Pancreatic Cancer Death Rates by Race Among US Men and Women,

Transcription

1 DOI: /jnci/djt292 The Author Published by Oxford University Press. All rights reserved. For Permissions, please Article Pancreatic Cancer Death Rates by Race Among US Men and Women, Jiemin Ma, Rebecca Siegel, Ahmedin Jemal Manuscript received April 1, 2013; revised August 8, 2013; accepted August 12, Correspondence to: Jiemin Ma, PhD, MHS, Surveillance and Health Services Research Program, American Cancer Society, 250 Williams St, Atlanta, GA ( Background Methods Results Conclusions Few studies have examined trends in pancreatic cancer death rates in the United States, and there have been no studies examining recent trends using age-period-cohort analysis. Annual percentage change in pancreatic cancer death rates was calculated for 1970 to 2009 by sex and race among adults aged 35 to 84 years using US mortality data provided by the National Center for Health Statistics and Joinpoint Regression. Age-period-cohort modeling was performed to evaluate the changes in cohort and period effects. All statistical tests were two-sided. In white men, pancreatic cancer death rates decreased by 0.7% per year from 1970 to 1995 and then increased by 0.4% per year through Among white women, rates increased slightly from 1970 to 1984, stabilized until the late 1990s, then increased by 0.5% per year through In contrast, the rates among blacks increased between 1970 and the late 1980s (women) or early 1990s (men) and then decreased thereafter. Age-period-cohort analysis showed that pancreatic cancer death risk was highest for the 1900 to 1910 birth cohort in men and the 1920 to 1930 birth cohort in women and there was a statistically significant increase in period effects since the late 1990s in both white men and white women (two-sided Wald test, P <.001). In the United States, whites and blacks experienced opposite trends in pancreatic cancer death rates between 1970 and 2009 that are largely unexplainable by known risk factors. This study underscores the needs for urgent action to curb the increasing trends of pancreatic cancer in whites and for better understanding of the etiology of this disease. J Natl Cancer Inst;2013;105: Cancer of the pancreas remains one of the deadliest cancer types, and it is the fourth leading cause of cancer death in each sex and in men and women combined in the United States (1). In 2013, it is estimated that approximately Americans will die from this disease (1). To date, the etiology of pancreatic cancer is not well understood. Known modifiable risk factors include cigarette smoking (2) and obesity (3,4). There is some evidence that intake of red or processed meat and high-temperature cooking may increase the risk of pancreatic cancer (5,6). Other suspected lifestyle risk factors include low vegetable and fruit intake, physical inactivity, and alcohol drinking (7,8). An examination of long-term disease trends by different demographic characteristics may provide crucial clues to a better understanding of the disease. However, few studies have examined the trends in pancreatic cancer incidence or mortality rates in the United States (9 12). The most recently published study found that the trends in pancreatic cancer death rates during 1969 to 2002 varied by sex and race (10). Nevertheless, no studies have examined recent trends using age-period-cohort analysis, which has been reported to be more useful than conventional cross-sectional analysis in evaluating trends (13). Therefore, in this study, we describe the long-term trends in pancreatic cancer death rates in the United States between 1970 and 2009 and examine cohort and period effects by sex and race using age-period-cohort modeling. Period effects are usually generated by factors that affect all ages concurrently, such as improvements in diagnostic practice and treatment. In contrast, cohort effects are attributed to risk factors that vary in prevalence by generation, such as smoking. Methods Study Population Mortality data for pancreatic cancer between 1970 and 2009 were obtained from the SEER*Stat database of the National Cancer Institute s (NCI) Surveillance, Epidemiology and End Results (SEER) Program, as reported by the National Center for Health Statistics (14). Pancreatic cancer deaths were selected as the underlying cause of death on the death certificates according to the coding and selection rules of the eighth revision of the International Classification of Disease (ICD-8) for deaths occurring between 1694 Articles JNCI

