Estimates and Projections of the Global Economic Cost of 29 Cancers in 204 Countries and Territories From 2020 to 2050

Key Points Question What is the estimated economic cost and cost distribution of 29 cancers in 204 countries and territories from 2020 to 2050? Findings In this decision analytical modeling study, the global economic cost of cancers from 2020 to 2050 was estimated to be $25.2 trillion (in international dollars at constant 2017 prices). The economic burden and the health burden were distributed unequally across countries, world regions, and country income groups. Meaning Results of this study suggest that global efforts to contain projected increases in the burden of cancers are warranted.

We aimed to quantify each type of cancer's impact on economic output through healthcare expenditures and through productivity losses due to mortality and morbidity.For each country and each cancer, we did the following analysis: Step 1.We identified the disease burden of cancer (in terms of mortality, morbidity, and treatment costs).
Step 2. We constructed economic projections for two scenarios: a status quo scenario, in which gross domestic product (GDP) is projected to grow based on current estimates and projections of disease prevalence, and a counterfactual scenario, in which cancer prevalence is eliminated from the beginning of the time frame.The economic projections utilize a macroeconomic production function and can be further decomposed into two parts: a) Projections of effective labor supply.b) Projections of physical capital accumulation.
Step 3. We calculated the economic loss as the discounted cumulative difference in projected annual GDP between these two scenarios by discount rate.
[6][7] Production function Consider an economy in which time  = 1,2, … , ∞ evolves discretely.Building on Lucas, 8 we considered the following production function for this economy: where   is aggregate output;   is the technological level at time , which we assume evolves exogenously;   is the physical capital stock (i.e., machines, factory buildings, etc.); and   represents aggregate human capital.The parameter  is the elasticity of final output with respect to physical capital.The aggregate production function recognizes that output is not only produced with physical capital and raw labor as in the Solow framework, 9 on which the original EPIC model is based, 10 but with effective labor, of which health is a crucial determinant.
Physical capital evolves according to  +1 = (1 − )  +   −   −   = (1 − )  +     , (2)   where  refers to the depreciation rate,   refers to the saving rate,   refers to the costs of ongoing treatment of cancer (or the types of cancer under consideration), and   refers to the amount of consumption.From Equation (2), it follows that the saving rate is defined as Note that aggregate output   is used for three purposes: (i) to pay treatment costs   (hospitalization, medication, etc.), (ii) to consume the amount   , and (iii) to save.
Individuals of age-sex group (, ) with age  and sex  are endowed with ℎ  (,) units of human capital and supply ℓ  (,) units of labor from age 15 up to their retirement.Children younger than 15 and retirees older than the age of  do not work.R varies by country and could correspond to a high age (e.g., some people older than 80 could also be working).In the theoretical derivations, R indicates the upper bound of the summation.In our simulations, we used labor projections data from the International Labour Organization, and positive values for the labor force exist for cohorts older than 65.We discretized age less than 65 into 13 groups every five years and discretized all ages older than 65 into 1 group.Thus, we have 28 age-sex groups marked as G. Aggregate human capital in the production function (1) is then defined as the sum over the age-sex-time-specific effective labor supply of each age-sex group: = ∑ ℎ  (,) ℓ  (,)   (,) where   (𝑎,𝑠) denotes the number of individuals in age-sex group (, ).Note that aggregate human capital increases with the number of working-age individuals who live in the economy (i.e., with a higher   = ∑   (,) (,)∈ ), with individual human capital endowment (i.e., with a higher ℎ  (𝑎,𝑠) for at least one group), and with labor supply (i.e., with a higher ℓ  (𝑎,𝑠) for at least one group).
We followed Mincer 11 and constructed the average human capital for each age-sex group with age a according to an exponential function of education and work experience: ℎ  (,) = exp [ 1 (  (,) ) +  2 ( −   (,) − 5) +  3 ( −   (,) − 5 where  1 is the semi-elasticity of human capital with respect to average years of education as given by   (𝑎,𝑠) , and  2 and  3 are the semi-elasticities of human capital with respect to the experience of the workforce ( −   (,) − 5) and the experience of the workforce squared ( −   (,) − 5)

2
, respectively.Here, we assumed a school entry age of 5 years throughout.

