showing the number of direct comparisons for each MTC analysis: (a) 5% weight loss 3 months, (b) 5% weight loss 6 months, (c) 5% weight loss 12 months, (d) 10% weight loss 3 months, (e) 10% weight loss 6 months, (f) 10% weight loss 12 months, (g) weight change 3 months, (h) weight change 6 months, (i) weight change 12 months, (j) BMI change 3 months, (k) BMI change 6 months, (l) BMI change 12 months.

5% weight loss

3 months

Sibutramine 15 mg had the largest probability (81.4%) of being the best intervention in terms of having a 5% body weight loss, with a 10-fold odds of having a good outcome compared with placebo [OR 9.95, 95% credible interval (CrI) 3.10 to 32.71]. No statistically significant difference was found between orlistat and placebo and standard care and placebo.

6 months

At 6 months the combination of orlistat and sibutramine was the most efficacious intervention compared with placebo (OR 16.99, CrI 2.45 to 62.01). No difference was found between standard care and placebo. Orlistat, sibutramine 15 mg and sibutramine 10 mg were all more beneficial than placebo.

12 months

Compared with placebo, similar results were seen at 12 months to those at 3 and 6 months. With the addition of rimonabant, all active treatments were significantly better than placebo, with no difference seen between placebo and standard care.

10% weight loss

3 months

Only one three-arm trial reported data on 10% weight loss at 3 months. No statistically significant differences between placebo and sibutramine 10 mg or sibutramine 15 mg were seen.

6 months

The active treatments (orlistat, sibutramine 10 mg, sibutramine 15 mg and a combination of orlistat and sibutramine) were all superior to placebo. No statistically significant difference was seen between placebo and standard care (OR 2.76, 95% CrI 0.21 to 12.1).

12 months

Comparable results to those seen for 5% weight loss at 12 months were seen for 10% weight loss at this time point. Sibutramine 15 mg had the largest probability (57.9%) of being the best intervention, followed by rimonabant (31.2%) and sibutramine 10 mg (10.7%).

Weight change from baseline

3 months

In comparison with placebo, the active drugs were all associated with greater weight loss, ranging from a mean change of –2.65 kg (95% CrI –4.00 kg to –1.31 kg) for orlistat to –11.23 kg (95% CrI –17.17 kg to –5.15 kg) for rimonabant. No statistically significant difference was seen between placebo and standard care.

6 months

As previously seen, all active treatments were associated with greater weight loss than placebo, with the combination of orlistat and sibutramine producing the greatest weight loss (–9.67 kg compared with placebo) and the largest probability (93.1%) of being the best intervention. Standard care was also more effective than placebo at 6 months (–1.95 kg, 95% CrI –3.83 kg to –0.11 kg).

12 months

All treatments (orlistat, sibutramine 10 mg and 15 mg, and rimonabant) and standard care were superior to placebo at 12 months, with mean weight change in comparison with placebo ranging from –6.35 kg (sibutramine 15 mg) to –2.89 kg (standard care).

Body mass index change from baseline

3 months

In line with the results for weight change at 3 months, all active treatments were associated with significant reductions in BMI compared with placebo. No difference was seen between standard care and placebo (0.50 kg/m², 95% CrI –0.51 kg/m² to 1.48 kg/m²).

6 months

At 6 months none of the active treatments or standard care was significantly better than placebo. For example, metformin was associated with an increase in BMI compared with placebo (0.18 kg/m², 95% CrI –4.05 kg/m² to 4.37 kg/m²).

12 months

At 12 months orlistat and sibutramine 15 mg resulted in significantly greater BMI loss than placebo. No difference was seen between sibutramine 10 mg and placebo or standard care. Sibutramine 15 mg had the largest probability (62.4%) of being the best intervention in terms of BMI loss.

Comparison with pair-wise meta-analysis

The pair-wise and MTC analyses generally agreed for the two binary outcomes – 5% and 10% body weight loss. Across all outcomes, where inconsistency was seen the intervention effect size was generally in the same direction but with a different magnitude. For the continuous outcomes there was more disagreement. For example, for weight change at 3 months, there were 15 pair-wise comparisons, eight of which were consistent with the MTC results and seven of which were not, with the MTC results showing greater weight loss for the drug interventions than the pair-wise results (Table 7a–l).