SEARCH

SEARCH BY CITATION

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgment
  8. Disclosure
  9. REFERENCES
  10. Supporting Information

Objective: Given links between obesity and cancer, we estimated incident cancer burden due to overweight and obesity at the state level in the United States.

Methods and Procedures: Using state rankings by per capita burden of incident cancer cases diagnosed in 2003 that were related to overweight and obesity, we examined the frequency with which states ranked in the highest and lowest quintiles of weight-related burden for cancers of the postmenopausal breast, endometrium, kidney, colon, and prostate. In this study, data from the Behavioral Risk Factor Surveillance System (BRFSS), US Census, US Mortality Public Use Data Tapes, and National Cancer Institute Surveillance, Epidemiology, and End Results (SEER) Program were used.

Results: Western states had the lowest weight-related cancer burden for both sexes. Iowa, South Dakota, and West Virginia had the highest burden for all three types of male cancers. West Virginia is the only state that ranked in the quintile of highest weight-related burden for all four cancers considered in women.

Discussion: For certain cancers, including endometrial, postmenopausal breast, and colon cancers, states with high burdens clustered in geographic regions, warranting further inquiry. Although state ranks for the total cancer burden and the prevalence of overweight and obesity correlated with state ranks for weight-related incident cancer burden, they often served poorly as its proxy. Such a finding cautions against simply targeting states with high overweight and obesity or high total burdens of cancers for which overweight and obesity are risk factors, as this approach may not reach areas of unrecognized burden.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgment
  8. Disclosure
  9. REFERENCES
  10. Supporting Information

Given the rising prevalence of obesity across a number of states in the United States (1), understanding the magnitude and breadth of the growing problem of overweight and obesity in terms of their burden on society is needed. Efforts have been made to estimate the economic costs (2,3), the loss of productivity (4,5), the disability and years of life lost (6,7), and the proportion of overall mortality related to overweight and obesity (8). Considerable evidence has also accumulated to support links between overweight and obesity and increased risk of several types of cancer, including colon, endometrial, kidney, and postmenopausal breast (9). However, only limited efforts have been made to characterize the burden of cancer that is related to overweight and obesity. The few reports that have focused on the United States have described the burden for the entire country (10,11), but did not look at state-level differences. Although national estimates can be useful for population-based planning strategies, they may not be applicable for state-level activities where most cancer control programs are planned and implemented (12).

In this study, we describe one approach to characterize the US state-level burden of incident cancer diagnosed among adults in 2003 related to excess bodyweight for selected types of cancer: colon, endometrial, kidney, postmenopausal breast, and prostate. The approach was based on a method described by Bergström et al. (13) that estimated the weight-related incident cancer case burden for 15 European countries. We extended this approach by conducting a sensitivity analysis to investigate the potential impact of statistical uncertainty on the estimation of weight-related burden of incident cancer. We found that the estimates of weight-related incident cancer cases per state were sensitive to the risk ratios (RRs) used, but the ranking of states by those estimates were more stable. Accordingly, we present the ranks of states by incident cancer burden associated with overweight and obesity rather than the estimates of weight-related incident cancer cases per state.

Methods and Procedures

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgment
  8. Disclosure
  9. REFERENCES
  10. Supporting Information

The steps taken in this analysis are depicted in Figure 1. To estimate the disease burden attributable to overweight and obesity for each state, we first calculated the “population attributable risk proportion” (PARP) for overweight and obesity per state, a standard epidemiological approach (14). This proportion was used with the expected total number of newly diagnosed cancer cases to calculate the number of incident cancer cases due to overweight and obesity. Of various PARP formulas available, we used the only one that incorporated the variance of the exposure (i.e., state prevalence rates of overweight and obesity) needed to estimate the unique weight-related cancer burden for each state (14). This formula required use of unadjusted RRs, which were not directly available from the report by Bergström et al. (13). Therefore, we converted the adjusted RRs of Bergström et al. (13) to unadjusted RRs before applying the PARP formula (see below, “Sensitivity Analysis”). In our analysis, we relied on the studies included in the meta-analysis conducted by Bergström et al. (13). Although this method excluded more recently published studies and retained other studies that might inadequately reflect the experience in the United States (e.g., studies conducted outside the United States or having small numbers of ethnic minorities), doing so helped in keeping our approach similar to that used by Bergström et al. (13).

image

Figure 1. Analytic steps to estimate state-level cancer case burdens and ranks due to overweight and obesity in the United States. NCI SEER, National Cancer Institute Surveillance, Epidemiology, and End Results.

Download figure to PowerPoint

In accordance with the work of Bergström et al. (13), we describe our results separately by state and sex for the following types of cancer: colon, kidney, prostate, endometrial, and postmenopausal breast. We did not analyze data for gallbladder cancer because case numbers by sex in our sample were insufficiently sized to produce stable estimates at the state level. Even though the evidence supporting an association between risk and obesity is presently inconclusive (15), we evaluated data for prostate cancer, one of the leading incident cancers among men.