Impact of cancer on labor supply
Following Bloom et al. 2 and Chen et al., 3 5 6 12 the evolution of labor supply in the status quo scenario is given by   (,) = ℓ  (,)   (𝑎,𝑠) with   (,) = [1 −  −1 (−1,) ] −1 (−1,) , (6)   where   (𝑎,𝑠) is the overall mortality rate of the age-sex group with age a and sex s at time .Mortality and morbidity reduce effective labor supply.The reduction of the population size   (,) captures the mortality effect.
Let  , (𝑎,𝑠) denote the mortality rate of people in each age-sex group due to cancer and let  −, (,) be the overall mortality rate due to causes other than cancer.Then we have (1 −   (,) ) = (1 −  , (,) )(1 −  −, (,) ), (7)   Next, we considered the mortality effect of cancer.In general, it reduces labor supply by reducing the population   (,) (through  , (,) ).In the counterfactual case, where cancers are eliminated from time  = 0 onward, the evolution of labor supply is defined similarly to Equation ( 6), but with a different overall mortality rate ( −, (,) instead of   (,) ).For simplicity, we assumed that the number of births is the same in both cases at each point in time .
In the counterfactual scenario, the size of the cohort with age  and sex s at time  ( ̅  (,) ) evolves according to Following Bloom et al., 2 the loss of labor due to mortality accumulates over the years according to The reduction of the labor participation rate ℓ  (,) captures the morbidity effect because people with an illness typically reduce their labor supply, either by reducing working hours or by leaving the workforce.
Following Bloom et al., 2 the labor participation rate in the counterfactual scenario ℓ ̅  (,) can be calculated as where  (𝑎,𝑠) measures the size of the morbidity effect relative to the relevant mortality rate, and  is the probability of a patient not recovering from cancer morbidity in each year.
Because the impact of morbidity is hard to estimate directly, we first defined the relative effect for each age-sex group: (,) = loss of labor due to morbidity of group (, ) loss of labor due to mortality of group (, ) .( 11) Next, we assumed that the following holds in any given year for each age-sex group: where  (𝑎,𝑠) represents the years lived with cancer and  (,) represents the years of life lost due to cancer.Notice that  (,) can be calculated from the corresponding DALY data reported by the Global Burden of Disease Study. 1 In sum, by reducing the prevalence of cancer, the counterfactual scenario is associated with an increase in labor supply as compared with the status quo scenario.We approximated the change in labor supply (at time  for each age-sex group) by For the more general case of a partial reduction in the prevalence of cancer by a factor , we obtained the loss of labor for each age-sex group at time  as Bloom et al. ( 2020) 2 provides the detailed mathematical proof.

Impact of cancer on physical capital accumulation
Cancers also impede the accumulation of physical capital because savings finance part of the treatment costs.Following Bloom et al. 2 and Chen et al., 3 physical capital accumulation in the counterfactual scenario can be written as where an overbar indicates the counterfactual scenario and where  is the fraction of the treatment cost that is diverted to savings.The counterfactual saving rate is thus defined by For more details, see Because cancers are assumed to be eliminated in the counterfactual scenario, the resources that were devoted to their treatment can now be used for savings or for consumption.Notice that this creates an income effect that, in reality, could affect the division of households' income between savings and consumption.For tractability, we assumed that aggregate investment consists of two parts in the counterfactual scenario: a fixed share γ  of total output and an additional part from   that would otherwise have been used to pay to treat cancers: Similarly, for the case of a partial reduction in cancer prevalence by , we have The intuition is that if cancers are partially eliminated, the treatment cost that is diverted to savings should be added back proportionally.

Education
Age-specific educational attainment data are from the Barro-Lee Educational Attainment Database, 13 which provides educational attainment data by five-year age-sex groups up to 2010.For 2010-2030, no agespecific data are available, but the database provides projections for the population aged 15-64.We approximated the age-specific estimates by assuming that educational attainment for each age-sex group grows at the same rate.Because the Barro-Lee database presents data in five-year intervals, linear interpolation was adopted to extend the estimates for each year.For 2030-2050, we projected educational attainment by assuming the same growth rate as during 2010-2030 for each age and sex group.

