### Data and study design

This study applied a top-down costing approach and performed a macrosimulation model to estimate the future costs of colorectal cancer attributable to red and processed meat consumption, using the Brazilian population as a case study. We used the following data: 1. Relative risks (RR) from WCRF/AICR meta-analyses [13]; 2. Prevalence data (%) of red and processed meat consumption in adults aged 20 years or older who relied exclusively on the public health system from representative national surveys; 3. Nationwide registries of federal direct healthcare costs of inpatient and outpatient procedures in the SUS in adults aged 30 years or older with cancer. The parameters used in the model are available in Supplementary Material A.

We estimated the potential impact of red and processed meat on federal direct healthcare costs of cancer, assuming a 10-year time lag between exposure and outcome via comparative risk assessment. We used the potential impact fraction (PIF) equation and the Monte Carlo simulation method to estimate the attributable costs and their 95% uncertainty intervals, considering the theoretical-minimum-risk exposure and other counterfactual (alternative) red and processed meat consumption scenarios. We assessed cancer costs attributable to red and processed meat, multiplying PIF by the direct healthcare costs of cancer.

### Relative risk estimates and cancer sites

We considered in our study only colorectal cancer as there is strong evidence of association (convincing or probable) with red and processed meat according to the WCRF/AICR [7]. The WCRF/AICR method of grading evidence has been designed to operationalize the criteria identified by Bradford Hill as contributing to an inference of causation from observational data. Strong evidence of association sustains a judgment of a convincing causal (or protective) relationship that justifies making recommendations designed to reduce cancer risk. This evidence is robust enough to be unlikely to be altered in the foreseeable future as new evidence accumulates [7]. We detailed the list of the 10^{th} Revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10) codes in Supplementary Material B.

We obtained the \({RR}_{x}\) for colorectal cancer incidence by sex from the WCRF/AICR dose–response meta-analysis [13], considering the increment of \(x\) g/day of red and processed meat (\(x\)=100 and 50, respectively). It is important to enphazise that this meta-analysis [13] represents the last update from WCRF concerning the evidence from observational studies (cohort, nested case–control and case-cohort designs) on the association between foods, nutrients, physical activity, body adiposity and the risk of colorectal cancer in men and women. The WCRF/AICR dose–response meta-analysis is part of the CUP (Continuous Update Project), a global network of researchers that evaluates cancer prevention research.

We converted these measures per increment of 1 g /day of the exposure (\({RR}_{1}\)) using the following equation [14]:

$${RR}_{1}=exp(\frac{\mathrm{log}({RR}_{x})}{x}).$$

To obtain the RR for each exposition category (RRc) (Supplementary Material C), we used the following equation [15]:

$${RR}_{c}={RR}_{1}^{{M}_{c}-ref},$$

where Mc represents the median value in each category, and ref represents the reference category value (< 70 g/day for red meat and 0 g for processed meat). The reference category reflected the theoretical minimum risk exposure level [7] and the Brazilian INCA recommendations [8].

### Assessment of red and processed meat consumption prevalence

We obtained an estimated prevalence of red and processed meat consumption in the adult population aged ≥ 20 years from the National Household Budget Survey (Pesquisa de Orçamentos Familiares – POF), a cross-sectional nationally representative survey conducted in Brazil in 2008–2009 [16] and 2017–2018 [17]. We considered only adults aged 20 years or older who reported not having health insurance to obtain the prevalence and 95% confidence interval for each red and processed meat consumption by sex.

POF collected two 24-h real-time food records from 34,003 participants in 2008–2009 [16] and 37,690 participants in 2017–2018 [17]. Using a food portion table, we converted reported food amounts into grams [18]. We estimated red meat consumption (g/day) based on the consumption of all types of meat from mammals, such as beef, horse, goat, lamb, mutton, and pork, whereas processed meat (g/day) on the consumption of meat preserved by smoking, curing, salting, the addition of chemical preservatives (e.g., bacon, chorizo, corned beef, ham, pastrami, salami, and sausages). We displayed categories of red and processed meat consumption in grams/day in Figs. 1 and 2, respectively. The reference category (< 70 g/day of red meat and 0 g/day of processed meat) aimed to reflect the theoretical minimum risk exposure level [7] as well as the recommendations of the Brazilian INCA [8].

The POF microdata is available in the public domain via the Brazilian Institute of Geography and Statistics (IBGE) at http://www.ibge.gov.br (Supplementary Material D). We incorporated the complex sample design into all estimates using RStudio version 1.4.1103.

