Cost-effectiveness Analysis of Nutrition Facts Added-Sugar Labeling and Obesity-Associated Cancer Rates in the US

Key Points Question What is the estimated association between added-sugar labeling and obesity-related cancer rates in the US? Findings This economic evaluation of Nutrition Facts added-sugar labeling and obesity-related cancer rates estimated that implementing the policy was associated with a reduction of 30 000 new cancer cases, 17 100 cancer deaths, and $1600 million in medical costs among US adults over a lifetime. This policy would generate net savings of $704 million from a societal perspective and $1590 million from a health care perspective. Meaning These model findings suggest added-sugar labeling would be associated with reduced costs and lower rates of obesity-related cancers.


eAppendix 1. Baseline cancer incidence and methods of cancer incidence projections for 13 types of cancer eAppendix 4. Estimate the association between added sugar labeling policy and added sugar intake
In order to understand the impact of the Nutrition Facts added sugar labeling policy, we must understand how both consumers and industry will respond to the policy as illustrated by the logical framework.
Because there is no much literature on the impact of added sugar labeling specifically, some assumptions must be made using the best available evidence from a mixture of sources. In Huang et al., the authors used the association estimate of calorie labeling, a reduction of intake by 6.6% (95% CI: 4.4% to 8.8%), as a proxy for added sugar labeling impact. 9,10 This policy impact estimate was chosen to represent a more conservative estimate than the larger impact observed from labeling on other dietary constituents such as sodium and trans fat. 9 This impact on consumer behavior alone was assumed to take effect during the first year of implementation and no further reduction thereafter. For industry reformulation, Huang et al. assumed no reformulation in the first year of labeling implementation, then 8.25% (95% CI: 7.5% to 9.0%) of the sugarcontaining products would be reformulated each of years 2 to 5 during the intervention period to achieve 25% reduction in added sugar content in these products, with no additional reformulation thereafter. [10][11][12] In sum, this represents an 8.25% x 4 years x 25% = 8.25% net reduction in added sugar amounts of U.S. sugar-containing products over the intervention period. 10

eAppendix 5. Methods of estimating policy implementation costs
We estimated the costs of implementing the Nutrition Facts added sugar labeling for both government and industry, including government administration cost, monitoring and evaluation costs, industry compliance costs and reformulation costs, based on FDA's budget report, 13 the Nutrition Review Project report, 14 and FDA's RIA 11 (eTable 2). a Policy intervention costs were inflated to 2015 US (December) dollars using the Consumer Price Index. b Given no range of uncertainty was provided in source materials, we assumed 20% uncertainty around these costs.

eTable 2. Implementation cost estimates for the Nutrition Facts added sugar labeling policy (in 2015 US dollars)
Added sugar labeling is one of many provisions in FDA's rule to update the Nutrition Facts label. The cost of implementing all provisions is fixed for government (administration and monitoring and evaluation) and industry compliance. Therefore, we attributed 50% of the costs for implementing the entire labeling policy to the costs of added sugar labeling for government administration (#1), government monitoring and evaluation (#2), and industry compliance cost (#3) as this approach generated more conservative estimates. Uncertainty for the costs associated with government administration (#1) and government monitoring and evaluation (# 2) was not provided in the source materials. 13, 14 We assumed that uncertainty is 20% around these costs.
For annual costs, namely the government monitoring and evaluation cost (#2) and the industry reformulation cost (#4), we applied 3% discounting rate recommended by the Second Panel on cost-effectiveness in health and medicine 15 to reflect the present value of future costs of government monitoring and evaluation and industry reformulation. The model is a closed cohort model, so we computed the discounted present value of per-person costs and total national costs for persons alive at implementation who remained alive in each subsequent year (not for the larger total US population in each year, which also has growth from immigration and new persons reaching the threshold age). The year-specific discounting factor is estimated by 1/(1+3%) (t-1) (t is number of years of policy intervention, t=1, 2, 3, …, lifetime) (eTable 3): eTable 3. Discounting factor in each year over a lifetime As our model estimated the costs and health outcomes based on a closed cohort and the population size decline over time, we need to express the annual costs in proportion to the population at risk. The population at risk was estimated based on the proportion of death (Pdt, t=1, 2, 3, …) in each year. We first obtained the proportion of people who are alive at each year by calculating 1-Pdt (t=1, 2, 3, …). Then we multiplied the baseline population size of 235 million by the proportion of people who are alive in each year (eTable 4). We then estimated the per-person annual cost for cost categories #2 and #4, by dividing the annual cost estimated in the second year of implementing the policy among all US population by the population size in the second year. Specifically, for government monitoring and evaluation, the per person annual cost is estimated $251,824/233,719,989=$0.00108, and that for industry reformulation is $44,550,571/233,719,989=$0.190615. Taken together, to estimate the discounted annual cost of #2 and #4, we multiplied the population at risk, the per person annual cost estimated at year-2, and the year-specific discounting factor, using: discounted annual cost = population at risk x per-person annual cost x 1/(1+3%) (t-1) .