Mortality/morbidity
The age-sex-specific mortality and morbidity (measured in years of life lost and years of life lost to disability) of cancers up to 2019 are from the recently updated GBD estimates. 1To extend the estimates beyond 2019, we assumed that the mortality rate of cancers grows at the same rate as during 2010-2019 for each country.Morbidity estimates were obtained similarly.If the projected mortality rate grew too large (i.e., if it more than doubled the current annual rate in 30 years), we limited the mortality rate's growth rate to 2%.

GDP projection
The GDP estimates (in constant 2017 international dollars or INT$) up to 2020 are from the World Bank database. 14The GDP growth rates for 2021-2027 are from the International Monetary Fund's World Economic Outlook as of April 2022. 15We assumed that growth beyond 2027 will be the same as in 2015-2019.

Physical capital
For each country, the physical capital stock (in constant 2017 INT$) was obtained from the Penn World Table projections. 16abor participation For each country, the labor participation rates (by five-year age-sex groups) are from the International Labour Organization database for 2010-2020. 17For estimates beyond 2020, we estimated the labor participation rates for the five-year age groups using a logistic regression.

Population
For each country, the population by five-year age group was obtained from the population dynamics database built by the Department of Economic and Social Affairs (DESA) of the United Nations. 18For those countries without five-year age group data, we used the total population from DESA to impute the economic burden.

Saving rate and health expenditure
We obtained country-specific saving rates and health expenditures from the World Bank database. 19For the projection, we assumed that the saving rate remains constant (at the average from 2010-2019), while health expenditures (as a percentage of GDP) grow at the same rate as in 2000-2019. Parameter

Treatment costs
5][26]  We calculated the country-level cancer-related treatment costs for the countries with data and extrapolated costs for the countries without data, assuming that the per case treatment cost for cancer was proportional to the health expenditure per capita of the country, as previous studies have assumed. 12 28 29The intuition here is that health expenditure per capita could be a good metric of the cost of treating a certain disease (cancer for example) across countries.This is an approximate estimate for tractability due to lack of data.Under this assumption, share of cancer-related treatment costs out of all health expenditures (all treatment costs) can be calculated via the cancer prevalence rate.Specifically, we used the U.S. data as the base share of cancer-related treatment costs out of all health expenditures and scaled this figure by the ratio of cancer prevalence between other countries and the United States.For years after 2010, we assumed that the treatment costs of cancers grow at the same rate as per capita health expenditures for each country.
For the 60 countries and territories with incomplete data (mostly on education, physical capital, and the saving rate in eTable 3) but reliable data for GDP and DALYs, we used a linear projection to approximate the economic burden of cancer.We found that the ratio of economic burden to GDP is highly correlated to DALYs.Thus, we imputed economic burden results based on an ordinary least squares linear regression of the relationship between GDP and DALYs.For the imputation, we then used the formula where and are the parameter estimates for the intercept and the slope obtained in the linear regression.
eTable 4 shows the regression coefficients with p-values.All p-values of the coefficient of DALYs are less than 0.0001.The average number of DALYs is 3,233, which means the constant value is far less than the coefficient of DALYs multiplying DALYs.Therefore, the large p-value of the coefficient of constant value is also acceptable for some cancers.To improve the fitting result, we replaced DALYs with all healthy data such as the mortality, incidence, and prevalence rates; DALYs; years of life lost to disability; and years of life lost for each cancer.Due to the multicollinearity of these IHME variables, we performed principal component analysis (PCA) and set the two main components of the PCA results as independent variables.We got a better fit for the linear regression for almost all cancers except non-melanoma skin cancer.In this paper, we used the primary results for almost all cancers except other neoplasms using PCA-refined methods (R-squared is 0.7).