2 1970 and 1978, according to the ninth revision of the ICD (ICD-9) for deaths occurring between 1979 and 1998, and according to the ten revision of the ICD (ICD-10) for deaths occurring between 1999 and To facilitate age-period-cohort analysis, we restricted analyses to deaths occurring in those aged 35 to 84 years, which accounted for approximately 90% of all pancreatic cancer deaths. Statistical Analysis Pancreatic cancer death rates were adjusted to the 2000 US standard population. Joinpoint regression (15) was applied to evaluate temporal trends in pancreatic cancer death rates by sex and race using NCI s Joinpoint Regression Program 3.5.4, which subsequently compares models by starting with no joinpoint and testing whether one or more joinpoints need to be entered into the model to best fit the data. In this study, a maximum of two joinpoints were allowed, which means that the final model could have one to three trend segments depending on how many joinpoints fit the data best. Annual percentage change (APC) was calculated for each trend segment detected. In age-period-cohort analysis, the data were classified into ten 5-year age groups (aged 35 39, 40 44,, 80 84) and eight 5-year calendar periods (periods , ,, ), spanning seventeen 10-year partially overlapping birth cohorts indicated by midyear of birth (years 1890, 1895,, 1970). Ageperiod-cohort modeling was performed within a Poisson regression framework. The expected log rate for each age-period-cohort group (ρ apc ) could be expressed as ρ apc = α a + π p + γ c, where α a, π p, and γ c denote age, period, and cohort effects, respectively, and these parameters can be further partitioned into linear (α L, π L, and γ L ) and nonlinear components ( α a, π p, and γ c ) (13,16). With the fundamental constraint that c = p a and an additional constraint that π L = 0, the above expression could be transformed into the following form: ρ ac = µ + ( αl + π L) a+ ( π L + γ L) c+ αa+ π p+ γ c (equation A) In equation A, µ is grand mean, ( αl + π L) is longitudinal age trend, and ( π L + γ L) is net drift. Although neither period ( π L ) nor cohort effect ( γ L ) is identifiable, period and cohort deviations ( π and γ p c ), longitudinal age trend ( αl + π L), net drift ( π L + γ L), and local changes in log-linear trends of period and cohort effects can be estimated unambiguously (13,16). In this study, we tested the slope changes of period and cohort deviations, which is equivalent to the Tarone Chu method of testing effect change (17). A positive slope change means an accelerated increasing effect, a reversed decreasing trend, or a decelerated decreasing trend; and a negative slope change means a decelerated increasing effect, a reversed increasing trend, or an accelerated decreasing trend. To account for post hoc selection of comparison groups (either including or not including a common time point), statistical significance level for slope changes of period and cohort deviations was set as P less than.025 (two-sided). Statistical significance level for other statistics in this study was set as P less than.05 (two-sided). In addition, we graphically presented birth cohort rate ratios, which compared the death rate of each birth cohort with that of the 1910 birth cohort. We chose the 1910 birth cohort as the reference in men because those born approximately 1910 had the highest rates, and we used this reference cohort for women for consistency. Each birth cohort rate ratio includes two components (ie, cohort deviation and net drift) (18) and its calculation was described elsewhere (19). Because net drift is the sum of log-linear cohort and period effect, birth cohort rate ratios would be cohort relative risks adjusted for age and nonlinear components (deviations) of period effects. Results Table 1 and Figure 1 show the trends in pancreatic cancer death rates by sex and race among those aged 35 to 84 years. In white men, pancreatic cancer death rates decreased by 0.7% per year from 24.8 per in 1970 to 20.4 per in 1995 and then increased by 0.4% per year through 2009 to 21.5 per ; in white women, the rates increased by 0.4% per year during the period from 1970 to 1984 from 14.6 to 15.3 per , remained stable between 1984 and 1998, and then increased again by 0.5% per year through 2009 to 15.9 per In contrast, the rates for black men increased by 0.5% per year between 1970 and 1989 from 29.0 to 31.3 per and then decreased by 0.9% per year through 2009 to 27.5 per ; in black women, the rates increased by 1.3% per year between 1970 and 1989 from 18.3 to 23.1 per and then decreased by 0.5% through 2009 to 20.9 per However, rates were consistently higher in blacks than in whites. Figure 2 presents cohort rate ratios of pancreatic cancer death rate by sex and race, which represent smoothed and summarized pattern of observed age-specific rates by birth cohort (Supplementary Figure 1, available online). In white men, cohort relative risks peaked in 1895, plateaued until 1910, and then dropped continuously to 0.6 in Among black men, the Table 1. Joinpoint analysis of pancreatic cancer death rates by race and sex, ages years, , United States* Male Female Trend 1 Trend 2 Trend 1 Trend 2 Trend 3 Years APC Years APC Years APC Years APC Years APC All Races White Black * APC = annual percentage change. The 1995 rate was excluded from analysis as an outlier. Statistically significant, P less than.05 (two-sided permutation test). jnci.oxfordjournals.org JNCI Articles 1695

3 Figure 1. Trends in pancreatic cancer death rates by race and sex, ages 35 to 84 years, 1970 to 2009, United States. Rates were age-adjusted to the 2000 US standard population. Scattered points were observed rates. Lines were fitted rates according to joinpoint regression, which allowed up to two join-points. Solid triangle = black male; open triangle = white male; solid circle = black female; open circle = white female. cohort relative risk increased from 0.8 in 1890 to 1.0 in 1910 and then decreased thereafter. For white and black women, cohort relative risks peaked around 1920 to After the peak, the relative risks slowly decreased until the trend was interrupted by elevated risks for those born around 1950 to Testing of the slope change in cohort deviations showed that the slope increases around 1950 to 1960 are statistically significant among both white women (slope difference = 0.009; 95% confidence interval [CI] = to 0.016; P =.023) and black women (slope difference = 0.026; 95% CI = to 0.047; P=.017) (Supplementary Figure 2, available online). In contrast, the slope decreased around 1950 to 1960 among both white men (slope difference = 0.008; 95% CI = to 0.002; P =.007) and black men (slope difference = 0.024; 95% CI = to ; P <.001), compared with 1935 to 1945 cohorts (Supplementary Figure 3, available online). Figure 3 depicts period deviations by sex and race. In white men, the slope of period deviations changed around both the late 1980s and the late 1990s. Testing of the slope change indicated that the increase since the late 1990s was statistically significant (slope difference = 0.005; 95% CI = to 0.007; P <.001). Similarly, the slope of period deviations for white women also has statistically significantly (slope difference = 0.006; 95% CI = to 0.008; P <.001) increased since the late1990s. Because of the relatively small number of deaths, there were large variations in period deviations for black men and women. However, increased period deviations around the late 1990s were also seen among these groups, although the changes were not statistically significant and were smaller than in whites. Discussion Our study found opposite trends in pancreatic cancer death rates between whites and blacks for those aged 35 to 84 years during the period from 1970 to 2009 in the United States. In whites, pancreatic cancer death rates have increased since the late 1990s, after a 25-year decrease in men and a 15-year stable trend in women. In contrast, pancreatic cancer death rates among blacks increased during the 1970s and 1980s and then decreased thereafter. Two major characteristics of these race-specific patterns are delayed decreases in death rates in blacks and recent increases in death rates in whites. Decreased smoking prevalence has been widely recognized as the main contributor to the decreases in pancreatic cancer death rates (9,12,20). However, reasons for the delayed decreases in pancreatic cancer death rates among blacks are unclear. Smoking prevalence has decreased in both blacks and whites since 1965, although black men