Age adjustment in ecological studies: using a study on arsenic ingestion and bladder cancer as an example
© Guo; licensee BioMed Central Ltd. 2011
Received: 8 June 2010
Accepted: 20 October 2011
Published: 20 October 2011
Despite its limitations, ecological study design is widely applied in epidemiology. In most cases, adjustment for age is necessary, but different methods may lead to different conclusions. To compare three methods of age adjustment, a study on the associations between arsenic in drinking water and incidence of bladder cancer in 243 townships in Taiwan was used as an example.
A total of 3068 cases of bladder cancer, including 2276 men and 792 women, were identified during a ten-year study period in the study townships. Three methods were applied to analyze the same data set on the ten-year study period. The first (Direct Method) applied direct standardization to obtain standardized incidence rate and then used it as the dependent variable in the regression analysis. The second (Indirect Method) applied indirect standardization to obtain standardized incidence ratio and then used it as the dependent variable in the regression analysis instead. The third (Variable Method) used proportions of residents in different age groups as a part of the independent variables in the multiple regression models.
All three methods showed a statistically significant positive association between arsenic exposure above 0.64 mg/L and incidence of bladder cancer in men and women, but different results were observed for the other exposure categories. In addition, the risk estimates obtained by different methods for the same exposure category were all different.
Using an empirical example, the current study confirmed the argument made by other researchers previously that whereas the three different methods of age adjustment may lead to different conclusions, only the third approach can obtain unbiased estimates of the risks. The third method can also generate estimates of the risk associated with each age group, but the other two are unable to evaluate the effects of age directly.
Keywordsecological study age adjustment direct standardization, indirect standardization, standardized morbidity ratio, arsenic drinking water bladder cancer
In spite of their limitations, ecologic studies are often used in epidemiologic research. These studies gather study participants into groups, mostly according to the geographic area of residence, and treat the whole group of people as a unit . In ecological studies, distribution of age in the unit population may affect the results, and therefore evaluating and adjusting for age effects are desirable in many cases. In regression analyses, a common approach is to apply direct standardization method  to obtain the age-standardized risk of each unit population and then use it as the dependent variable. Another common approach, especially frequently applied in the SMR ("standardized morbidity ratio" or "standardized mortality ratio") studies, is to adopt indirect standardization method to obtain age-standardized risk ratio  of each unit population and then use it as the dependent variable instead. The third technique is to treat age as a predictor and adjust for its effects by adding independent variables [4, 5].
Different approaches, however, may lead to different results. Rosenbaum and Rubin have shown that the third approach can generate unbiased risk estimates as those obtained from simple linear regression models using data on individual participants. They also showed that the first approach generally cannot generate unbiased risk estimates and suggested avoiding its use except when the bias can be shown to be negligible for the purposes of the study. However, they did not provide any empirical examples in the paper. Likewise, the second approach has been recognized as one of the "common errors in disease mapping" by Ocaña-Riola , but no empirical examples were included in the paper. Therefore, the objective of this study is to compare these three approaches using an ecologic study on the associations between arsenic levels in drinking water and the incidence of bladder cancer in Taiwan as an empirical example. On the basis of the three approaches, three regression models were applied to model the relations between each of the outcome measurements (age-adjusted mortality rate, standardized mortality rate, and crude mortality rate) and exposure levels.
The association between consumption of artesian well water and cancer in Taiwan has been documented since the 1960s [7–18], and bladder cancers had the highest relative risks [13, 19]. All the three methods have been used to study the associations between arsenic ingestion and cancer in Taiwan [7–9], but most previous studies were limited to 6 townships in the southwestern coast area. The current study evaluated the associations between arsenic ingestion and bladder cancer using data on 243 townships in Taiwan, including the 6 most frequently studied.
Arsenic levels in drinking water
Data on the arsenic levels in drinking water were obtained from a nation-wide survey conducted by the Taiwan Provincial Institute of Environmental Sanitation  using the standard mercuric bromide stain method . According to the standard solutions used in that survey, drinking water arsenic levels can be grouped into 10 categories: "undetectable" (test result compatible with the blank control) "trace" (test result between the blank control and the 0.01 mg/L standard), "0.01 mg/L," "0.02 mg/L," "0.03-0.04 mg/L," "0.05-0.08 mg/L," "0.09-0.16 mg/L," "0.17-0.32 mg/L," "0.33-0.64 mg/L," and " > 0.64 mg/L." While Taiwanese laboratories applied this method, the limit of detection (LOD; defined as the value of the mean plus three times of the standard deviation obtained from repetitive testing of blank controls) was 0.04 mg/L . Therefore, in the data analyses, all levels at or below the LOD were combined together as a single category " < 0.05 mg/L."