eTable 4. Estimating the relationship among percentage of economic loss, the two main PCA components of IHME data, and the indicator of high-income countries
We then imputed the percentage of the economic loss in total GDP using the coefficients from the regression and calculated the economic loss for the 60 countries with incomplete data as listed in eTable 3.
The results are merged into the main paper.
eAppendix 5. Geographic distribution and discounted estimates of economic burden of cancers  Our model has several strengths and limitations, as summarized in eTable 11.First, we had to rely on imputations to calculate cancer-related health expenditures, which are based on data for the United States and then projected to other countries.This can either underestimate or overestimate the country-specific treatment costs for cancers.We also made assumptions about projections for labor participation rate, mortality, and morbidity data.This also may either underestimate or overestimate the results in some countries.Second, we had to impute the economic burden of cancers for 60 out of 204 countries and territories; however, this does not compromise our results, given that the 144 countries for which we had complete data account for 92.7% of the global population.Third, we did not account for changes in labor force participation of family members of people with cancer requiring informal care.With respect to this limitation, our findings should provide the lower bound of the economic cost of cancers.Finally, we did not consider unemployment, nor did we include an explicit treatment of price movements and endogenous savings in the framework.
Our analysis uses a simulation model grounded in dynamic macroeconomic theory to advance understanding of the global macroeconomic cost of cancers in several important ways.First, the calculations are based on recently developed methods and rely on the best available global data, including data from the recently updated Global Burden of Disease Study 2019, the Barro-Lee education database, the World Bank, and the International Labour Organization.Second, this study is the first to account for cancer's influence on economic growth through a morbidity effect for 204 countries and territories in the world.Third, our framework is the first to consider productivity loss among people with different education and experience levels-the lack of which is a key limitation of previous studies.Fourth, our work shows the causal relationship between cancers and GDP.It avoids issues of reverse causality because we did not estimate the relationship but rather constructed it from our simulated production function.Fifth, we provided a detailed, step-by-step description of our methods and the data sources and specific parameters we used in our analysis in the SI Appendix.Sixth, we included sensitivity analyses to account for underlying uncertainty by adjusting the mortality and morbidity data based on upper and lower bounds of the GBD data.

eAppendix 1 . 1 ***White areas represent countries with insufficient data eFigure 4 .
Figures S1-S4show the health burden of cancer.The numbers are based on the Global Burden of Disease Study (2020).1

eAppendix 7 .eFigure 7 . 8 .
Contribution of treatment costs and human capital eFigure 7 shows the contribution of treatment costs to the total macroeconomic burden of cancers by country income group and by World Bank region.eFigure 8 shows the contribution of human capital to the total macroeconomic burden of cancers by country income group and by World Bank region.Contribution of treatment costs to the total economic cost of cancers by country income group and World Bank region eFigure Contribution of human capital to the total economic cost of cancers by country income group and World Bank region eAppendix 8. Strengths and limitations Bloom et al. (2020) 2 and Chen et al. (2018). 3

12 eTable 1. Parameter values and data sources
values and data sources eTable 1 shows parameter values and data sources in the model, where definitions for parameters are consistent with Bloom et al. (2020) 2 and Chen et al. (2019). 5

Economic cost of cancers as a percentage of total GDP in 2020-2050 Tables
S5-S7 show the total discounted economic burden of cancers in 2020-2050 for each country, by World Bank region, and by World Bank income group, using discount rates of 2%, 0%, and 3%.eTable 5. Total macroeconomic cost, economic cost as a share of GDP in 2020-2050, and per capita economic cost attributable to cancers, by World Bank region and country (in 2017 INT$),

Total macroeconomic burden attributable to cancers in 2020-2050 using a discount rate of 0% for 204 countries, by World Bank region (in 2017 INT$)
Please note that results for countries marked with an asterisk in column 2 are imputed due to missing data.

Cancer type responsible for the most DALYs in 2019 for each country (TBL = tracheal, bronchus, and lung) eTables
8-9 illustrate the distribution of macroeconomic burden of cancers by World Bank region and income group with discount rates of 0% and 3%.eTable 8.

Total macroeconomic cost, economic cost as a share of total GDP in 2020-2050, and per capita economic cost attributable to cancer mortality and morbidity, by World Bank region, by World Bank income group, and globally with discount rate of 3%
eAppendix 6. Differences in macroeconomic loss and lifetime disease burden eTable 10 illustrates the differences in economic costs across regions or income groups that also differ in terms of GDP, population, and DALYs.

eTable 10. Comparison of macroeconomic loss and lifetime disease burden by World Bank region and country income group
Please note that each country is classified into World Bank regions as in eTable 5.The seven World Bank regions do not include Cook Islands, Niue, Palestine, and Tokelau.