We ranked the 50 states and Washington, DC from highest (#1) to lowest (#51) burden for each type of cancer by sex, divided states into quintiles and compared (i) how often they were ranked in the quintiles of highest and lowest burden, and (ii) the quintile differences between state rankings for weight-related incident cancer, total cancer burden (i.e., total number of cancer cases per 100,000 state residents), and prevalence of overweight and obesity. (For some cancer/sex combinations, insufficient data were available to estimate the state burden, and fewer than 51 states were divided into quintiles. When the number of states being divided into quintiles was 51 and 41, we included the additional state in the middle quintile (e.g., 10, 10, 11, 10, 10). When the number of states being divided was 48, the quintile sizes were 8, 9, 9, 9, and 8). We chose to focus only on changes and differences across quintiles, rather than smaller changes and differences in rank, to focus on larger differences that were more likely to be meaningful. We used Spearman's rank correlation (16) to assess a variety of comparisons including the relationship between weight-related cancer burden and sex.

The prevalence of overweight and obesity by state

To obtain the state-level prevalence of overweight and obesity, we used the 2002 Behavioral Risk Factor Surveillance System (BRFSS) data from the Centers for Disease Control and Prevention (http:www.cdc.govbrfsstechnicalinfodatasurveydata2002.htm). Individuals were grouped by BMI (calculated as weight in kilograms divided by height squared in meters) into three weight categories: BMI <25 kg/m2, BMI = 25.0 to 29.9 kg/m2 (overweight), and BMI ≥30 kg/m2 (obese) (17). We then estimated for each state the proportion of individuals in the weight categories described above, by incorporating the BRFSS-supplied sampling weights that adjusted for study design, nonresponse (i.e., refusal to participate), and noncoverage (i.e., inability to survey people without telephones). The weights were also adjusted for oversampling and undersampling by gender, age, and race and ethnicity, so that the overall weighted estimates were representative of the US population in 2003. Finally, we adjusted the proportions by age distribution to the 2000 Census standard population (18). This final step was performed to make the population proportions comparable with other data sources used in the analyses. States ranked by the prevalence of overweight and obesity combined are presented in Figure 2 and Table 1.

image

Figure 2. US census regions are outlined on the map as follows: the Northeast (Maine, New Hampshire, Vermont, Massachusetts, Rhode Island, Connecticut, New York, New Jersey, Pennsylvania); the Midwest (Ohio, Indiana, Illinois, Michigan, Wisconsin, Minnesota, Iowa, Missouri, North Dakota, South Dakota, Nebraska, Kansas); the South (Delaware, Maryland, District of Columbia, Virginia, West Virginia, North Carolina, South Carolina, Georgia, Florida, Kentucky, Tennessee, Alabama, Mississippi, Arkansas, Louisiana, Oklahoma, Texas); and the West (Montana, Idaho, Wyoming, Colorado, New Mexico, Arizona, Utah, Nevada, Washington, Oregon, California, Alaska, Hawaii) (22). (a) Prevalence of overweight and obesity among men, 2002. (b) Prevalence of overweight and obesity among women, 2002.

Download figure to PowerPoint

Table 1.  Age-adjusted prevalence of overweighta and obesity, and rank by total excess weight for men and women aged ≥20 years by state and for the United States, 2002b
inline image

State-specific PARP associated with overweight and obesity

We calculated the PARP using the computational formula shown below (Formula #5, Table 1 in Rockhill et al. (14)). This calculation produced a specific PARP for each state based on the prevalence rates of overweight and obesity for that specific state. All procedures were performed separately by sex for each type of cancer.

  • image

where pi is the proportion of population in i th exposure level: overweight (i = 1) and obese (i = 2), RRi is the crude RR comparing i th exposure level: overweight (i = 1) and obese (i = 2) with unexposed group (normal weight).

Calculation of cancer case burden associated with overweight and obesity

Calculating the precise state- and sex-specific incident cancer case burden associated with overweight and obesity requires the total number of new cancer cases diagnosed for each type of cancer for each state, by sex. As complete cancer registration has not yet been achieved in many states and the precise number of new cases diagnosed each year in every state is unknown, the numbers of new cancer cases for each state had to be estimated. To estimate cases for each state, we used the procedure developed by the American Cancer Society (19). This procedure obtains the expected total number of new cases diagnosed in the year of analysis for each type of cancer being evaluated. We estimated this number using an autoregressive quadratic time trend model based on cancer mortality data (1979–2002, National Center for Health Statistics (20)), age-specific, cancer-specific incidence rates (1979–2000, National Cancer Institute Surveillance, Epidemiology, and End Results (SEER) Program (21)), and population data (US Census Bureau (18)).

