The Potential Cost-Effectiveness and Equity Impacts of Restricting Television Advertising of Unhealthy Food and Beverages to Australian Children

Television (TV) advertising of food and beverages high in fat, sugar and salt (HFSS) influences food preferences and consumption. Children from lower socioeconomic position (SEP) have higher exposure to TV advertising due to more time spent watching TV. This paper sought to estimate the cost-effectiveness of legislation to restrict HFSS TV advertising until 9:30 pm, and to examine how health benefits and healthcare cost-savings differ by SEP. Cost-effectiveness modelling was undertaken (i) at the population level, and (ii) by area-level SEP. A multi-state multiple-cohort lifetable model was used to estimate obesity-related health outcomes and healthcare cost-savings over the lifetime of the 2010 Australian population. Incremental cost-effectiveness ratios (ICERs) were reported, with assumptions tested through sensitivity analyses. An intervention restricting HFSS TV advertising would cost AUD5.9M (95% UI AUD5.8M–AUD7M), resulting in modelled reductions in energy intake (mean 115 kJ/day) and body mass index (BMI) (mean 0.352 kg/m2). The intervention is likely to be cost-saving, with 1.4 times higher total cost-savings and 1.5 times higher health benefits in the most disadvantaged socioeconomic group (17,512 HALYs saved (95% UI 10,372–25,155); total cost-savings AUD126.3M (95% UI AUD58.7M–196.9M) over the lifetime) compared to the least disadvantaged socioeconomic group (11,321 HALYs saved (95% UI 6812–15,679); total cost-savings AUD90.9M (95% UI AUD44.3M–136.3M)). Legislation to restrict HFSS TV advertising is likely to be cost-effective, with greater health benefits and healthcare cost-savings for children with low SEP.


Introduction
Childhood obesity is a significant public health issue worldwide [1]. Obesity in childhood is a risk factor for overweight and obesity in adulthood [2], resulting in both short-term and long-term negative health effects and highlighting the importance of obesity prevention for children and adolescents. Childhood and adolescent overweight and obesity also have a socioeconomic gradient. Youth with a low socioeconomic position (SEP) are at a greater risk of overweight and obesity compared to youth (iii) exploring thresholds for minimum BMI effect required to achieve cost-effectiveness. This will provide important information on the potential population level impact of the intervention, plus new information on the intervention's potential to reduce health inequities across socioeconomic groups.

Current Practice
Australian legislation currently prohibits TV advertising during 'P' classified programs [10]. TV networks must screen at least 260 h of Children's ('C') programs per year in 'C' time bands (7 am-8:30 am and 4 pm-8:30 pm weekdays, 7 am-8:30 pm weekends and school holidays). 'C' programs must not contain advertisements of more than 5 min in total per 30 min and content must not mislead or deceive children, put undue pressure on purchases or contain promotions by popular children's characters immediately before, during or after a 'C' or 'P' period [10].

The Intervention
The proposed intervention was defined as legislation to implement time-based restrictions of unhealthy food and beverage marketing to children under 16 years of age on FTA TV until 9:30 pm [28,29]. Peak TV viewing periods for Australian children are from 8 am-9 am and 7 pm-8 pm [30], and therefore legislation would reduce exposure to HFSS advertising. It should be noted that older adolescents and adults may also benefit from reduced exposure to HFSS advertising, however modelling has been undertaken in line with current policy recommendations (focusing on benefits for children under 16 years) [26,27]. Whilst children aged under five years may also be exposed to less HFSS TV advertising as a result of the intervention, this was considered less likely given their relatively young age and higher likelihood of watching 'P' programs. The intervention effect estimation was therefore limited to children aged between five and 15 years. We assumed that baseline viewing in the intervention population occurs during the time period of the proposed intervention on FTA TV, and tested assumptions around viewing patterns on FTA vs. non-FTA TV in sensitivity analyses.

