SUPPORTING INFORMATION

The cost-effectiveness of shopping to a predetermined grocery list to reduce overweight and obesity

Summary of systematic review

Background

Much has been made of the potential benefits of strategies drawn from behavioural economics such as pre-commitment in addressing the obesity epidemic.1 However, little is known about effectiveness of such strategies in influencing grocery purchasing patterns, body-weight and obesity-related illness among overweight and obese individuals.

Objectives

To assess the effectiveness of pre-commitment strategies in influencing food purchasing patterns, body-weight and obesity-related illness among overweight individuals.

Data sources

Searches of the following citation databases were conducted in June 2011 using the closest possible approximation of the search strategy described in Table S1 for each database.

  • ISI Web of Knowledge: Science Citation Index Expanded 1899 to 29 June 2011, Social Sciences Citation Index 1898 to 29 June 2011, Arts & Humanities Citation Index 1976 to 29 June 2011;
  • OVID: EBM Reviews – Cochrane Central Register of Controlled trials, 2nd Quarter 2011, EBM Reviews – Cochrane Database of Systematic Reviews 2005 to June 2011, EBM Reviews – Cochrane Methodology Register 2nd Quarter 2011, EBM Reviews – Health Technology Assessment 2nd Quarter 2011, PsychINFO, 1987 to June Week 4, 2011, MEDLINE, 1948 to 29 June 2011;
  • Informit: Australian Public Affairs-FT, 1995 to 29 June 2011, Health Collection, 1977 to 29 June 2011, Health & Society: 1980 to 29 June 2011;
  • CSA Illumina: ERIC, 1966 to 30 June 2011, International bibliography of the social sciences, 1951 to 29 June 2011; and
  • EBSCO host: Business Source Complete, 1886 to 29 June 2011, EconLit, 1886 to 29 June 2011

We also searched forward citations of all included studies using Google Scholar and hand-searched backward citations in reference lists of included studies.

Study eligibility criteria

Participants:adultsor childrenof above normalweight.

Interventions:interventions that seek to modify the mode (e.g. in-person versus on-line), frequency, duration,product selection strategy or providerthat subjects utilise for grocery shopping via a pre-commitment mechanism either with or without an adjuvant intervention.Interventions that relied on other mechanisms of action such as changes in the relative price of ‘healthy’ foods were excluded.

Outcomes: outcomes of interest included purchasing volumes or patterns, consumption volume or pattern, weight, BMI, skin-folds, body composition, risk factors for obesity or obesity-related illness, disease status or severity, mortality.

Study design: health technology assessments, systematic reviews, meta-analyses, randomised controlled trials, controlled clinical trials, comparative studies and cohort studies were eligible for inclusion. Protocols, studies in which the ascertainment of study subjects has not been described, editorials, letters, case series, case reports and narrative reviews were excluded.

Characteristics of the publication (date, language, specific journals):we excluded studies published in languages other than English and studies published outside the date ranges specified in the search strategy and in journals not indexed in the databases specified in the search strategy.

Data extraction

We designed a data extraction form suitable for use as both an electronic and paper form. Two of the named authors (DM & GM) independently piloted the data extraction form in the first five included studies selected for inclusion. The data extraction form was revised and updated during piloting in response to identified problems. For each included study, we extracted publication characteristics (citation, publication type, study aims/objectives, experimental design, recruitment and randomisation procedures), participant characteristics (number of participants per group, baseline demographics by group, baseline weight/BMI by group, baseline food environment by group, blinding), and outcomes (outcomes measured, blinding, follow-up, outcomes reported, participants observed, participants analysed, results, incomplete data). We also extracted detailed information regarding the design and delivery of the intervention and control conditions; with the objective of describing 'who did what to whom, when, where and how often?'.2 Two authors (DM & PG) independently completed a data extraction form for all studies meeting the inclusion criteria.

Measures of treatment effect

For continuous outcomes such as consumption volumes and weight gain, we expressed treatment effects as the difference of group means. None of the included studies reported on dichotomous outcomes such as obesity-related disease status.

Study appraisal

Included studies were assessed at the study level for a 'low risk’, ‘high risk’ or ‘unclear risk’ of material bias with respect to seven domains: random sequence generation; allocation concealment; blinding of participants and personnel; blinding of outcome assessment; incomplete outcome data; selective reporting; and other sources of bias using the Cochrane Collaboration’s revised tool.3While it was our intention to assess risk of bias across studies that were homogeneous with respect to type of intervention and outcome, the review failed to identify multiple studies for any intervention type.

Synthesis methods

Studies were grouped according to intervention type and separate analyses were conducted for each type of intervention. While it was our intention to meta-analyse studies that were homogeneous with respect to type of intervention and outcome, the review failed to identify multiple studies for any intervention type. We provide a narrative summary of the evidence across types of intervention.