The state-specific estimated total number of new cancer cases diagnosed in 2003 for each sex was then multiplied by the state- and sex-specific PARP to obtain for each type of cancer the number of incident cases associated with overweight and obesity per state. (Note: the PARP was first divided by 100 to multiply as a proportion and not a percentage.) When the estimated total number of new cases for a specific type of cancer in a state was ≥50 (<50 cases were likely to generate unstable estimates), we calculated the number of cancer cases attributed to excess weight. For endometrial cancer, the number of incident cases could not be estimated because the classification of corpus uteri cancers by anatomic subsite from SEER cancer registries did not allow us to identify cases of endometrial cancer separately. Before 1992, <5% of the corpus uterine cancer and cancer of the uterus that were not otherwise specified cases diagnosed in the SEER areas were classified as endometrial cancers compared with as high as 95% of the cases after 1992 (21). Therefore, we used cancer of the corpus uterus, not otherwise specified as a surrogate for endometrial cancer because the majority of cancer cases for corpus and uterus not otherwise specified were endometrial cancer cases.

As states with larger populations would be expected to have higher absolute numbers of cancer cases, we standardized the case burdens per state population by calculating the “incident cancer burden per 100,000” associated with overweight and obesity. Weight-related cancer burden estimates were calculated by taking the number of new cancer cases associated with overweight and obesity per state and dividing by the sex-specific state populations as delineated in 5-year age categories by the 2000 US Census for adults ≥20 (except for postmenopausal breast cancer) and then finally multiplying by 100,000 (18). For postmenopausal breast cancer, we used the US Census state populations for women aged ≥50 years to calculate case burden due to excess weight (18). To provide a visual depiction of the state-specific estimated cancer burden associated with overweight and obesity, US maps were created using ArcGIS software 9.0 (ESRI, Charlotte, NC) on which individual states and US census bureau regions were outlined (22).

Sensitivity analysis

Because of the potential for introducing statistical uncertainty from modifying and applying the approach used by Bergström et al. (13), we conducted a sensitivity analysis. This analysis estimated the extent of the possible bias introduced by “crude” RRs (i.e., modified age-adjusted RRs) from a plausible range (i.e., lower and upper bounds created by the 25th and 75th percentiles) of RRs taken from the individual studies included in Bergström's meta-analysis (13) to provide the state-level number of incident cancer cases associated with overweight and obesity per 100,000 state residents (i.e., the state-level weight-related incident cancer case burden). Applying this range of RRs allowed us to evaluate the impact of variation in RRs on calculating the state cancer case burdens due to excess bodyweight, providing some sense of the error that might result from bias in our risk estimate. This approach had the advantage of being based directly upon results obtained from the many peer-reviewed studies that collectively represented a plausible range of RRs for the association between excess bodyweight and cancer risk (23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,3940,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57).

We calculated the crude RR for each individual study included in the Bergström's meta-analysis (13) and then computed an adjustment factor (AF) as:

  • image

These values were then ranked across studies by sex for each type of cancer and we selected the AFs at the median, 25th and 75th percentiles to set plausible values (see below) for RRs used to estimate the state-specific PARP for each type of cancer by sex.

To calculate the lower and upper bounds and the median for the RRoverweight and RRobese respectively, we used the following formulas:

  • image

were, RRBergström is the estimated dose-response per unit increase in BMI; LB 95% CI is the lower bound of the 95% confidence interval (CI) for RRBergström; and UB 95% CI is the upper bound of the 95% confidence interval (CI) for RRBergström. (Note: Multiplying by 5 and 10, respectively represented the unit change in BMI from the normal to overweight and from the normal to obese categories). These choices for upper and lower bounds resulted in a wide range of RRs within each cancer site. To see the impact of different RRs on the estimated state cancer burden, we examined the ratio of the estimate for the state with the highest weight-related burden to the estimate for the state with the lowest burden for each set of RRs (from Bergström's analysis and our three scenarios in the sensitivity analysis). This ratio captured both the intrastate and interstate variation resulting from using each set of RRs, where a ratio of 1.00 would indicate no effect of variation of the RRs on the case burden estimates.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgment
  8. Disclosure
  9. REFERENCES
  10. Supporting Information

Sensitivity analysis

The result of our sensitivity analysis indicated that the state case burdens of cancer due to excess bodyweight were greatly influenced by the choice of RR used in the PARP formula; we observed large variation in the state-level weight-related cancer case burdens as a result of varying the RRs (Table 2). For example, in Nevada, the state with the lowest burden, the estimated weight-related case burden of endometrial cancer per 100,000 women was more than doubled when RRs for the upper bound were applied (11.19 cases per 100,000 women) as compared with when the RRs for the lower bound were used (4.66 cases per 100,000 women). By contrast, the estimates for the District of Columbia, which had the highest burden, were 18.00 cases per 100,000 women and 41.83 cases per 100,000 women based on the upper and lower bounds for the RRs, respectively.