Assessment of Benefit
A scoping review of the literature was conducted to inform the evidence of effect. Searches were conducted by one author (VB), using the Scopus and EBSCOHost academic databases. "Gold standard" evidence to inform modelling would consist of BMI effect estimates from reduced exposure to HFSS TV advertising from high-quality randomised controlled trials (RCTs) conducted in natural settings [19,23]. A search was conducted for such evidence (Supplementary Information 2), with no relevant studies identified in the literature.
Because of this absence of "gold standard" evidence, we defined the assessment of benefit of the intervention using a method previously used [19], whereby the effect is estimated based on the relationship between food promotion and consumption behaviours [16,[31][32][33]. The logic pathway of the intervention effect is given (Figure 1).
A scoping search was conducted for reviews and systematic reviews reporting the effects of TV advertising on consumption in children and adolescents ( Supplementary Information 2). Results of keyword searches were reviewed, with the meta-analysis by Boyland et al. [33] identified due to its immediate relevance and most recent publication date (2016). Boyland et al. [33] reviewed the evidence for an association between exposure to HFSS food advertising and energy intake and reported mean differences between exposure and control conditions. In order to model the effect of the intervention to BMI, estimates of a change in kilocalorie (kcal) (energy intake) per minute exposed was required. We conducted a meta-analysis of all studies in the review by Boyland et al. [33] that reported results for children that could be converted to kcal effect per minute exposed to TV advertisements using a random effects model. Heterogeneity was explored using Cochran's Q and I 2 . Full meta-analysis methods and the characteristics of included studies are reported in Supplementary Information 3. Studies included in the meta-analysis reported effects from acute exposure in controlled settings. Consumption under experimental conditions may not necessarily reflect consumption in natural settings, due to factors such as scrutiny, context and differences in time horizons and choice sets [34]. However, limited evidence on the external validity of such study findings exists. We therefore applied a crude adjustment factor for translation of experimental effects to the real-world, assuming that 50% of the effect would be maintained under non-experimental conditions (Table 1). Compensatory intake effect at mealtimes was also included, given evidence suggesting that children may compensate for snacks between meals by consuming less at mealtimes [35]. We therefore reduced the effect estimate, to conservatively estimate overall impact of exposure to HFSS TV advertisements on energy intake in children by assuming increased energy intake was a product of snacking between meals ( Table 1). The derived estimate of effect was multiplied by estimates of time per day spent watching HFSS TV ads by age and SEP (Table 1, Supplementary information 4). The change in mean energy intake as a result of the intervention was then converted to change in mean BMI using the validated equations by Hall et al. [36], with assumptions tested in sensitivity analyses.

Assessment of Costs
Fixed costs were estimated as the cost of passing legislation [37] (Table 1). Ongoing costs comprised the salary costs of two administration and compliance officers [21], including salary oncosts [38][39][40]. Potential loss of revenue to TV networks resulting from reduced advertising was tested in sensitivity analysis. From a societal perspective, the base case assumed that potential loss of revenue for the food and advertising industries would be compensated by revenue for the advertisement of other products or consumer expenditure on other goods and services. Data on healthcare costs were obtained from the Australian Institute of Health and Welfare (AIHW) for 2001 [41], and inflated to 2010 Australian dollar prices using the Health Price Index [42].

Cost-Effectiveness Modelling
Cost-effectiveness modelling was undertaken from a limited societal perspective, with the time horizon for estimating costs, cost-savings and health benefits being rest-of-life or 100 years for the 2010 Australian population. Costs, cost-savings and health benefits were discounted at 3% and were presented as 2010 values.
Modelling was undertaken as (i) a whole population analysis, and (ii) an analysis by area-level SEP, using the Socio-Economic Indexes for Areas (SEIFA) IRSD quintiles 1 (most disadvantaged) and 5 (least disadvantaged) [43]. A proportional multi-state, multiple cohort life table model was used to estimate obesity-related health outcomes and healthcare cost-savings. The model used data from the Australian Health Survey 2011-2012 [44] and disease epidemiology from the Global Burden of Studies included in the meta-analysis reported effects from acute exposure in controlled settings. Consumption under experimental conditions may not necessarily reflect consumption in natural settings, due to factors such as scrutiny, context and differences in time horizons and choice sets [34]. However, limited evidence on the external validity of such study findings exists. We therefore applied a crude adjustment factor for translation of experimental effects to the real-world, assuming that 50% of the effect would be maintained under non-experimental conditions (Table 1). Compensatory intake effect at mealtimes was also included, given evidence suggesting that children may compensate for snacks between meals by consuming less at mealtimes [35]. We therefore reduced the effect estimate, to conservatively estimate overall impact of exposure to HFSS TV advertisements on energy intake in children by assuming increased energy intake was a product of snacking between meals ( Table 1). The derived estimate of effect was multiplied by estimates of time per day spent watching HFSS TV ads by age and SEP (Table 1, Supplementary Information 4). The change in mean energy intake as a result of the intervention was then converted to change in mean BMI using the validated equations by Hall et al. [36], with assumptions tested in sensitivity analyses.