The original survey data are available for 65269 wells in 243 townships, with an average of about 269 wells in each township. Because the survey was specifically for the arsenic level in drinking water, the standard report form did not include other chemical characteristics of the well water. As in most of the similar ecological studies, the number of users of each well was not recorded. The survey performed almost all the measurements between 1974 and 1976. At the time of survey bottled water was generally unavailable, and therefore it can be assumed that most quantity of drinking water was taken from the same source all days in the surveyed areas.
Distribution of arsenic exposure levels in well water
Arsenic level (mg/L)
Average % of wells in each township
5 605 000
5 144 000
10 749 000
Collection of other data
Cases of bladder cancer diagnosed between January 1, 1980 and December 31, 1989 were identified using the computerized database of the National Cancer Registry Program, which is operated by the Department of Health. Gender, age, diagnoses, and township of residence were reported for each registered case. Cases with ICD-O codes  from 188.0 to 188.9 were defined as bladder cancer cases, and 3068 cases, including 2276 men and 792 women, were identified.
Demographic data on the residents in each township at the end of 1985, the midpoint of the ten-year study period, were obtained from the Department of Internal Affairs. The numbers of residents of seven age groups were calculated: 0-19 years, 20-29 years, 30-39 years, 40-49 years, 50-59 years, 60-69 years, and above 69 years.
An urbanization index developed by Wu  on the basis of 19 socioeconomic factors was adopted to assess the associations between urbanization and incidence of bladder cancer. The study townships had urbanization indexes ranging from -1.410 to 3.257 (mean = 0.224, standard deviation = 1.128).
The magnitude of cigarette sales was used to evaluate effects of smoking. In Taiwan, cigarette selling was a monopoly business operated by the Tobacco and Alcohol Monopoly Bureau during the study period. Sales records collected from the Bureau in a previous study  were adopted to estimate the number of cigarettes sold per capita per year in each township, which had a range of 14.94 to 689.93 (mean = 63.76, standard deviation = 66.11). The unit for of cigarette sales used in the analyses was 100 cigarettes.
For comparison, three different methods were applied to analyze the data, but they required different information. To account for the fact that the size of the population was different across the townships, in all three approaches, the population in each township was used as the weighting factor in regression models.
where for each township, Xj is the proportion (as percentage) of residents with arsenic exposures in category j, U is the urbanization index, and T is the number of cigarettes (in hundreds) sold per capita. Because the exposure category " < 0.05 mg/L" was used as the reference, X1 = percentage of residents in the "0.05-0.08 mg/L" category, X2 = percentage of residents in the "0.09-0.16 mg/L" category, and so on. In this case, α (intercept) is the estimated background cumulative incidence rate, βj indicates the rate difference (RD) associated with each 1% increase in residents in category j, γ indicates the RD associated with each one-unit increase in urbanization index, and δ indicates the RD associated with each 100 cigarettes sold per capita.
where Xj, U, and T are defined as in Model 1. In this case, α' is the estimated background ratio, βj' indicates the increase in SIRatio associated with each 1% increase in residents in category j, γ' indicates the increase in SIRatio associated with each one-unit increase in urbanization index, and δ' indicates the increase in SIRatio associated with each 100 cigarettes sold per capita. In this model, SIRatio (a rate ratio) needs to be forced to take the value 1 when the arsenic exposure is within the reference category (" < 0.05 mg/L") and all other variables are set to their reference categories. This can be accomplished through coding the " < 0.05 mg/L" group as the reference category.
where CIR is the crude cumulative incidence rate, Xj, U, and T are defined as in Model 1, and Ak is the proportion (as percentage) of residents in age group k in each township. Because there are seven age groups, six independent variables derived from dummy variables at the individual level were used in the regression model . Therefore, the age group "0-19 years" was used as the reference, A1 = percentage of residents in the age group "20-29 years," A2 = percentage of residents in the age group "30-39 years," and so on. In this case, α" is the estimated background cumulative incidence rate; βj", γ", and δ" are defined as βj, γ, and δ in Model 1 respectively; and θk indicates the RD associated with each 1% increase in residents in age group k.