Results

Figure S1 summarises the identification and selection of studies. Two papers were identified as meeting the inclusion criteria.4,5 Wing et al.4 investigated (among other interventions) the impact of shopping to a list among 163 overweight women from the United States (aged 15-55). Participants were randomly assigned to 6 months of SBT or SBT combined with provision of detailed meal plans and corresponding shopping lists (SBT+List). Gorin et al.5 studied the effect of using online grocery shopping to modify the household food environment among 28 overweight participants from the United States (aged 21-65, 89% female). Participants were randomly assigned to 8-weeks of standard behavioural weight loss therapy (WLT) or WLT combined with online purchase and home delivery of groceries (Home Delivery). Wing et al. detected a statistically significant treatment effect in favour of SBT+List with respect to the number of foods in the home including low fat meats, low calorie entrees and fruit/vegetables. Similarly, Gorin et al. detected a statistically significant treatment effect in favour of Home Delivery with respect to total food items in the home and the number of high-fat food items in the home. While both studies also measured weight loss, only Wing et al. observed a significant treatment effect (in favour of SBT+List) with respect to weight loss. The Wing et al. trial was assessed as having a high (blinding of outcome assessment for dietary intake and food stored in the home) or unclear risk of bias (random sequence generation; allocation concealment; blinding of outcome assessment for weight) across several domains. The Gorin et al. trial was assessed as having a high (blinding of outcome assessment for food availability; incomplete outcome data for weight and for food availability) or unclear risk of bias (random sequence generation; allocation concealment; blinding of outcome assessment for weight) across several domains.

Limitations

It is possible that widening the search strategy to achieve greater coverage of the grey literature and relaxing the study selection criteria to include studies published in languages other than English may have yielded some additional studies.

Conclusions

There is limitedevidence of a treatment effect with respect to food availability in overweight, adult females when Home Delivery is added to WLT as compared to WLT alone. There is limitedevidence of a treatment effect with respect to food availability and weight loss in overweight, adult females when ‘shopping to a list’ is added to SBT as compared to SBT alone.

Table S1.Medline search strategy.

Database: Ovid MEDLINE(R)
Unless otherwise stated, search terms are free text terms; MESH=Medical subject heading (MEDLINE medical index term); tw=text word; pt=publication type; sh=MESH; adj=adjacent; ti=title; ab=abstract.
MEDLINE search terms will be amended as required for use in each database included in our search strategy.
Search strategy:
------
  1. shop$.tw.
  2. grocer$.tw
  3. supermarket.tw
  4. 1 or 2 or 3
  5. weight.tw
  6. obesity.tw
  7. overweight.tw
  8. body mass index.sh
  9. 5 or 6 or 7 or 8
  10. consum$.tw
  11. diet$.tw
  12. nutrition.tw
  13. (purchas$ or buy$ or bought).tw
  14. 10 or 11 or 12 or 13
  15. 4 and 9 and 14

Figure S1. PRISMA Flow diagram for identifying studies for inclusion in analysis.

1

Figure S2. Extrapolated average change in BMI for participants in shopping to a list pre-commitment intervention.

This figure shows the changes in BMI from the trial (weeks 0 to 78) and the extrapolated change in BMI from 78 weeks post baseline. Markers indicate values as reported in Wing et al. Extrapolation assumes 0.39kg weight gain per month on week 26 weight in the SBT group and 0.43kg weight gain per month on week 26 weight in the SBT+List group.

1

Figure S3. Markov model.

1

Cost-effectiveness modelling methods

The Markov model

The Markov model used in the analysis is shown in Figure S3. Ellipses represent health states, with the arrows demonstrating the possible transitions within the model. The loops on the health states signify the possibility of remaining in that state on a subsequent cycle. All individuals entered the model in the pre-intervention weight state, free of co-morbidities. The dotted transition leading from post-intervention weight back to pre-intervention weight represents gradual weight regain over multiple cycles. Both groups regained weight until they reached the baseline BMI of the intervention group.

Transition probabilities

As data on disease risk and mortality rates were not collected alongside the trial, multiple published studies were used to source data on transition probabilities. The following section details the estimates and the respective data sources.

Relative risks

The relative risks of contracting relevant co-morbidities per one unit increase in BMI are presented in Table S2. The estimates are for both genders because the risks for males and females were found to not significantly differ in the source studies. Where available, age-specific relative risks are reported and these were used for the starting age of the cohort in the Markov model.