Table 2.  Risk ratio (RR) for overweight (OW) and obese (OB) and range of state cancer case burdens associated with excess bodyweight per 100,000 sex-specific state residents by RRs (i.e., Bergström, lower and upper bounds, median) and type of cancer
inline image

As expected, the ratios of the estimates for the state with the highest weight-related burden to the estimates for the state with the lowest burden for each set of RRs were constant regardless of scenario (upper or lower bound; median) but varied by type of cancer and sex: 2.50, colon for men; 3.35, colon for women; 2.46, kidney for men; 2.38, kidney for women; 3.80, endometrial; 2.21, postmenopausal breast; and 3.45, prostate. Although specific case burdens varied considerably in response to varying the RRs, the state rankings of the weight-related case burdens were less strongly affected.

For each type of cancer and for both sexes separately, the ranked state burdens (i.e., cases due to overweight and obesity per 100,000) from the four sets of RRs (i.e., Bergström, lower bound, median, and upper bound) were highly correlated (r > 0.97, P < 0.0001 for all correlations, data not shown). In general, few states changed rank when different RRs were applied and of those that did, the majority changed rank by only one or two places. Such a result indicates that using RRs chosen from a wide range of possible RRs had only a modest impact on the ordered state ranks. Therefore, we focused on state ranks rather than on absolute state cancer case burdens associated with excess bodyweight. The remainder of the findings regarding PARP and rank of estimated weight-related cancer case burden per 100,000 are based on the median RR value computed from the sensitivity analysis.

Burden of incident cancer associated with excess bodyweight by cancer type

When states were sorted by burden of incident cancer associated with overweight and obesity, three ranked in the highest quintile of states for the three types of cancer evaluated among men (Table 3): Iowa, South Dakota, and West Virginia. Among women, West Virginia also ranked in the highest quintile of burden for all four types of cancer examined, the only state to do so (Table 4). Four other states ranked in the highest quintile for three of four types of cancer examined among women: Indiana, Iowa, Mississippi, and Pennsylvania. In terms of lowest weight-related cancer burden, three states were ranked in the lowest quintile for all three types of cancer evaluated among men: California, Colorado, and Utah. Only Colorado consistently ranked in the lowest quintile of weight-related incident cancer burden for the four types of cancer examined among women. In general, states with high burden of weight-related incident cancer did not cluster by geographic region. However, most of the states with the lowest weight-related burden of incident cancer were in the West for both men and women. Patterns of weight-related cancer burden for specific types of cancer are described as follows.

Table 3.  State ranks from lowest (51) to highest (1)a by burden of cancer associated with excess bodyweight and type of risk ratio (RR)b used in population attributable risk proportion (PARP) equation among adult men aged ≥20 years
inline image
Table 4.  State ranks from lowest (51) to highest (1)a by burden of cancer associated with excess bodyweight and type of risk ratio (RR)b used in population attributable risk proportion (PARP) equation among adult women aged ≥20 years
inline image

Endometrial cancer. The majority of states ranking in the highest quintile of incident endometrial cancer burden due to overweight and obesity were in the Midwest and the Northeast Of these states, New Jersey, New York, Ohio, Pennsylvania, Virginia, West Virginia, and Washington, DC clustered together in the Mid-Atlantic region (Figure 3a). Several of these states (i.e., Ohio, Pennsylvania, and West Virginia) also ranked among those that had the highest burden of weight-related incident postmenopausal breast cancer. Most of the states in the lowest quintile of weight-related endometrial cancer burden were in the West: Arizona, Colorado, Idaho, Nevada, New Hampshire, Oklahoma, Rhode Island, South Dakota, and Utah.

image

Figure 3. US census regions are outlined on the map as follows: the Northeast (Maine, New Hampshire, Vermont, Massachusetts, Rhode Island, Connecticut, New York, New Jersey, Pennsylvania); the Midwest (Ohio, Indiana, Illinois, Michigan, Wisconsin, Minnesota, Iowa, Missouri, North Dakota, South Dakota, Nebraska, Kansas); the South (Delaware, Maryland, District of Columbia, Virginia, West Virginia, North Carolina, South Carolina, Georgia, Florida, Kentucky, Tennessee, Alabama, Mississippi, Arkansas, Louisiana, Oklahoma, Texas); and the West (Montana, Idaho, Wyoming, Colorado, New Mexico, Arizona, Utah, Nevada, Washington, Oregon, California, Alaska, Hawaii) (22). (a) Estimated incident endometrial cancer cases per 100,000 women due to overweight and obesity, 2003. (b) Estimated incident postmenopausal breast cancer cases per 100,000 women aged ≥50 years due to overweight and obesity, 2003. (c) Estimated incident colon cancer cases per 100,000 men due to overweight and obesity, 2003. (d) Estimated incident colon cancer cases per 100,000 women due to overweight and obesity, 2003. (e) Estimated incident kidney cancer cases per 100,000 men due to overweight and obesity, 2003. (f) Estimated incident kidney cancer cases per 100,000 women due to overweight and obesity, 2003. (g) Estimated incident prostate cancer cases per 100,000 men due to overweight and obesity, 2003.