Assessment of Costs
Fixed costs were estimated as the cost of passing legislation [37] (Table 1). Ongoing costs comprised the salary costs of two administration and compliance officers [21], including salary on-costs [38][39][40]. Potential loss of revenue to TV networks resulting from reduced advertising was tested in sensitivity analysis. From a societal perspective, the base case assumed that potential loss of revenue for the food and advertising industries would be compensated by revenue for the advertisement of other products or consumer expenditure on other goods and services. Data on healthcare costs were obtained from the Australian Institute of Health and Welfare (AIHW) for 2001 [41], and inflated to 2010 Australian dollar prices using the Health Price Index [42].

Cost-Effectiveness Modelling
Cost-effectiveness modelling was undertaken from a limited societal perspective, with the time horizon for estimating costs, cost-savings and health benefits being rest-of-life or 100 years for the 2010 Australian population. Costs, cost-savings and health benefits were discounted at 3% and were presented as 2010 values.
Modelling was undertaken as (i) a whole population analysis, and (ii) an analysis by area-level SEP, using the Socio-Economic Indexes for Areas (SEIFA) IRSD quintiles 1 (most disadvantaged) and 5 (least disadvantaged) [43]. A proportional multi-state, multiple cohort life table model was used to estimate obesity-related health outcomes and healthcare cost-savings. The model used data from the Australian Health Survey 2011-2012 [44] and disease epidemiology from the Global Burden of Disease study [45]. Modelling by SEP group incorporated SEIFA quintile specific data, including disease incidence, mortality rate, BMI distribution and population number [37]. Potential impact fractions estimated the proportional reduction in disease incidence that would occur through reduced exposure to a risk factor as a result of an intervention. Disease-specific life tables estimated mortality and morbidity for nine obesity-related diseases (ischaemic heart disease, hypertensive heart disease, ischemic stroke, diabetes, colorectal cancer, kidney cancer, breast cancer, endometrial cancer and osteoarthritis). Interventions were compared against a "no intervention" comparator, where the distribution of BMI in the 2010 reference year Australian population remained unchanged [46].
All modelling was undertaken in Microsoft Excel 2016. Uncertainty analysis around key input parameters was estimated using Monte Carlo simulation (2000 iterations) using the Excel add-in Ersatz (version 1.35) [47] to estimate 95% uncertainty intervals (95% UI). Quality of life in children was incorporated using values from the literature [48]. Cohort-based modelling allowed disease-related benefits which are not present in childhood to be estimated due to lingering BMI effects. It was assumed that the BMI effect as a result of the intervention at age 15 years was maintained into adulthood, allowing the estimation of health benefits and healthcare cost-savings over the lifetime. Incremental cost-effectiveness ratios (ICERs) were calculated by dividing the difference in the net cost of the intervention by the difference in the net health benefit. ICER results were presented on a cost-effectiveness plane. The AUD50,000 per health adjusted life year (HALY) threshold was used to determine cost-effectiveness, as per Australian benchmarks [49]. Sampled from a pert distribution, minimum 1.5 maximum 6.5 from a 2012 Australian review of outcomes for studies that reported non-core TV advertising during children's peak viewing times (based on television audience patterns, generally weekday evenings and weekend mornings) [15]. Most likely 3.1 taken from Australian study 2017 [13].