consistently had a higher rate than white men over the years (21,22). In addition, there are no delays for blacks in experiencing decreased lung cancer death rates (23). This indicates that smoking may act differently in causing these two cancers. Alternately, there may be some factors that have differentially modified the impacts of smoking on pancreatic cancer trends. Further studies on the mechanisms by which smoking causes pancreatic cancer are warranted. A recent pooled analysis suggested that smoking is more likely to be a late-stage carcinogen to pancreatic cancer because pancreatic cancer risk reduced very quickly after cessation (2). Our study supports this finding by observing a 20-year-earlier decrease in pancreatic cancer death rates than in lung cancer death rates among white men (23). Our age-period-cohort analysis also found that the peak year of pancreatic cancer death risk by birth cohort was 10 to 15 years earlier than that of smoking prevalence and that of lung cancer death rate (24 26). Another notable finding from the age-period-cohort analysis is that women born circa 1950 to 1960 seemed to have an elevated death risk of pancreatic cancer. As these women age, we would expect to see further increased rates in white women and a possible reverse of current decreasing trends in black women, if they continue to smoke. Similar findings were also seen for lung cancer death rates (19). Reasons for the recent increases in pancreatic cancer death rates among whites are also unclear. Increased period effects since the late 1990s indicate that factors affecting all ages may have contributed to these increases. One of the possible contributors is obesity, which has been linked with a 20% increased death risk from pancreatic cancer. Since the 1980s, obesity prevalence has increased by a factor of two- to threefold among all age and ethnic groups in the United States (27,28). However, given the possible long latency period between obesity and pancreatic cancer death and the absence of increase in mortality rates in blacks, the contribution of fairly recent increases in obesity prevalence is expected to be small. Nevertheless, the obesity epidemic will likely lead to future increases in pancreatic cancer death rates because younger birth cohorts have been reported to have a higher propensity of being obese (29). Another possible contributor to the increased pancreatic cancer death rates among whites is improved diagnostic techniques. Historically, pancreatic cancer ascertainment has been a challenge because of the deep anatomical position of the pancreas (30). Since the 1980s, improvements in imaging-guided fine-needle aspiration have led to a rapid increase in microscopically and cytologically confirmed pancreatic cancer cases among both blacks and whites in the SEER 9 registries (data not shown), which may in turn have led to higher pancreatic cancer incidence rates. Similarly, improvements 1696 Articles JNCI

4 Figure 2. Birth cohort rate ratios of pancreatic cancer death rate by race and sex, United States. The reference group is 1910 birth cohort; shaded areas denote the 95% confidence intervals of rate ratios. The horizontal line is a reference line for rate ratios equal to one, and the vertical line represents 1910 birth cohort. jnci.oxfordjournals.org JNCI Articles 1697

5 Figure 3. Period deviations of pancreatic cancer death rates by race and sex, 1970 to 2009, United States.The dotted line is a reference line of zero period deviation. Shaded area represents pointwise 95% confidence interval of period deviation. P value denotes the statistical significance for the change in the slopes of the period deviations before and after the 1995 to 2000 time period. A P value of was considered statistically significant. Statistical significance of slope difference was tested using two-sided Wald test. The 95% confidence intervals of slope differences for white men, white women, black men, and black women were to 0.007, to 0.008, to 0.008, and to 0.010, respectively. in case ascertainment have been recognized as a contributor to increased pancreatic cancer rates in several European countries (31 34). Improvements in diagnostic techniques may have attenuated the unexplained declines in pancreatic cancer death rates in blacks. Because of lack of reliable long-term data, it is difficult to evaluate the potential