Download figure to PowerPoint

Postmenopausal breast cancer. Half of the states ranking in the highest quintile of incident weight-related postmenopausal breast cancer burden were in the Midwest: Illinois, Indiana, Michigan, North Dakota, and Ohio Four of these Midwestern states clustered together in the eastern part of the region that bordered two states (Pennsylvania and West Virginia) that also ranked in the highest quintile of postmenopausal breast cancer burden due to overweight and obesity, but were in other census regions (Figure 3b). Six of the ten states with the lowest ranked burden of postmenopausal breast cancer due to overweight and obesity were in the West: Colorado, Hawaii, Montana, Nevada, New Mexico, and Utah.

Colon cancer. By state, the rank for burden of weight-related incident colon cancer was generally similar for both men and women (r = 0.78, P < 0.0001). For men, while most states in the highest quintile of burden due to overweight and obesity were in the South (i.e., Kentucky, Louisiana, Mississippi, Oklahoma, and West Virginia), some of these states clustered in two areas: the Midwest (i.e., Iowa, Nebraska, and South Dakota) and Appalachia, specifically, Kentucky, Pennsylvania, and West Virginia (Figure 3c). For women, states with the highest weight-related colon cancer burden were more often in the Midwest than in the other census regions: Indiana, Iowa, Nebraska, and South Dakota. For men, three states with the highest weight-related colon cancer burden, Iowa, Nebraska, and South Dakota, were geographically contiguous (Figure 3d). In addition to Iowa, Nebraska, and South Dakota, three other states also ranked in the highest quintile of weight-related colon cancer burden for both men and women: Mississippi, Pennsylvania, and West Virginia.

The majority of states in the quintile of lowest weight-related colon cancer for both men and women clustered in the West census region. Eight of these states were the same for both women and men: Alaska, California, Colorado, Hawaii, Idaho, New Mexico, Utah, and Washington.

Kidney cancer. The state burden of weight-related incident kidney cancer was correlated by sex, but only moderately (r = 0.50, P = 0.0009). Four states ranked for both men and women among the top ten states with the highest burden: Indiana, Iowa, Oklahoma, and West Virginia (Figure 3e, f). Five states among the top ten states with the highest burden among men were in the Midwest, and four states clustered together: Iowa, Nebraska, North Dakota, and South Dakota. For women, several states within the highest quintile of weight-related kidney cancer burden were also in the Midwest (i.e., Indiana, Michigan, and Missouri). Four states with burdens among the top quintile of burden for women, including Missouri, Oklahoma, and some in the South (i.e., Arkansas, Louisiana), clustered together.

For both men and women, the West had the largest number of states ranking among the lowest quintile of weight-related kidney cancer burden. States ranking in the lowest quintile of burden for both men and women included California, Colorado, Georgia, Maryland, and New York. Some states with low kidney cancer burden among women clustered together in southern New England (i.e., Connecticut, Massachusetts, New York).

Prostate cancer. By census region, the majority of states with the highest ranked cancer burden for weight-related prostate cancer were in the South (i.e., Alabama, Arkansas, Florida, Louisiana, South Carolina, and West Virginia), including four contiguous states (Figure 3g). Most states ranked in the lowest quintile for weight-related prostate cancer burden were in the West: Alaska, California, Colorado, Hawaii, Utah, and Washington. Four other states ranking in the lowest quintile by prostate cancer burden due to overweight and obesity clustered together in the Northeast: Maine, New Hampshire, New York, and Vermont.

Using total incident cancer burden as proxy for weight-related cancer burden

The correlation between state ranks for total cancer burden and those for the incident cancer burden associated with excess bodyweight ranged from 0.85 to 0.96 across cancer types and sexes, indicating strong relationships (P < 0.0001 for all correlations). Thus, relying upon the state total incident burden of cancer as an indicator of weight-related incident cancer burden might seem to be a reasonable approach in the absence of accurate data on the burden of incident cancer associated with excess bodyweight (i.e., case burden/100,000). However, for some states, such reliance would inaccurately represent the weight-related incident cancer burden (Table 5). In particular, for every type of cancer evaluated among women, the state ranks for weight-related cancer burden of Alabama, Michigan, and North Carolina were at least one quintile higher than the ranks for total cancer burden for each respective cancer type, whereas those for Connecticut and Florida were at least one quintile lower than their respective state ranks for total cancer burden. For cancers among men, Texas and Indiana had ranks for weight-related cancer burden that were consistently at least one quintile higher than the ranks for total cancer burden for each type of cancer. Only New Mexico had a rank for weight-related cancer burden that was consistently at least one quintile lower than its rank for total cancer burden for each of the three types of cancer estimated among men.