Sensitivity Analyses
The base case assumes no loss of advertising revenue for broadcasters, given the likelihood of new advertising contracts to fill existing advertising periods. The effect of this assumption was tested in sensitivity analyses, assuming a short-term (one year) loss of revenue before networks recoup these costs through new advertising contracts (Table 1)

(Supplementary Information 5).
Multi-variate sensitivity analysis was undertaken by varying key input parameters in a "worst-case" analysis (Table 1). Worst-case analysis incorporated network loss of revenue, a smaller effect estimate derived from the meta-analysis (Supplementary Information 5) and greater adjustment for translation of effect from experimental to real-world settings. Worst-case analysis also varied the assumption that all TV viewing occurred on FTA TV. Pay TV and streamed services have gained popularity amongst Australian TV viewing audiences, although limited data exists on the proportion of children's viewing time using these services. A recent report by Deloitte Australia [54] cited that 22% of total viewing time is now via paid or streamed services, although this was based on a non-representative survey of people aged over 14 years. In lieu of more rigorous data, the worst-case analysis adjusted the mean minutes spent watching TV per day to assume that 22% of total viewing time was on streamed services with no exposure to HFSS advertising (for example, services like Netflix).

Threshold Analysis
Given the relative uncertainty around the evidence of effect, a threshold analysis was undertaken to determine the minimum effect size required to return a mean ICER less than the AUD50,000 per HALY threshold [49]. Analysis was undertaken using base case intervention costs, and additionally incorporating loss of short-term revenue to TV networks (Table 1).

Implementation Considerations
The broader impacts of an intervention that may not necessarily be captured within an economic evaluation should also be considered in order to take into account the wide range of factors important to decision-makers when setting priorities [55]. Cost-effectiveness results are discussed alongside implementation considerations, which qualitatively assess the strength of evidence, feasibility, acceptability and sustainability of the proposed intervention.

Results
Results suggest that an intervention restricting TV advertising of HFSS food and beverages to children would cost AUD5.9M (95% UI AUD5.8M-AUD7M). The intervention would result in a small mean decrease in energy intake (approximately 115 kJ/day) and a small mean BMI reduction (0.352 kg/m 2 ) at the population level ( Table 2). The cost-effectiveness modelling showed that the intervention would be dominant (i.e., cost-saving and health promoting), resulting in 88,396 HALYs saved (95% UI 54,559-123,199) and total cost-savings of AUD777.9M (95% UI AUD369.8M-AUD1.2B) at the population level over the lifetime (Table 2). Table 2. Cost-effectiveness results of restricting HFSS TV advertising.

Results
Children ( Importantly, the intervention may reduce health inequities, resulting in 1.5 times more HALYs and 1.4 times higher total cost-savings in children living in the most disadvantaged areas (Q1) compared to the least disadvantaged areas (Q5) ( Table 2). The cost-effectiveness planes in Figure 2 demonstrate the dominance of the intervention, at both the population level and for the most disadvantaged (Q1) and least disadvantaged (Q5) quintiles.  The intervention would also be dominant when short-term loss of revenue to TV networks is included (probability of: cost-effectiveness 100%, dominance 99.9%) (Supplementary information 5). Worst-case sensitivity analysis results also suggested that the intervention would remain dominant when varying input parameters to result in a much smaller effect estimate (population level mean The intervention would also be dominant when short-term loss of revenue to TV networks is included (probability of: cost-effectiveness 100%, dominance 99.9%) (Supplementary Information 5). Worst-case sensitivity analysis results also suggested that the intervention would remain dominant when varying input parameters to result in a much smaller effect estimate (population level mean BMI reduction 0.13 kg/m 2 , Supplementary Information 5), combined with a short-term loss of revenue to TV networks (probability of: cost-effectiveness 99.5%, dominance 83.5%).
Threshold analysis demonstrated that the effect size (expressed as the relationship between 'minutes of exposure to TV ads for HFSS food' and 'energy intake') could be reduced by more than two orders of magnitude (from a lower range estimate of 15.5 kcal per minute exposed, to <0.1 kcal per minute exposed) for the intervention to remain cost-effective. In this scenario, assuming base case intervention cost, a BMI reduction of 0.0004 kg/m 2 would result in a mean ICER under the AUD50,000 cost-effectiveness threshold (mean ICER AUD44,688 per HALY saved (95% UI AUD28,815-79,516)). Under the base case scenario for intervention benefits, short-term loss of revenue to TV networks could be more than seven times higher (~$700 million), and the intervention would remain dominant (mean ICER per HALY saved 95% UI dominant-AUD7215).