Models 1 and 3 generate estimates of RD's, but Model 2 generates estimates of incremental rate ratios. Therefore, estimates of RD's from Models 1 and 3 were then divided by the estimates of background rates (α and α" respectively) to obtain estimates of incremental rate ratios to facilitate the comparison among the three methods.
Estimates of incremental relative risks (IRRs) for bladder cancer of men obtained by three different methods
> 0.64 mg/L
> 69 years
p Value for the Modelf
Estimates of incremental relative risks (IRR) for women obtained by three different methods
> 0.64 mg/L
> 69 years
p Value for the Modelf
All three methods showed a significant positive association between urbanization index and incidence of bladder cancer for men, but the Indirect Method gave a lower risk estimate than the other two. (Table 2) Again, although all three methods showed a positive association between urbanization index and incidence of bladder cancer for women, when the Variable Method was used, the association would be determined as not significant by the general practice of setting the significant level at 0.05 (p = 0.16). (Table 3)
For both men and women, no significant association was found between cigarette sales and incidence of bladder cancer by any of the methods. (Tables 2 and 3) The risk estimates obtained by different methods were quite different, and that obtained by the Indirect Method was between those obtained by the Direct and the Variable Methods for both genders.
To determine the validity of these methods, comparisons between the regression models for ecological data with those for individual data need to be made. Because Rosenbaum and Rubin have shown that the first approach cannot generate unbiased risk estimates and that the third approach can generate unbiased risk estimates, the focus of discussion is placed on the validity of the second approach.
where, as defined in Model 2, Xj denotes the percentage of residents in category j, and T is the average number of cigarettes (in hundreds) sold per capita. From Model 5 at the individual level to Model 6 at the township level, all regression coefficients (Bj', C', and F') remain the same. Comparing Model 2 with Model 6, one can find that while the independent variables X j , U, and T are the same, SIRatio, which has been adjusted for age, is not identical or proportional to N/nS. Therefore βj', γ', and δ' are not unbiased estimates of Bj'/100, C', and 100F', and α' does not necessarily has an expected value of 1.
This study showed that different methods of age-adjustment may lead to different results and that the two methods (Direct and Indirect) frequently applied to adjust for age in ecological studies can not generate unbiased risk estimates, although the control of the effect of age may be achieved. Even though none of the risks associated with age groups was significant in this study, application of either method might lead to different conclusions, such as a significant positive association between urbanization index and bladder cancer women. On the other hand, the Variable Method not only adjusts for the effects of age, but also generates risk estimates associated with different age groups. Therefore, even though the Direct and Indirect Methods may generate valid age-adjusted risk estimates for individual unit population directly for the purposes of comparison among populations, one should not use these estimates as dependant variables for the purpose of age-adjustment when applying regression models to analyze ecological data.
When the effects of age themselves are not of primary interest, a reasonable alternative to conduct age-adjustment in ecological studies is to use the mean age of each unit population as an independent variable in the regression model. But, a study showed that this approach is based on the assumption of a linear dose-response relationship between age and the outcome of interest . That study also showed that such assumptions are not necessary when one uses independent variables derived from dummy variables at the individual level in the regression models, as in the Variable Methods described above. When more data are available, other approaches may also be applied. For example, when we have age-specific data on both dependent and independent variables, we may conduct separate regression analyses for different age groups. However, such data are often unavailable, as in the case presented in this paper. We should also note that the transformation of Model 5 at the individual level to Model 6 at the township level only holds for linear regression, not for logistic or Poisson regressions, although they may be more appropriate for some cases.