### Counterfactual (alternative) scenarios for red and processed meat consumption

We proposed four counterfactual (alternative) scenarios of population-wide reduction in red and processed meat consumption (observed in 2017–2018) to be achieved in Brazil in 2030 to save direct healthcare costs with cancer in 2040 (Figs. 1 and 2). The counterfactual scenarios considered reduction in one serving per week (Scenario 1: -120 g/week of red meat and -50 g/week of processed meat) and shifting red and processed meat categories, which considered a scenario where everyone consumes: for red meat: Scenario 2) < 140 g/day; Scenario 3) < 210 g/day; Scenario 4) < 280 g/day; for processed meat: Scenario 2) < 50 g/day; Scenario 3) < 100 g/day; Scenario 4) < 150 g/day.

### Federal direct healthcare costs of cancer in the Brazilian SUS in 2030 and 2040

We retrieved registries of federal direct healthcare costs of inpatient and outpatient cancer-related procedures between 2008 and 2019 from the SIH/SUS and SIA/SUS (Supplementary Material D**)**. We used the 10^{th} Revision of the International Statistical Classification of Diseases and Related Health Problems (ICD-10) codes for recovering cancer procedures from information systems (Supplementary Material B). We stratified the direct healthcare costs by sex. Assuming a 10-year time lag between exposure and outcome, we considered the procedures approved for payment in adults with cancer aged 30 years or older in 2030 and 2040.

We performed a simple linear regression to predict the future costs of cancer (dependent variable) as a function of time (independent variable) up to 2030 and 2040 based on the values practiced over time between 2008 and 2019. Controlling for potential confounders while examining the possible determinants of cost is crucial. Once our outcome was the direct healthcare costs over time, it was unnecessary to control for confounders because we observed their effect in the observed costs used to fit the regression model [19]. We transformed the monetary values in Brazilian Real (R$) to United States Dollar (US$), considering the purchasing power parity (PPP) of 2018 (conversion factor 2.226) to current costs and of 2019 (conversion factor 2.281) to future costs [20].

#### Cancer costs attributable to red and processed meat consumption

Based on the abovementioned intermediate inputs of the models, we calculated the (PIF) for colorectal cancer by sex and counterfactual (alternative) scenario using the following equation [21]:

$$PIF= \frac{{\sum }_{i=1}^{n}{P}_{i} {RR}_{i}-{\sum }_{i=1}^{n}{P^{\prime}}_{i} {RR}_{i} }{{\sum }_{i=1}^{n}{P}_{i }{RR}_{i}},$$

where \({P}_{i}\) is the proportion of the population at the level \(i\) of red and processed meat consumption in a given year, \({P{\prime}}_{i}\) is the proportion of the population at the level \(i\) of red and processed meat consumption in a given counterfactual (alternative) scenario, and \({RR}_{i}\) is the RR of colorectal cancer at the level \(i\) of red and processed meat consumption. We displayed the levels \(i\) for red and processed meat consumption in Figs. 1 and 2. Of note, the PIF equals the Population Attributable Fraction (PAF) when the counterfactual (alternative) scenario represents the theoretical minimum risk exposure level [21, 22].

To estimate the fraction of colorectal cancer costs attributable to combined red and processed meat consumption, we used the joint PIF/PAF equation [23], which assumes the absence of interaction between risk factors:

$$Joint PIF=1-{\prod }_{i=1}^{n}(1-{PIF}_{i}).$$

To assess the cancer costs attributable to red and processed meat, we multiplied PIF by the total colorectal cancer costs. We considered the prevalence in 2008–2009 and 2017–2018 and the costs of cancer in 2018 and 2030, respectively, assuming at least a 10-year time lag between exposure and outcome (i.e., based on the average follow-up time of prospective cohort studies [13]). Finally, we calculated the potential savings in cancer costs in 2040 if reduced red and processed meat consumption occurred in Brazil to levels fixed in the counterfactual (alternative) scenarios in 2030.

We quantified the uncertainty in all modeled estimates using the Monte Carlo simulation approach [24, 25] with 10,000 iterations. The simulation works thoroughly, producing a draw from the distributions of a) baseline prevalence per red and processed meat consumption category considering a binomial distribution; b) the log of the RR per exposure category for the association of red and processed meat consumption with colorectal cancer risk assuming a normal distribution. We calculated PIF by sex for the 50^{th}, 2.5^{th}, and 97.5^{th} percentiles as the central estimate and 95% uncertainty intervals across all simulations. Negative values of PIF derived from the Monte Carlo simulation were rounded to 0, assuming that reducing red and processed meat consumption values may not increase the risk of cancer and, consequently, the attributable costs. We used R Studio version 1.3.1093 for analysis.

## Add Comment