influence of some suspected risk factors (eg, meat and vegetable intake) on pancreatic cancer trends. However, it is highly likely that these suspected and some unknown risk factors may have modified the effects of known risk factors on pancreatic cancer trends. Because there have been few improvements in pancreatic cancer survival over the last three decades (23), treatment would have had limited impact on pancreatic cancer death trends. Strengths of this study include its long study period and the use of age-period-cohort analysis, which allowed us to examine pancreatic cancer trends in a more systematic way than have previous studies. Our study was not without limitations. First, using mortality as the outcome, we were unable to evaluate the trends by pancreatic cancer type. The two major types of pancreatic 1698 Articles JNCI

6 cancer endocrine and exocrine pancreatic cancers may have different trends because their risk factors, diagnosis, treatment, and prognosis are different. The dominance (about 95%) of exocrine pancreatic cancer suggests our findings are applicable for exocrine cancers but are not necessarily applicable for endocrine cancers. Second, during the study period, the coding of disease changed from ICD-8 to ICD-9 in 1979 and to ICD-10 in The comparability ratio between ICD-9 and ICD-10 for pancreatic cancer is (35), and the comparability ratio for malignant neoplasms of digestive organs and peritoneum between ICD-8 and ICD-9 is 1.03 (pancreatic cancer data unavailable) (36). This suggests that the changes in pancreatic cancer coding would have had a minimal impact on pancreatic cancer trends. In summary, we found race-distinct trends in pancreatic cancer death rates during the period from 1970 to 2009 in the United States that are largely unexplainable by known risk factors. The decreasing trend in blacks over the past 10 to 15 years is particularly interesting because the prevalence rates of factors (ie, obesity, diabetes, and improved diagnosis) that are likely contributing to the recent increases in pancreatic cancer death rates in whites have also increased in blacks. With a full manifestation of the consequences of the fairly new obesity epidemic, pancreatic cancer could become a major public health problem in the future in the United States if no urgent interventions are to take effect. High death rates were also found in women born circa 1950 to 1960, which may reflect their increased smoking initiation rates. Based on current knowledge, smoking cessation and body weight control would be the two key intervention measures. A better understanding of the etiology of pancreatic cancer would help formulate more effective measures to curb the foreseeable increases in pancreatic cancer burden. References 1. Siegel R, Naishadham D, Jemal A. Cancer statistics, CA Cancer J Clin. 2013;63(1): Lynch SM, Vrieling A, Lubin JH, et al. Cigarette smoking and pancreatic cancer: a pooled analysis from the pancreatic cancer cohort consortium. Am J Epidemiol. 2009;170(4): Arslan AA, Helzlsouer KJ, Kooperberg C, et al. Anthropometric measures, body mass index, and pancreatic cancer: a pooled analysis from the Pancreatic Cancer Cohort Consortium (PanScan). Arch Intern Med. 2010;170(9): Berrington de Gonzalez A, Sweetland S, Spencer E. A meta-analysis of obesity and the risk of pancreatic cancer. Br J Cancer. 2003;89(3): Larsson SC, Wolk A. Red and processed meat consumption and risk of pancreatic cancer: meta-analysis of prospective studies. Br J Cancer. 2012;106(3): Anderson KE, Mongin SJ, Sinha R, et al. Pancreatic cancer risk: associations with meat-derived carcinogen intake in the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (PLCO) cohort. Mol Carcinog. 2012;51(1): Hidalgo M. Pancreatic cancer. N Engl J Med. 2010;362(17): Vincent A, Herman J, Schulick R, Hruban RH, Goggins M. Pancreatic cancer. Lancet. 2011;378(9791): Weiss W, Benarde MA. The temporal relation between cigarette smoking and pancreatic cancer. Am J Public Health. 1983;73(12): Zhang J, Dhakal I, Ning B, Kesteloot H. Patterns and trends of pancreatic cancer mortality rates in Arkansas, : a comparison with the US population. Eur J Cancer Prev. 2008;17(1): Zhang J, Dhakal I, Yan H, Phillips M, Kesteloot H. Trends in pancreatic cancer incidence in nine SEER Cancer Registries, Ann Oncol. 