The relative risks were based on estimates presented in the Global Comparative Quantification of Health Risks6 and estimates used in a cost-effectiveness analysis by Forster et al. (Appendix 2, Table 1).7 The colorectal cancer estimate was sourced from a meta-analysis of six studies (4 studies from the USA, 1 study from Sweden and 1 study from Australia).8 Relative risks for osteoarthritis by unit increases in BMI are rare and as such there was only one study (in the USA) which provided the basis for this estimate.9 The CHD and Stroke estimates were sourced from the only meta-analysis to provide relative risks for unit increments of BMI.10 The Asia-Pacific Cohort Studies Collaboration (APCSC) included data from 33 studies in the Asia Pacific region. 10% of the studies were from Japan, 15% from mainland China, 55% from other parts of Asia (Singapore, Taiwan, Hong Kong and Republic of Korea) and 20% from Australian and New Zealand. The type 2 diabetes estimates were sourced from a meta-analysis of the only 3 cohort studies analysed by APCSC that had repeated measures of diabetes and baseline BMI. These studies were from Australia, New Zealand and Japan.

Table S2. Relative risks (RR) of disease per 1 unit increase of BMI.

Age / RR / Source populations / Source years
Colorectal cancer / 35+ / 1.03 (1.01-1.05) / USA, Sweden, Australia / 1950-1986a
Osteoarthritis / 35+ / 1.04 (1.03-1.06) / USA / 1988-1994
CHD / 35-44 / 1.12 (1.05-1.19) / Asia-Pacific region / 1961-1992a
45-59 / 1.10 (1.08-1.14)
60-69 / 1.06 (1.03-1.08)
Stroke / 35-44 / 1.14 (1.05-1.23) / Asia-Pacific region / 1961-1992a
45-59 / 1.10 (1.03-1.16)
60-69 / 1.08 (1.03-1.13)
Type II Diabetes / 35-44 / 1.19 (1.06-1.32) / Australia, N.Z., Japan / 1975-1992a
45-69 / 1.14 (1.05-1.23)
70+ / 1.10 (1.03-1.16)

95% confidence intervals shown in parentheses ().

a Range of start years across the multiple studies

Baseline risks

Relative risks were applied to baseline incidence rates in order to derive transition probabilities to each disease state dependent on BMI. Baseline incidence rates were required to be BMI specific, thus they were linked to the BMI of the study population from which they were calculated, or, where this was not possible, linked to the mean population BMI for the country of the study population. Baseline estimates and transition probabilities are detailed in Table S3.

1

Table S3. Baseline incidence rates and transition probabilities.

Gender / Incidence per 1000 persons per year / Incidence per person per 6 months^ (SE) / Source study / Source
population / Source age (yrs) / Source year
From weight state (BMI=27)a
Colorectal cancer / Male / 0.54 / 0.0003 (2.0x10-6)* / 11 / United Kingdom / ≥15 / 1996
Female / 0.37 / 0.0002 (1.4x10-6)*
Osteoarthritis / Both / 25.00 / 0.0126 (9.5 x10-5)* / 12 / United Kingdom / ≥55 / 1991-1996
CHD / Male / 4.14 / 0.0021 (1.6x10-5)* / 13 / United Kingdom / 25-74 / 1996-1998
Female / 1.47 / 0.0007 (5.6x10-6)*
Stroke / Both / 1.42 / 0.0007 (5.4x10-6) / 14 / United Kingdom / ≥15 / 1996
Type II Diabetes / Male / 4.86 / 0.0024 (1.5x10-5) / 15 / United Kingdom / 10-79 / 1996-2005
Female / 3.66 / 0.0018 (1.3x10-5)
From diabetes state
Colorectal cancer / Both / 0.98 / 0.0005 (1.6x10-5)# / 16 / United States / 30-73 / 1974-1994
CHD / Male / 36.30 / 0.0183 (0.00058)# / 17 / Finland / 45-77 / 1982-1996
Female / 31.60 / 0.0159 (0.00049)#
Stroke / Male / 10.82 / 0.0054 (0.00017) / 18 / United Kingdom / ≥35 / 1992-1999
Female / 13.16 / 0.0066 (0.00020)

^ Converted to 6 month risks using the formula: 6 month risk = 1-[1-(annual risk)]1/2. Transition probabilities from each weight state (BMI=28 to 33) were first calculated as annual rates (using annual relative risks and annual incidence risks) and then converted to 6 month risks.

* Standard error estimated by taking the ratio of the standard error to the mean for the incidence from overweight to stroke.

a The same incidence rates used for transition probabilities from osteoarthritis state to colorectal cancer, CHD, Stroke and Type 2 diabetes states

# Standard error estimated by taking the ratio of the standard error to the mean for the incidence from diabetes to stroke.

Note: unless otherwise stated, standard errors were taken from the source study.