The occurrence of urinary cancers has been noted among arsenic intoxicated patients in the 1950s , and high incidence of urinary cancers associated with arsenic in drinking water has reported in Taiwan [9, 19] and many other countries [27–32]. Users of Fowler's solution (containing potassium arsenite)  and wine growers exposed to arsenical pesticides through drinking alcoholic beverages and spraying pesticides  had increased risks of bladder cancer. Therefore, the associations between arsenic ingestion and bladder cancer observed in this study are supported by the scientific literature. The data on the dose-response relationships, however, are quite limited. A study in Japan  observed cases only at the highest level among the three studied, and other studies in Taiwan also support the association between exposure to arsenic levels above 0.3 mg/L in drinking water and mortality of bladder cancer [14, 35]. The irregularities in the dose-response relationships at lower exposure levels observed in this study among women might due to the effects of un-controlled confounders or the cell-types specificity of the carcinogenic effect of arsenic on the urinary system . Therefore, the association between exposure to arsenic levels below 0.3 mg/L in drinking water and occurrence of bladder cancer needs further evaluation. In fact, a recent meta-analysis showed that arsenic levels below < 0.2 mg/L alone did not appear to be a significant independent risk factor for bladder cancer .
Absence of an association between cigarette sales and bladder cancer in this study is consistent with the findings of a previous study in Taiwan which showed associations between smoking and lung cancer, but not bladder cancer, after adjusting for exposures to arsenic . The number of cigarettes sold in a township might not be a good surrogate measurement of the number of cigarettes actually consumed by the residents, and it is also possible that excess risks associated with smoking were too small to be detected by either study.
The ecologic study presented in this paper shares with all the other ecologic studies several major limitations inherent in the ecological study design, such as "ecological fallacy." Although it might be minimized by using smaller population units such as "village," this study had to use township as the unit because the National Cancer Registry Program coded the residences of cases by township. On the other hand, for a relatively rare disease like bladder cancer among Taiwanese, estimates of incidence rates will be unstable if the unit population is too small, especially when the duration of observation is relatively short. Risk estimates generated by this study might be affected by possible incomplete reporting of cancer cases because the reporting is not mandatory by law. If there is a correlation between reporting rates and arsenic exposure levels, proper validation studies on the registry, which are not currently available, are necessary to determine the direction and magnitude of the biases. Nonetheless, in a previous Taiwan study on skin cancer, which has a much more serious problem of under-reporting than bladder cancer because of its low case fatality rate, the comparison between the relative risks obtained by analyzing cancer registry data and those obtained by conducting physical examinations on all study participants to achieve complete case ascertainment showed that the possible under-reporting had little effect on the estimates of relative risks . Except for the " < 0.05 mg/L" category (the reference exposure category), no township had all the wells in a single exposure category, and so it is impossible to validate the risks estimates in the current study. In addition, the current study cannot account for the migration of residents of each township over the 10-year study period. Further studies with exposure data for each individual (such as case-control or cohort studies) as well as studies evaluating the effects of other co-existing risk factors are needed to confirm the hypotheses generated in this study.
Although ecological study design has some major limitations, it is widely applied in epidemiology. Like in other types of study designs, adjustment for age is often necessary in ecological studies, but different methods may lead to different conclusions. The current study compared three common approaches of age adjustment using a study on the associations between arsenic in drinking water and incidence of bladder cancer in 243 townships in Taiwan was used as an example. Whereas all three methods showed a statistically significant positive association between arsenic exposure above 0.64 mg/L and incidence of bladder cancer in both genders, they reached different conclusions on the other exposure categories. Even for the category above 0.64 mg/L, the risk estimates obtained by different methods were different. Using proportions of residents in different age groups as a part of the independent variables in the multiple regression models is the only approach of the three that can obtain unbiased estimates of the risks and also generate estimates of the risk associated with each age group. This approach is recommended for age adjustment in ecological studies.
crude cumulative incidence rate
incremental relative risk
standardized incidence rate
standardized incidence ratio.
This work was funded in part by Grants NSC-87-2314-B006-090 and NSC-89-2320-B006-015 from the National Science Council, Taiwan, R.O.C. The author would also like to thank the Taiwan National Cancer Registry Program and the Taiwan Provincial Department of Environmental Protection for their support in the conduct of this study, Ms. Shu-Fong Tsai for providing data on cigarette sales, and Dr. Chao-Yu Guo for assisting the revision of the manuscript.