Table 5.  Underestimation and overestimation of state ranks of weight-related incident cancer burden that results from using as a proxy either the total burden of incident cancer (by type) or the prevalence of overweight and obesity and states for which both total incident cancer burden and prevalence of overweight and obesity would have under and overestimated state rank of weight-related incident cancer burden by type of cancer and by sex
inline image

Using prevalence rates of excess bodyweight as proxy for weight-related cancer burden

Another potential surrogate for the burden of weight-related incident cancer is the prevalence rate of overweight and obesity. Correlations were moderately strong between the state ranks for prevalence of excess bodyweight and the state ranks for incident cancer burden associated with excess bodyweight ranged from 0.38 to 0.79 across types of cancer and sexes, indicating moderately strong relationships (P ≤ 0.007 for all correlations). However, reliance on the state prevalence of overweight and obesity as a proxy for the burden of weight-related incident cancer would underestimate the burden for some states. For all the four types of cancer evaluated among women, two states had weight-related cancer ranks that were at least one quintile higher than the respective state ranks for prevalence of excess bodyweight: Ohio and South Carolina. For all the three types of cancer evaluated among men, three states had weight-related cancer burden ranks that were at least one quintile higher than the corresponding state rank for excess bodyweight: Florida, Louisiana, and Oregon.

Several states had ranks for incident cancer associated with excess bodyweight that were lower than their ranks for prevalence of overweight and obesity, which would have overestimated weight-related incident cancer burden. For women, state weight-related cancer burden ranks for Kentucky and Minnesota were at least one quintile lower than the corresponding ranks for overweight and obesity for all four types of cancer. Among men, five states had ranks for weight-related cancer that were lower by at least one quintile than the ranks for excess bodyweight for all three types of cancers evaluated: Illinois, Kansas, Texas, Utah, and Washington.

States with weight-related incident cancer burdens higher than both total cancer burden and prevalence of excess bodyweight

For some states, the rank for weight-related cancer burden differed by at least one quintile of rank from their corresponding ranks for both their total cancer burden and prevalence of overweight and obesity. The states for which this was observed varied according to the type of cancer. For example, among cancers in women, North Carolina and South Carolina had higher state ranks for weight-related burdens of colon and kidney cancer than they had for either total cancer burden or prevalence of overweight and obesity. North Carolina's rank for weight-related burden of endometrial cancer was also higher than its ranks for total cancer burden and prevalence of overweight and obesity. Among men, Louisiana's state ranks for weight-related cancers of the kidney and prostate would have been underestimated by either total cancer burden or prevalence of overweight and obesity. Other states had weight-related cancer burden that was underestimated by their respective ranks for total cancer burden and prevalence of overweight and obesity: Alabama, Michigan, Virginia (endometrial cancer); Pennsylvania (postmenopausal breast cancer); and Arkansas (colon cancer among women).

Conversely, the ranks for Minnesota's and Maryland's weight-related cancer burden among women would have been overestimated by their ranks for total cancer burden and prevalence of overweight and obesity for cancers of the endometrium and colon. Minnesota's rank for weight-related kidney cancer burden among women would also have been overestimated by its ranks for total cancer burden and prevalence of overweight and obesity. Other states had weight-related cancer burden that was overestimated by their respective ranks for total cancer burden and prevalence of overweight and obesity: Washington (endometrial cancer) and Florida (postmenopausal breast cancer).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgment
  8. Disclosure
  9. REFERENCES
  10. Supporting Information

Understanding the impact of incident cancer associated with overweight and obesity can aid individual states in developing tailored strategies addressing overweight and obesity and weight-related cancer (58). A metric such as a state's case burden of weight-related incident cancer would be useful to measure progress and to compare between states. However, a wide range of estimates for individual state case burdens of incident cancer associated with overweight and obesity resulted from variation in RRs from epidemiological studies, as shown in our sensitivity analysis, such that the estimates were insufficiently accurate to describe the absolute or relative magnitude of weight-related cancer burden. By comparison, state ranks for weight-related incident cancer burden were sufficiently stable to report despite variation in RRs.