1

Mortality rates

Mortality rates, representing transition to obesity-related death, are documented in Table S4. Age-specific mortality rates were not available by cause of death and as such, the same rates for disease-related death were applied throughout the model. Mortality rates for stroke and type 2 diabetes are representative of the UK population, whereas other rates were taken from studies based on smaller regions within the UK. The WHO Monica Project defines CHD case fatality as the proportion of deaths within 28 days of a coronary event,19 thus in the absence of fatality statistics over the 6 month period it was assumed that an individual who survived the initial 28 days did not die of CHD in the remainder of the 6 month period. Probability of transition to the absorbing state of unrelated death was assumed constant across disease states, but to vary according to the age of the cohort throughout the model. Data was based on life-tables for the UK obtained from the Office of National Statistics.20 Overweight-specific transitions were not employed to avoid double counting.

Table S4. Mortality rates

Gender / Mortality per person per 6 months (CI)^ / Source
study / Source
population / Source age (yrs) / Source year
Colorectal cancer* / Male / 0.151 (0.137-0.166) / 21 / United Kingdom / ≥15 / 2004-2006
Female / 0.157 (0.146-0.169)
CHD / Male / 0.483 (0.457-0.507) / 19 / United Kingdom / 35-64 / 1985-1994
Female / 0.464 (0.421-0.507)
Stroke / Male / 0.118 (0.107-0.129) / 22 / United Kingdom / 70 (14.9)# / 2005
Female / 0.159 (0.148-0.171) / 75 (14.8)#
Type II Diabetes / Male / 0.031 (0.030-0.031) / 23 / United Kingdom / ≥35 / 1992-1999
Female / 0.031 (0.030-0.031)

^ Risks for colorectal cancer, stroke and type II diabetes were obtained as annual figures and converted to six month risks using the formula: 6 month risk = 1-[1-(annual risk)]1/2. 95% confidence intervals shown in parentheses ().

* mortality rates for colon cancer only. Confidence interval not reported – estimated by taking the standard deviation to mean ratio for stroke mortality rate.

# mean (SD) age

Note: unless otherwise stated, confidence intervals were taken from the source study.

Costs

Table S5. Costs per patient per 6 month intervention.
Type of costs / Time per session / price / Cost for 26 sessions / Cost per patient^ / Source
SBT: 26 sessions
Therapist session / 1 hour / £73.00/hour / £1,898 / £94.90 / 24
Therapist preparation / 30 mins / £36.00/hour / £468 / £23.40 / 24
Calorie/fat guide book / £6.67/book / £6.67 / Amazon.com
SBT total / £124.97
Dietician - meal plan / grocery list preparation / 30 mins / £25.00/hour / £325 / £16.25 / 24
SBT + LIST total / £141.22

^ Assumes 20 patients per session as achieved in the trial.

Costs are in £ sterling, 2010 prices.

Note: in sensitivity analyses, SBT total cost ranged from £62.49 (best case) to £249.94 (worst case) and SBT+List total cost ranged from £70.61 (best case) to £282.44(worst case).

Table S6. Costs of disease states.
Disease State / 6 month cost / Source study / Source population
Overweight or obese* / Male / £80.92 / 25 / United Kingdom
Female / £103.18
Colorectal cancer^ / £3,613.86 / 26 / United Kingdom
Osteoarthritis# / £423.62 / 27 / Australia
CHD / £838.91 / 28 / United Kingdom
Stroke / Initial year / £9,095.08 / 29 / United Kingdom
Subsequent year / £1,222.26
Type II Diabetes / £757.00 / 28 / United Kingdom

* Includes cost of statins treatment and monitoring costs, based on proportion of obese persons with elevated cholesterol.30

^Unweighted average of the costs of laparoscopic and open resection including chemotherapy and radiotherapy.

# Australian cost converted to British Pounds using the GDP purchasing power parity for 2005.

Costs are in £ sterling, 2010 prices.

Note: in sensitivity analyses, all 6 month disease state costs were doubled (best case) and halved (worst case).

Utilities

The utility weights used in the model are presented in Table S7. These were derived by taking utility decrements associated with each health condition from a reference utility estimate of normal weight individuals in the UK population.31 Utilities for overweight and obese states were calculated by subtracting from the reference utility estimate, the utility decrements associated with overweight and obesity, estimated by Jia and Lubetkin.32 These utility decrements were calculated, adjusting for the main comorbidities associated with an elevated BMI and are therefore assumed to be utilities of overweight/obese individuals free from comorbidities. Although the utility decrements are for the US population, they have been found to be comparable to estimates for the UK population.33 Utilities associated with obesity-related disease states included in the model (type 2 diabetes, colorectal cancer, osteoarthritis, coronary heart disease (CHD) and stoke) were derived by subtracting from the utility estimates for overweight and obesity, the utility decrements associated with each disease determined by Ara and Brazier.34 Due to the interrelationship between comorbidities related to overweight and obesity, it was assumed that the utility decrements reflected individuals who were affected by the health condition and any other health condition (taken from Table A2 of Ara and Brazier).