- Walter SD: The ecological method in the study of environmental health. I. Overview of the method. Environ Health Perspect. 1991, 94: 61-65.View ArticlePubMedPubMed CentralGoogle Scholar
- Monson RR: Occupational Epidemiology. 1990, Boca Raton: CRC Press, Inc, 72-74. 2Google Scholar
- Greenland S, Rothman KJ: Introduction to Categorical Statistics in Modern Epidemiology. 1998, Philadelphia: Lippincott-Raven Publishers, 2Google Scholar
- Rosenbaum PR, Rubin DB: Difficulties with regression analyses of age-adjusted rates. Biometrics. 1984, 40: 437-443. 10.2307/2531396.View ArticlePubMedGoogle Scholar
- Guo H-R, Lipsitz SR, Hu H, Monson RR: Using ecological data to estimate a regression model for individual data: the association between arsenic in drinking water and incidence of skin cancer. Environ Res. 1998, 79: 82-93. 10.1006/enrs.1998.3863.View ArticlePubMedGoogle Scholar
- Ocaña-Riola R: Common errors in disease mapping. Geospatial Health. 2010, 4: 139-154.View ArticlePubMedGoogle Scholar
- Tsai S-F: Ecological Correlation Study on major Risk Factors of Various Malignant Neoplasms in Taiwan; 1972-1983. Master thesis. 1987, National Taiwan University, Department of Public HealthGoogle Scholar
- Chen C-J, Wang C-J: Ecological correlation between arsenic level in well water and age-adjusted mortality from malignant neoplasms. Cancer Res. 1990, 50: 5470-5474.PubMedGoogle Scholar
- Guo H-R, Chiang H-S, Hu H, Lipsitz SR, Monson RR: Arsenic in drinking water and incidence of urinary cancers. Epidemiology. 1997, 8: 545-550. 10.1097/00001648-199709000-00012.View ArticlePubMedGoogle Scholar
- Yeh S: Relative incidence of skin cancer in Chinese in Taiwan: with special reference to arsenical cancer. Natl Cancer Inst Monogr. 1963, 10: 81-107.Google Scholar
- Chen K-P, Wu H-Y, Yeh C-C, Cheng Y-J: Color Atlas of Cancer Mortality by Administrative and Other Classified Districts in Taiwan Area: 1968-1976. 1979, Taipei: National Science Council, Taiwan R.O.CGoogle Scholar
- Tseng WP, Chu HM, How SW, Fong JM, Lin CS, Yeh S: Prevalence of skin cancer in an endemic area of chronic arsenicism in Taiwan. J Natl Cancer Inst. 1968, 40: 453-463.PubMedGoogle Scholar
- Chen C-J, Chuang Y-C, Lin T-M, Wu H-Y: Malignant neoplasms among residents of a blackfoot disease endemic area in Taiwan: high-arsenic artisan well water and cancers. Cancer Res. 1985, 45: 5895-5899.PubMedGoogle Scholar
- Wu M-M, Kuo T-L, Hwang Y-H, Chen C-J: Dose-response relation between arsenic concentration in well water and mortality from cancers and vascular diseases. Am J Epidemiol. 1989, 130: 1123-1132.PubMedGoogle Scholar
- Brown KG, Chen C-J: Significance of exposure assessment to analysis of cancer risk from inorganic arsenic in drinking water in Taiwan. Risk Analysis. 1995, 15: 475-484. 10.1111/j.1539-6924.1995.tb00340.x.View ArticlePubMedGoogle Scholar
- Guo H-R: The lack of a specific association between arsenic in drinking water and hepatocellular carcinoma. J Hepatol. 2003, 39: 383-388. 10.1016/S0168-8278(03)00297-6.View ArticlePubMedGoogle Scholar
- Guo H-R: Arsenic level in drinking water and mortality of lung cancer. Cancer Cause Control. 2004, 15: 171-177.View ArticleGoogle Scholar
- Guo H-R, Wang N-S, Hu H, Monson RR: Cell-type specificity of lung cancer associated with arsenic ingestion. Cancer Epidemiol Biomarker Prev. 2004, 13: 638-643.Google Scholar
- Guo H-R, Chen C-J, Greene HL: Arsenic in drinking water and cancers: A descriptive review of Taiwan studies. Environ Geochem Health. 1994, 16 (suppl): 129-138.Google Scholar
