Potential of trans fats policies to reduce socioeconomic inequalities in mortality from coronary heart disease in England: cost effectiveness modelling study

Objectives To determine health and equity benefits and cost effectiveness of policies to reduce or eliminate trans fatty acids from processed foods, compared with consumption remaining at most recent levels in England. Design Epidemiological modelling study. Setting Data from National Diet and Nutrition Survey, Low Income Diet and Nutrition Survey, Office of National Statistics, and health economic data from other published studies Participants Adults aged ≥25, stratified by fifths of socioeconomic circumstance. Interventions Total ban on trans fatty acids in processed foods; improved labelling of trans fatty acids; bans on trans fatty acids in restaurants and takeaways. Main outcome measures Deaths from coronary heart disease prevented or postponed; life years gained; quality adjusted life years gained. Policy costs to government and industry; policy savings from reductions in direct healthcare, informal care, and productivity loss. Results A total ban on trans fatty acids in processed foods might prevent or postpone about 7200 deaths (2.6%) from coronary heart disease from 2015-20 and reduce inequality in mortality from coronary heart disease by about 3000 deaths (15%). Policies to improve labelling or simply remove trans fatty acids from restaurants/fast food could save between 1800 (0.7%) and 3500 (1.3%) deaths from coronary heart disease and reduce inequalities by 600 (3%) to 1500 (7%) deaths, thus making them at best half as effective. A total ban would have the greatest net cost savings of about £265m (€361m, $415m) excluding reformulation costs, or £64m if substantial reformulation costs are incurred outside the normal cycle. Conclusions A regulatory policy to eliminate trans fatty acids from processed foods in England would be the most effective and equitable policy option. Intermediate policies would also be beneficial. Simply continuing to rely on industry to voluntary reformulate products, however, could have negative health and economic outcomes.