eAppendix 6. Annual health-related costs among cancer patients and the general population without cancer
The annual health-related costs data include: 1) medical expenditure, 2) productivity loss form missed work days or disability, and 3) patient time cost associated with receiving care for cancer survivors by age (under 65 vs. above 65 years old) and phase of care (initial, continuing, end-year of life); 4) medical expenditure, 5) productivity loss, and 6) patient time cost for individuals without cancer by age and status of end year of life. The description of data source and data structure were provided in eTable 5. We extracted the raw data for each of the costing component from the published literatures. 2,[16][17][18][19][20] The overall assumptions for data extraction include: 1) health-related costs for breast cancer among post-menopausal females, advanced prostate cancer, esophageal adenocarcinoma, and stomach cardia cancer, by age, sex, and phase of cancer care, were the same as those for breast cancer, prostate cancer, esophagus cancer, and stomach cancer; 2) if no data available for a specific cancer type, we assumed the costs for that cancer type were the same as the estimates of costs for all-cancer sites, e.g., medical expenditure for all-cancer sites were used to replace the medical expenditures for multiple myeloma, gallbladder, liver, and thyroid cancers; 3) we extracted the costs for end-year of life due to cancer death and assumed that death due to other causes is not a competing outcome; 4) we assumed that the end-year life medical expenditure for individuals without cancer does not vary by the 32 subgroups.

eTable 5. Description of Data Source of Health-Related Expenditures
If a specific costing component was not reported directly in the raw data, we calculated the cost for that component based on available data. For example, the annual productivity loss for colorectal cancer were reported as a percentage of total health-related costs. 20 We multiplied the percentage and the total health-related costs to obtain the productivity loss for colorectal cancer. We also performed data imputation for unavailable data. For instance, the annual productivity loss for allcancer sites was reported by time interval since cancer diagnosis (diagnosed within one year vs. diagnosed greater than one year). 16 To obtain this costing component by the defined phases of care, we calculated the weighted means which was used as eTable 6. Defining population and 32 subgroups

. One-way sensitivity analysis of net costs of Nutrition Facts added sugar label and obesity-associated cancer rates to varying assumptions of key input parameters from (A) societal perspective and (B) healthcare perspective
1a) conservative policy impact assumed half of the base-case policy impact (consumer behavior: -3.3%; industry reformulation: -4.13%); 1b) optimistic policy impact assumed two times of the base-case policy effect (consumer behavior: -13.0%; industry reformulation: -16.5%); 2a) weaker diet-BMI association assumed half of the base-case diet-BMI association (healthy-weight: 0.05 kg/m 2 ; overweight/obese: 0.12 kg/m 2 ); 2b) stronger diet-BMI association assumed two times of the base-case diet-BMI association (healthy-weight: 0.20 kg/m 2 ; overweight/obese: 0.46 kg/m 2 ); 3) 2% annual increase in medical expenditure on cancer care; 4a) lower policy implementation costs assumed 25% of the total costs for implementing the Nutrition Facts label; 4b) higher policy implementation costs assumed 75% of the total costs for implementing the Nutrition Facts label; 5a) lower discounting rate assumed 0% discounting rate; and 5b) higher discounting rate assumed 5% discounting rate. Under base-case scenario (policy effect assumed consumer behavior: -6.6%, and industry reformulation: -8.25%; diet-BMI association assumed healthy-weight: 0.1 kg/m2, and overweight/obese: 0.23 kg/m2; medical expenditure on cancer care assumed 0% annual increase; policy implementation costs assumed 50% of the total costs for implementing the Nutrition Facts label; discounting rate assumed 3%), the policy was cost-saving from both societal and healthcare perspectives. The policy remained cost-saving for all sensitivity analyses from the healthcare perspective and from societal perspective with additional industry reformulation. With consumer behavior alone, the policy was cost-saving when under 1b, 2b, 3, 4a, and 5b and was cost-effective under 1a (ICER: $8330/QALY), 2a ($8320/QALY), 4b ($933/QALY), and 5b ($1680/QALY). -