Our use of ranks, in the absence of state-level weight-related case burdens per 100,000, revealed weight-related cancer burdens that may not be easily recognized by the states and regions that bear them. Repeated observations from national and regional data indicate that in spite of knowledge that the obese individuals have higher risk of many types of cancer (11), often are diagnosed at later stages (59), and have poorer survival (60), they participate less frequently in prevention activities (61). Alerting states and regions to their exceptional burden will allow them to address such burden by implementing appropriate strategies, such as targeted efforts offering effective, early detection screening technologies to overweight and obese people (61,62) and promotional campaigns to increase awareness about potential bias and stigma in provider-patient interactions that may deter cancer prevention activities (63) and about the need for overweight and obese individuals to practice prevention (62). Conversely, insight might be gleaned from comparing the demographic profiles and health practices of people in these states with those of states for which weight-related cancer burdens are low. Certainly, rankings are an imperfect measure—incapable of providing a sense of the absolute or relative burden—but arguably, ranking of a state in the highest quintile of burden could be considered of sufficient concern to warrant further inquiry and particularly for states that rank consistently with higher burden across multiple types of cancer. While to date no other method has been proposed to provide state-level burden of incident cancer, rankings allowed us to make several observations, including some about patterns of weight-related cancer case burden.

Overall, the lowest burden of weight-related incident cancer was in the western United States, and Colorado was the only state to rank in the lowest quintile of burden for all cancers considered for both men and women. We observed no obvious geographical pattern for the highest burdens for all cancers, although Iowa, South Dakota, and West Virginia consistently ranked within the highest quintile of weight-related cancer burden among the three cancers evaluated for men. West Virginia ranked within the highest quintile of weight-related cancer burden for all four cancers evaluated among women, the only state to do so.

We observed geographical patterns for specific cancers. There was an overlap of states with high burden for weight-related cancers of the endometrium and postmenopausal breast, and to an extent, cancers of the colon in the Midwest, Mid-Atlantic, and Appalachian regions. In these areas, we speculate that people may share weight-related cancer risk factors, such that interventions targeting such key risk factors may have broad impact preventing multiple types of cancer. Other factors that relate to both excess bodyweight and cancer burden in a community such as a region's socioeconomic profiles, cultural practices, access to health services, and other community resources may also inform public health leaders seeking targets for intervention and wide public health impact.

Of particular importance is our use of ranks to identify states with greater weight-related burden than is suggested by their burdens of either total cancer or prevalence of overweight and obesity. Even though state ranks for weight-related cancer burden were correlated with ranks for both total cancer burden and prevalence of excess bodyweight, state ranks for the latter two were imperfect proxies for weight-related cancer burden, underestimating it considerably in some cases. Reliance on either total cancer burden or prevalence of overweight and obesity as surrogates for the burden of weight-related incident cancer would result in undetected, and therefore unaddressed, burden for many states for several types of cancer.

Our findings extended those of other analyses focused on weight-related cancer burden in the United States An analysis by Polednak (10) estimated that 3.2% of all new cases for six types of cancer (i.e., endometrial, kidney, gallbladder, esophagus, gastric cardia, postmenopausal breast), or 41,383 new diagnoses in the United States in 2002, might have been avoided by preventing obesity (BMI ≥30 kg/m2). Another analysis by Calle et al. (11) estimated that 90,000 deaths due to cancer, equaling 14% of all cancer deaths in men and 20% of those in women, could be prevented each year if US adults could maintain normal weight (BMI <25 kg/m2). Together, the studies by Polednak (10) and Calle et al. (11) provide a sense of scale for the burden of weight-related cancer incidence and mortality at the national level. To this, our analysis adds identification of regions and states in the United States that bore the greatest burdens of weight-related incident cancer diagnosed in 2003.

Our study had several limitations. First, our estimates of weight-related cancer burden for the United States, like those of Polednak (10) and Calle et al. (11) for the United States, and those of Bergström et al. (13) for Europe, resulted from ecological analyses, in that height and weight data were not obtained directly from the individuals upon whom the cancer incidence (or mortality) was estimated. Such ecological analyses focus more upon the aggregated experience of groups than those of individuals, and to inform about population disease burden, such reports must rely upon other published studies for evidence associating risk factors with outcomes. Another limitation related to our use of cross-sectional data, specifically that the ascertainment of “exposure” (i.e., overweight and obesity) and “outcome” (i.e., cancer diagnosis) were concurrent in time, thus, causality cannot be determined. One could speculate that we have underestimated the actual weight-related incident cancer burden by applying today's rates of overweight and obesity in our calculations (rather than rates from an earlier time), as rates of excess bodyweight in the United States have risen steadily over the past 20 years and have never been higher than they are now (1). Moreover, assuming that other factors influencing cancer rates remain the same, the impact of today's rates of excess bodyweight is likely to result in higher incidence of weight-related cancer in the future than what we have reported here, which is based on current estimates of cancer incidence. Although it is possible to estimate the impact of today's rates of excess bodyweight on future cancer incidence, or alternatively the influence of historic rates of overweight and obesity on today's cancer incidence, performing such estimates were beyond the scope of this analysis.