Discussion
An intervention restricting the exposure of Australian children to TV advertising of HFSS food and beverages is likely to be cost-saving over the lifetime of the cohort under the base case and all sensitivity analyses modelled, with greatest health benefits and cost-savings accrued by children living in the most disadvantaged areas (Q1) compared to children living in the least disadvantaged areas (Q5). To the best of our knowledge, this is the first time that an intervention restricting HFSS TV advertising to children has been modelled taking into account differences in advertising exposure by SEP.
Results suggest the significant potential of legislation to restrict HFSS TV advertising to children in addressing health inequities, in a cost-effective manner. Given that the differences in effect estimates by SEP were determined by differences in TV viewing time within our analysis, the differences in health benefits and total cost-savings by measure of area-level disadvantage may actually be under-estimates. However, if the content of what children are watching on TV also differs by SEP (for instance, FTA vs. paid/streamed services with varying levels of advertisement exposure), the differences in health benefits and total cost-savings between children with low SEP and children with high SEP may vary in other ways. More evidence quantifying the differential level and type of exposure of HFSS advertising to children by SEP is required to better inform this analysis. Further, this study only examined differences between SEP quintile one and five. Analysis across the entire socioeconomic spectrum is required, in order to gain a complete understanding of the equity-related effects of legislation to restrict HFSS TV advertising across all socioeconomic groups.
The most significant challenge to date remains the availability of rigorous evidence of effect of the intervention to better inform analysis (Table 3). While this study is based on strong program logic and the best available evidence of effect size, there is currently only limited evidence from real-world implementations of advertising restrictions. As more jurisdictions implement advertising restrictions (e.g., Chile, Canada), this will present increased opportunities to evaluate population-level impacts. Results from our threshold analyses demonstrate the small effect size (compared to our modelled effect estimates and meta-analyses based on previously published studies) required for the intervention to be considered cost-effective, further supporting the logic for further intervention in this area. Legislation restricting the exposure of children to TV advertising of HFSS foods and beverages is being implemented internationally (Table 3). Ireland has recently announced a comprehensive ban with a push for a 9 pm watershed on advertising of HFSS products on TV and radio, along with regulation across other advertising mediums. Whilst the Code of Practice has not yet been released by the Irish Department of Health, this demonstrates the feasibility of the intervention given the necessary political will. Table 3. Implementation considerations, intervention to restrict HFSS TV advertising to children.

Strength of evidence
Direct evidence of BMI effect of TV advertising of food and beverages HFSS from RCTs is currently not available. The intervention is modelled using an effect estimate derived from meta-analysis of non-naturalistic experimental evidence.

Acceptability
Food, media industry acceptability Likely to be low. Marketing and advertising drives sales. Low

Political acceptability
To date, political motivation to enact legislation has been low but may vary by political party and over time. International experience in countries such as Ireland and the United Kingdom suggests the potential for political acceptability.

Consumer acceptability
Public support for government regulation of advertising of HFSS food and beverages to children is high [56].

Feasibility
This legislative intervention is feasible to implement in the Australian setting. High

Sustainability
The intervention is sustainable once implemented. The ACMA already has regulatory responsibilities and can oversee the regulation of TV HFSS advertising.
The sustainability of potential BMI effect is unknown, and more evidence is required on the effects of TV advertising of HFSS food and beverages in adults.

Equity
Children with low SEP may have more exposure to HFSS TV advertising than children with high SEP, due to differences in TV viewing practices. Positive

Side effects
Positive side effects The intervention may have an impact on the food preferences and consumption behaviours of older children and adults.

Negative side effects
The intervention may result in loss of revenue to TV networks (likely to be short-term effect). The intervention may result in loss of revenue to food companies (although over the longer term it may be expected that companies adapt to market conditions).