- Lo M-C, Hsen Y-C, Lin B-K: Second Report on the Investigation of Arsenic Content in Underground Water in Taiwan. 1977, Taichung: Taiwan Provincial Institute of Environmental SanitationGoogle Scholar
- American Public Health Association: Standard Methods of the Evaluation of Water and Water Waste. 1985, Washington, DC: American Public Health AssociationGoogle Scholar
- Kuo T-L: Investigation on the arsenic levels in well water in the blackfoot disease endemic area. Chin J Public Health. 1996, 15 (suppl): 116-125.Google Scholar
- World Health Organization: International Classification of Diseases for Oncology. 1976, Geneva: World Health Organization, 1Google Scholar
- Wu S-Y: Report on the Degree of Urbanization of Precincts and Townships in Taiwan. 1986, Taipei: Bureau of Statistics, Executive Yuan, Taiwan, R.O.CGoogle Scholar
- Waterhouse J, Shanmugaratnam K, Muir C, Powell J: Cancer Incidence in Five Continents. 1976, Lyon: IARC International Agency for Research on Cancer, 3:Google Scholar
- Sommers SC, McManus RG: Multiple arsenical cancers of the skin and internal organs. Cancer. 1952, 6: 347-359.View ArticleGoogle Scholar
- Biagini RE, Quiroga GC, Elias V: Chronic hydroarsenism in Urutau. Arch Argent Dermatol. 1974, 24: 8-11.Google Scholar
- Besuschio SC, Prez Desanzo AC, Croci M: Epidemiological association between arsenic and cancer in Argentina. Biol Trace Element Res. 1980, 2: 41-55. 10.1007/BF02789034.View ArticleGoogle Scholar
- Tsuda T, Babazono A, Yamamoto E, Kurumatani N, Mino Y, Ogawa T, Kishi Y, Aoyama H: Ingested arsenic and internal cancer: A historical cohort study followed for 33 years. Am J Epidemiol. 1995, 141: 198-209.PubMedGoogle Scholar
- Bates MN, Smith AH, Cantor KP: Case-control study of bladder cancer and arsenic in drinking water. Am J Epidemiol. 1995, 141: 523-530.PubMedGoogle Scholar
- Hopenhayn-Rich C, Biggs ML, Fuchs A, Bergoglio R, Tello EE, Nicolli H, Smith AH: Bladder cancer mortality associated with arsenic in drinking water in Argentina. Epidemiology. 1996, 7: 117-124. 10.1097/00001648-199603000-00003.View ArticlePubMedGoogle Scholar
- Han YY, Weissfeld JL, Davis DL, Talbott EO: Arsenic levels in ground water and cancer incidence in Idaho: an ecologic study. Int Arch Occup Environ Health. 2009, 82: 843-849. 10.1007/s00420-008-0362-9.View ArticlePubMedGoogle Scholar
- Cuzick J, Sasieni P, Evans S: Ingested arsenic, keratoses, and bladder cancer. Am J Epidemiol. 1992, 136: 417-421.PubMedGoogle Scholar
- Lchtrath H: The consequences of chronic arsenic poisoning among Moselle wine growers. J Cancer Res Clin Oncol. 1983, 105: 173-182. 10.1007/BF00406929.View ArticleGoogle Scholar
- Chen C-J, Kuo T-L, Wu M-M: Arsenic and cancers. Lancet. 1988, 1: 414-415.View ArticlePubMedGoogle Scholar
- Mink PJ, Alexander DD, Barraj LM, Kelsh MA, Tsuji JS: Low-level arsenic exposure in drinking water and bladder cancer: a review and meta-analysis. Regul Toxicol Pharmacol. 2008, 52: 299-310. 10.1016/j.yrtph.2008.08.010.View ArticlePubMedGoogle Scholar
- Chiou H-I, Hsueh Y-M, Liaw K-F, Horng S-F, Chiang M-H, Pu Y-S, Lin J-S, Huang C-H, Chen C-J: Incidence of internal cancers and ingested inorganic arsenic: A seven-year follow-up study in Taiwan. Cancer Res. 1995, 55: 1296-1300.PubMedGoogle Scholar
- Guo H-R: Arsenic in drinking water and skin cancer: Comparison among studies based on cancer registry, death certificates, and physical examinations. Arsenic: Exposure and health effects. Edited by: Abernathy CO, Calderon RL, Chappell WR. 1997, London: Chapman & HallGoogle Scholar
- The pre-publication history for this paper can be accessed here:http://www.biomedcentral.com/1471-2458/11/820/prepub
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.