. Potential Ruminant TFA categories are: Milk and milk products, Butter, Eggs and egg dishes, Meat and meat products, Fish and fish dishes. Within each of these potential ruminant categories, there still may be non-ruminant TFA. For example, the category Milk and milk products includes ice cream and dairy desserts, which can have TFA. The meat and fish categories include fried or breaded dishes, which can have TFA. The figures below are thus likely a slightly over-estimate of ruminant TFA. Therefore we used 0.4%E as the average ruminant TFA consumption. Since the differences by age and gender are not large, we used rounded values of 0.7%E for the average total TFA consumption for IMDQ3 and 0.4%E for the ruminant TFA consumption of all IMDQs.  Table B: TFA consumption from Low Income Diet and Nutrition Survey [16], which we treated as IMDQ5. As above (Table A), the average across age and gender is similar, therefore we used 1.3% as a simple rounded average. Since this data is slightly older than in Table A, we lowered the average here to 1.2%. It is not possible to calculate the ruminant TFA consumption in exactly the same way due to differences in the categories reported. It seems likely that individuals with low socioeconomic status would be more likely to purchase low cost / low quality products that contain industrial TFA (e.g. cheap ice cream, cheap ready meals) whereas those with higher socioeconomic status would be more likely to purchase the equivalent product without processed TFA (e.g. whole cuts of meat There are no data on food consumption outside the home in the UK. To approximate the consumption of TFA in restaurant food or fast food, we used food expenditure data (Table C). These UK data are very similar to recent data about the proportion of calories consumed outside the home in the United States, both being around 30%. These data from the USA can be found here: http://www.health.gov/dietaryguidelines/data-table-1.asp. Though we cannot necessarily conclude that the data from the USA are applicable to the UK, it is at least reassuring that they are similar. . We assumed that the two lowest income deciles were equivalent to IMDQ5, and so on. For modelling a restaurant ban, we assumed that industrial TFA consumption was proportional to the food expenditure away from home (e.g. 22% of industrial TFA consumed by IMDQ5 was at restaurants). We did a sensitivity analysis by running the reverse gradient (e.g. 40% of industrial TFA consumed by IMDQ5), which we refer to as the "fast food" scenario due to the tendency for unhealthy take-away establishments to be clustered in deprived neighbourhoods (Please find the information at the following from National We treated the meta-analysis relationship between TFA consumption and CHD incidence [4] as loglinear. A 23% increase in CHD incidence per 2%E from TFA is mathematically: log(1.23) / 2%E = 0.104 per 1%E. Then for a certain reduction in TFA consumption (e.g. a 0.3%E reduction), we calculated the reduction in CHD incidence as: 1-exp(0.104 per 1%E x TFA reduction) = e.g., 1-exp(0.104 x 0.3) = -0.032 (this is a 3.2% reduction).
Age-specific effects are not provided in the meta-analysis. Therefore we assumed that age-specific values would follow the same relationship as those linking cholesterol to CHD reported by the Prospective Studies Collaboration (2007) and adapted for the IMPACT-SEC model by Bajekal et al. (2012) [2], since cholesterol is the primary pathway through which TFA increase CHD risk. We assumed there would be no further stratification in the effect by gender or by IMDQ. If we consider the age group 65-74 as an example, the reduction in CHD incidence per 1%E from TFA would be: log(1.12) / 2E% = 0.057 per 1E%. Again following the above paragraph, a 0.3%E reduction in TFA would result in a reduction of 1.7% in CHD incidence ( 1 -exp(0.057 * 0.3) = -0.017 ). Age-specific TFA effect on CHD incidence per 2%E from TFA (product of above two rows) +33% +33% +23% +16% +12% +12% +11% We converted the proportional reductions in CHD incidence by first assuming that CHD mortality would reduce by the same amount, e.g. 1.7% reduction in incidence would lead to a 1.7% reduction in mortality. Then to convert the percentage reduction to absolute numbers, we used a projection of future CHD mortality by age group, gender and IMDQ. This projection was done using a Bayesian analysis of an age-period-cohort model which has been previously applied to England & Wales [55] and stratified by IMDQ for England [3].We continue with the example above, now illustrating that Men ages 65-74 in IMDQ5 are predicted in 2020 are predicted to have CHD mortality of 379 per 100,000. Reducing this by 1.7% as above would represent 6 CHD DPP per 100,000 in 2020.
The rates of CHD DPP (per 100,000) were converted to absolute numbers using two pieces of information. First, ONS published population counts by age group, gender, and IMDQ for 2002-2013. We first summed the five-year age bands into ten-year age bands, by gender. For each age-gender group, we calculated the proportion of the population in 2013 in each IMDQ. Thus we calculate that for the above example, a reduction in CHD deaths of 6 per 100,000 for Men 65-74 in IMDQ5 in the year 2015 would represent 22 CHD DPP in absolute numbers ( 6 per 100,000 x 365,000 ).
The CHD DPP for the year 2020 were translated to Life Years Gained using the data in the table below. We assumed that 26% of CHD DPP were among those with the disease state "diagnosed CHD", another 26% with "undiagnosed CHD", and the remaining 48% with "no CHD". From the example above, 22 CHD DPP for Men 65-74 in IMDQ5 would result in an average of 8.4 LYG per DPP (0.26 x 3.6 + 0.26 x 7.6 + 0.48 x 11.5), or in total 185 LYG (8.4 LYG per DPP x 22 DPP). As described in the main text (Table 1), the QALY gained would be 155 QALY (185 LYG x 0.84 QALY per LYG).
We repeated this process for all age groups and IMDQs, for both genders. There are a small number of CHD incidence cases among the age group 25-34, but the number of annual deaths is in the low single digits. The average annual DPP would therefore be less than zero at the scale of reduction we model and not meaningful to include. We did retain this age group in the cost savings portion of the analysis because the reduction in incidence could be non-zero, though still very small. The number of CHD deaths at baseline are for the "do-nothing" scenario based on past trends. The reduction is for a hypothetical policy option that reduces TFA consumption. The number of deaths with the policy implemented is the difference between the baseline and the reduction. This can be illustrated more easily on a plot: First, the number of CHD deaths is plotted on the vertical axis with the IMDQ on the horizontal axis. Then for each scenario (baseline and with policy), a simple linear regression is fitted. The Slope Index is the fitted difference between IMDQ5 and IMDQ1. This is simply the slope of the line multiplied by four because 5 minus 1 is 4. For the "baseline" example, the Slope Index is 10,000 (4 times 2,500), while for the scenario "with policy" the Slope Index is 8,000 (4 times 2,000). The change in slope index is thus a reduction of 2,000 (8,000 minus 10,000 = -2,000).

SUPPLEMENTARY RESULTS
First we provide the CHD DPP for each age group and gender. In some cases, the numbers are listed as simply less than 10 ("<10") to avoid giving the impression of over-precision.         Figure B: Net costs (negative is saving) vs. change in inequality of CHD mortality (negative is reduction). Net costs are from the optimistic scenario (Table 7) where industry reformulation costs are included in part of the normal product life cycle and thus zero. Compared to Figure 2 in the main paper, here we see less overlap in the cost estimates.
Next we present a one-way sensitivity analysis of the TFA consumption gradient, where we assume that each IMDQ has the same average TFA consumption. Table L shows the TFA consumption and  reductions modelled for each policy under this scenario (compare to Table 3 in main text).  Table 4 in main text). For each policy option, the percentage reductions in CHD deaths would vary little by IMDQ (moving down in a column), with the slight differences being only due to different age-gender distributions within each IMDQ. The net effect would be about one-third lower compared to the main analysis, but the absolute inequality reduction would be substantially lower. The restaurant scenario and the labelling scenario that favour IMDQ1 would be essentially neutral at reducing absolute inequality. The policies that benefit all IMDQs equally (total ban, labelling without a gradient) or that favour IMDQ5 (fast food) could still reduce absolute inequality of CHD mortality, but the effect would be small.
Next we show the cost savings associated with each policy option under the assumption of constant TFA consumption (Table N, equivalent to Table 6 in main text). The cost savings would be lower by about the same percentage as the CHD DPP (Table M), i.e. a little more than one-third lower.