Policy conclusion:
The intervention demonstrates significant potential for cost-effectiveness, positive equity effects and is feasible, sustainable and acceptable to the Australian general public. Our cost-effectiveness results are in keeping with previously published findings in Australia [21], The United States [23] and internationally [22], that interventions to restrict TV advertising of HFSS food and beverages are considered cost-effective ( Supplementary Information 1). Whilst results are not directly comparable due to methodological differences, there is now a growing body of evidence that the societal benefits of an intervention restricting the exposure of children to HFSS TV advertising outweigh the potential costs of legislation. In addition, our paper provides new evidence on the potential for cost-effectiveness of the intervention when incorporating what we consider to be realistic short-term loss of revenue to TV networks. In 2007, ACMA cited concerns regarding the impact of restricting food and beverage advertising to children on the production of children's TV programs by networks [57]. The inclusion of network costs, whilst relatively uncertain given the commercial sensitivity of trying to estimate loss of revenue for a hypothetical intervention, goes some way towards addressing this issue. Our analyses including loss of advertising revenue to networks demonstrates that the intervention remains cost-effective even when incorporating these costs ( Supplementary  Information 5), suggesting the potential for cost-effective government reimbursement if required.
By limiting the intervention population to children aged five to 15 years, we may have under-estimated the cost-effectiveness of the intervention at the population level. Studies have suggested that HFSS TV advertising may also have an effect on food-related behaviours and attitudes in older adolescents and adults, however at this time the evidence is inconclusive [33] and therefore we have not included potential health benefits and cost-savings accruing to older adolescents and adults as a result of the intervention into our analyses. Whilst this likely under-estimates cost-effectiveness, more evidence is required in order to capture the full public health benefits of legislation to reduce exposure to HFSS TV advertising across the entire child and adult population.
Limitations of our study include the assumption of maintenance of BMI effect into adulthood and the modelled effect estimate used for base case and sensitivity analyses, taken from meta-analysis of a small number of studies conducted in highly controlled experimental conditions. We have tried to compensate for this suspected bias by adjusting the effect estimate for a more plausible translation to a real-world setting, and for compensatory meal behaviours of children who snack. Adjustments were made using best-available evidence and logic pathways, however a more rigorous understanding of the effects of TV advertising on the consumption of HFSS food and beverages in children is clearly an area for significant future research.
It is likely that the food industry would oppose the introduction of more stringent TV advertising legislation, and this is recognised as a potential barrier to the implementation of this public health policy [58]. Whilst our analysis took a limited societal perspective, we have not included loss of revenue to the food industry as a result of the intervention. We have instead assumed that any loss of food industry revenue from a reduction in demand for HFSS food and beverages would be compensated in increased demand for non-discretionary foods or other items, and therefore doesn't result in a loss to society. However, this is yet to be rigorously and definitively tested. In addition to its effect on consumption, advertising is also designed to influence market share within a category, and the impact of advertising restrictions on individual companies is not analysed in this paper.
Finally, television advertising is only one medium used for marketing of HFSS food and beverages, and although traditional TV viewing is still the most dominant promotional channel, it may be waning [16]. Changes to technology have led to the development and implementation of comprehensive approaches to marketing focused on brand and relationship development, including advertising across mediums, sponsorship, product placement, brand mascots and celebrity endorsement, sales promotion, labelling and point-of-purchase displays [5]. This analysis did not take into account potential shifts in advertising from TV to other media in response to the intervention. Ideally this intervention would be combined with more comprehensive marketing restrictions across multiple forms of media, as well as form part of a broader obesity prevention strategy across health and non-health sectors (for example, fiscal, built environment, education interventions) [28,29]. Results from this study form part of a broader body of work into obesity prevention priority-setting in Australia, using standardised methods [37,46,59] and designed to better inform obesity prevention policy. Other interventions found to be cost-saving to date include sugar-sweetened beverage (SSB) taxation [37] and reformulation or package size caps of SSBs [46].

Conclusions
This study indicates the significant potential of legislation restricting TV advertising of HFSS food and beverages to Australian children as a cost-effective intervention, with greatest benefits for the most socioeconomically disadvantaged. All modelled scenarios and sensitivity analyses resulted in dominance (i.e., an intervention that is both cost-saving and improves long term health outcomes). Our analysis by measure of area-level socioeconomic disadvantage demonstrates greater health benefits and total cost-savings in those living in the most disadvantaged socioeconomic areas compared to those living in the least disadvantaged areas. Furthermore, results from our threshold analysis demonstrate the very small population level effect size required for the intervention to be considered cost-effective. Author Contributions: V.B. designed the study and conducted the meta-analysis. J.L.V. and A.L. contributed to model development. All authors reviewed model parameters and inputs. All authors reviewed input parameters for economic modelling. V.B. conducted modelling, and wrote first draft of the paper. All authors provided comments on the paper.