2007;18(7): Zheng T, Holford TR, Ward BA, McKay L, Flannery J, Boyle P. Time trend in pancreatic cancer incidence in Connecticut, Int J Cancer. 1995;61(5): Rosenberg PS, Anderson WF. Age-period-cohort models in cancer surveillance research: ready for prime time? Cancer Epidemiol Biomarkers Prev. 2011;20(7): Surveillance, Epidemiology, and End Results (SEER) Program. SEER*Stat Database: Mortality All COD, Aggregated With State, Total U.S. ( ) <Katrina/Rita Population Adjustment>, National Cancer Institute, DCCPS, Surveillance Research Program, Surveillance Systems Branch, Released April Accessed September 25, Kim HJ, Fay MP, Feuer EJ, Midthune DN. Permutation tests for joinpoint regression with applications to cancer rates. Stat Med. 2000;19(3): ; erratum: 2001;20(4): Rosenberg PS, Anderson WF. Proportional hazards models and age-period-cohort analysis of cancer rates. Stat Med. 2010;29(11): Tarone RE, Chu KC. Evaluation of birth cohort patterns in population disease rates. Am J Epidemiol. 1996;143(1): Rosenberg PS, Wilson KL, Anderson WF. Are incidence rates of adult leukemia in the United States significantly associated with birth cohort? Cancer Epidemiol Biomarkers Prev. 2012;21(12): Jemal A, Ma J, Rosenberg PS, Siegel R, Anderson WF. Increasing lung cancer death rates among young women in southern and midwestern States. J Clin Oncol. 2012;30(22): Wingo PA, Cardinez CJ, Landis SH, et al. Long-term trends in cancer mortality in the United States, Cancer. 2003;97(12 Suppl): Giovino GA, Schooley MW, Zhu BP, et al. Surveillance for selected tobacco-use behaviors United States, MMWR. 1994; 43(3): Garrett BE, Dube SR, Trosclair A, Caraballo RS, Pechacek TF. Cigarette smoking United States, MMWR Surveill Summ. 2011;60(Suppl): Howlader N, Noone AM, Krapcho M, et al. SEER Cancer Statistics Review, (Vintage 2009 Populations). National Cancer Institute. Bethesda, MD: National Cancer Institute; Garfinkel L. Trends in cigarette smoking in the United States. Prev Med. 1997;26(4): Harris JE. Cigarette smoking among successive birth cohorts of men and women in the United States during J Natl Cancer Inst. 1983;71(3): Jemal A, Chu KC, Tarone RE. Recent trends in lung cancer mortality in the United States. J Natl Cancer Inst. 2001;93(4): Flegal KM, Carroll MD, Kuczmarski RJ, Johnson CL. Overweight and obesity in the United States: prevalence and trends, Int J Obes Relat Metab Disord. 1998;22(1): Flegal KM, Carroll MD, Kit BK, Ogden CL. Prevalence of obesity and trends in the distribution of body mass index among US adults, JAMA. 2012;307(5): Robinson WR, Keyes KM, Utz RL, Martin CL, Yang Y. Birth cohort effects among US-born adults born in the 1980s: foreshadowing future trends in US obesity prevalence. Int J Obes (Lond). 2013;37(3): Lyon JL, Robison LM, Moser R, Jr. Uncertainty in the diagnosis of histologically confirmed pancreatic cancer cases. Int J Epidemiol. 1989;18(2): Bosetti C, Bertuccio P, Negri E, La Vecchia C, Zeegers MP, Boffetta P. Pancreatic cancer: overview of descriptive epidemiology. Mol Carcinog. 2012;51(1): Levi F, Lucchini F, Negri E, La Vecchia C. Pancreatic cancer mortality in Europe: the leveling of an epidemic. Pancreas. 2003;27(2): jnci.oxfordjournals.org JNCI Articles 1699

7 33. Arfe A, Malvezzi M, Bertuccio P, Decarli A, La Vecchia C, Negri E. Cancer mortality trend analysis in Italy, Eur J Cancer Prev. 2011;20(5): Bouvier AM, David M, Jooste V, Chauvenet M, Lepage C, Faivre J. Rising incidence of pancreatic cancer in France. Pancreas. 2010;39(8): Anderson RN, Minino AM, Hoyert DL, Rosenberg HM. Comparability of cause of death between ICD-9 and ICD-10: preliminary estimates. Natl Vital Stat Rep. 2001;49(2): Klebba AJ. Estimates of Selected Comparability Ratios Based on Dural Coding of 1976 Death Certificates by the Eigth and Ninth Revisions of the International Classification Of Disease. Monthly Vital Statistics Report. Hyattsville, MD: National Center for Health Statistics; Funding This work was supported by the intramural research department of the American Cancer Society. Notes The funding sources had no role in the study design, data collection, data analysis, data interpretation, writing of the manuscript, or decision to submit the manuscript for publication. The authors have no conflicts of interest to report. Affiliations of authors: Surveillance and Health Services Research program, American Cancer Society (JM, RS, AJ) Articles JNCI