Other limitations that are highly relevant to the public health impact of the burden of weight-related cancer relate to the interpretation of the PARP for excess bodyweight defined by BMI. BMI describes only the present state of an individual's body size that has developed over time and as a result of complex interactions between multiple causal factors. Misattribution of BMI as a risk factor could result if BMI data used in our approach were not the BMIs during critical phases of disease development or if true risk factors are absent when BMI is measured. Other problems of misclassification result from characterizing excess bodyweight using an algorithm that combines weight and height, as these measures alone may not be relevant aspects of body composition (e.g., body mass distribution, proportion of lean and fat mass (64)).

Other limitations relate to possible error inherited from data resources used in our approach. For example, self-reported height and weight data from the BRFSS, which is conducted by telephone, are vulnerable to reporting bias and may result in an underestimation of BMI. Also of concern is the low response rate to the BRFSS (http:www.cdc.govbrfss). Investigators relying upon ecological approaches will have to exercise caution, as we have tried to do, when interpreting their findings.

Another limitation related to PARP, which can be estimated for groups of exposures (e.g., overweight and obesity) (14), is the interpretation of BMI in the context of other factors that influence body size. These factors include behaviors (e.g., diet, dietary intake, lifestyle, and physical activity), as well as genetic factors, including polymorphisms that influence metabolic efficiency and the hormone profiles that reflect the genetic-environmental interface. Because these factors can each be used to characterize excess bodyweight, but overlap in such characterization, combining them in PARP formulas may not be appropriate and the interpretation of such composite PARP values may be complex. Although not the focus of this analysis, efforts made toward understanding this issue better may facilitate development of more effective state-level strategies against weight-related cancer while developing better methods to estimate disease burden in populations.

The rapidly rising rates of overweight and obesity in the United States and their impact on cancer are major public health issues. Policies and strategies against excess bodyweight and weight gain may be critical tools to prevent cancer, and the development of such strategies at a local level may be helped by knowledge of the state-level distribution of weight-related cancer burdens. Although our calculations for the weight-related cancer case burdens per state produced estimates that we felt were too inexact to use directly, we found their rankings to be useful for some description of weight-related cancer burden in the United States in the absence of state-level cancer incidence data linked directly to body size measures. For example, grossly captured patterns of greater burden borne by states such as West Virginia should not be ignored. We were able to observe geographical clustering of states with low weight-related and high weight-related cancer burdens, which may also be useful for conducting research on the causes of weight-related cancer, and also to reveal that total cancer burden and prevalence of overweight and obesity are imperfect proxies for weight-related cancer burden. The latter finding, perhaps more importantly, cautions against simply developing interventions for weight-related cancers that target states with high burden of excess bodyweight or high total burden of types of cancers for which excess bodyweight is a risk factor, as such interventions may not reach the states most affected by weight-related incident cancer.

The limitations to this analysis underscore the need for better data resources (e.g., cancer diagnosis linked directly to measured height and weight) and more consistency in the methods used in studies estimating risk of cancer associated with excess bodyweight (e.g., report of age-specific and sex-specific RR adjusted for critical confounders, such as caloric intake and physical activity). Despite these limitations, great incentive exists for states to recognize higher burden of weight-related cancer and to prevent and detect it early among the overweight and obese individuals, as they have greater risk, present at later stages, and have shorter survival from cancer. In summary, while we continue to address and investigate issues in the accurate estimation of the burden of weight-related cancer, efforts to remedy the problem of overweight and obesity in the United States, and its impact on cancer, should not be delayed.

Acknowledgment

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgment
  8. Disclosure
  9. REFERENCES
  10. Supporting Information

We thank Ms Youngping Hao at the American Cancer Society for her help creating the map figures and at The University of Texas M. D. Anderson Cancer Center, Ms Christine Wogan of Scientific Publications for her help editing the manuscript, Ms Jessica Lingerfelt of Epidemiology for her help with the references, and Ms Cheri McClellan of Epidemiology for her assistance with the tables. We also thank Beverly Rockhill Levine for providing helpful feedback about our analytical process. Also, during the initial phase of the project, Facundo Arganaraz from the National University of Tucumán (Tucumán, Argentina) was supported through the scholarship, “The Capital Experience/To Learn Doing (2001–2002),” sponsored by EDESA S.A. (Salta, Argentina) in conjunction with The Institute for Experiential Learning (Washington, DC) and Bernard F. Fuemmler was a fellow in the National Cancer Institute Cancer Prevention Fellowship Program before joining the faculty at Duke University.

REFERENCES

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgment
  8. Disclosure
  9. REFERENCES
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods and Procedures
  5. Results
  6. Discussion
  7. Acknowledgment
  8. Disclosure
  9. REFERENCES
  10. Supporting Information

Supporting Information

FilenameFormatSizeDescription
oby_1636_sm_oby2008228a_errcrr.fhtml0KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.