Population-based cancer registries that are NAACCR members and participate in the NCI’s Surveillance, Epidemiology, and End Results (SEER) Program, and/or the CDC’s National Program of Cancer Registries were used to obtain information on newly diagnosed invasive cancers and benign and borderline brain tumors. Incident cases were classified by site and histology according to the International Classification of Diseases for Oncology (ICD-O) edition in use at the time of diagnosis, converted to the Third Edition coding (¹³) and categorized according to SEER site groups (¹⁴).

Incidence data were not available uniformly for every period, geographic area, and racial and ethnic group in the United States (Supplementary Table 1, available online). The longest continuous incidence data were available from the nine original SEER registries (SEER 9) covering about 10% of the US population. Long-term (1975–2007) trends based on data from the SEER 9 registries are included in Supplementary Table 2 (available online). Data providing better coverage of the US population (about 14%) were available from the SEER 13 registries and form the basis of our long-term incidence trend (1992–2007) analysis for all races and ethnicities combined (¹⁵). Beginning in 1995, following the advent of the National Program of Cancer Registries, coverage of the US population increased dramatically. Data from NAACCR covering 40 population-based cancer registries were used to assess short-term (1998–2007) trends. Data from 46 NAACCR population-based cancer registries were used to estimate 5-year (2003–2007) average annual age-standardized incidence rates for all races and ethnicities combined and for each of the five major racial and ethnic populations (white, black, Asian and Pacific Islander [API], American Indian/Alaska Native [AI/AN] who reside in counties covered by the Indian Health Service [IHS] Contract Health Service Delivery Area [CHSDA], and Hispanic). The 40 and 46 registries met NAACCR’s data quality criteria for every year included in the analysis; these registries covered 83.6% and 93% of the US population, respectively.

All primary brain and ONS tumors (ICD-O-3 codes C70.0–72.9 and C75.1–75.3, respectively), including malignant, borderline, and benign behaviors diagnosed in 2004–2007 were identified from 46 states in the NAACCR dataset. A neuropathologist reviewed the brain and ONS site and histology combinations and recommended excluding 1771 cases (0.8%) from analysis because of unlikely combinations. Consistent with the SEER site re-code convention, tumors coded to the nasal and nasopharyngeal regions also were excluded.

Data on approximately 76 000 malignant and 137 000 non-malignant brain and ONS tumors were analyzed. Within the brain and ONS, seven major histological groups were used in analyses (¹⁶,¹⁷). Tumors of neuroepithelial tissue were divided into eight specific histological subgroups (¹⁶). Tumors of neuroepithelial tissue coded as nonmalignant by registries, but for which only a malignant behavior code existed in ICD-O-3, were considered malignant. Consistent with previous practice, pilocytic astrocytomas were considered malignant. Malignant and papillary meningioma and meningeal sarcomatosis were categorized as malignant, whereas all other benign and uncertain or atypical meningioma histologies were categorized as nonmalignant, according to the ICD-O-3 behavior codes. Childhood brain and ONS tumors also were grouped using International Classification of Childhood Cancers (ICCC) definitions (¹⁸).

Cause of death is based on death certificate information reported to state vital statistics offices and compiled into a national file through the CDC National Center for Health Statistics National Vital Statistics System (¹⁹) and categorized according to SEER anatomic site groups (¹⁴) to maximize comparability among ICD and ICD-O versions. The underlying causes of death were selected according to the version of the ICD codes and selection rules in use at the time of death (ICD-6 to ICD-10) (²⁰–²⁴). We examined long-term (1975–2007) mortality trends for all races and ethnicities combined. Short-term (1998–2007) trends and 5-year (2003–2007) average annual age-standardized death rates were calculated for all cancer sites combined and for the top 15 cancer sites for men and women in each of the five major racial and ethnic populations. Death rates for the AI/AN population were based on deaths in counties served by IHS CHSDA because estimated rates based on CHSDA counties have been reported to be more accurate for this group (¹⁰,²⁵).

County-level population estimates, summed to the state and national level, were used as denominators in calculations of incidence rates (²⁶). The National Center for Health Statistics and the Census Bureau collaborate to provide NCI with bridged single-race annual population estimates, with annual reestimates calculated back to the most recent decennial census to accommodate multiracial data (²⁷). The NCI makes slight modifications to the Hawaii population estimates based on additional local information (²⁶).

For most states, population estimates as of July 1 of each year were used to calculate annual incidence and death rates. For Louisiana, Alabama, Mississippi, and Texas, where residents were displaced by Hurricanes Katrina and Rita, NCI made adjustments to the 2005 incidence data and underlying population data. The national total population estimates are not affected by these adjustments (further details are available at http://seer.cancer.gov/popdata/methods.html).

Age-specific and age-standardized rates were expressed per 100 000 persons (or per 1 000 000 children), based on 2000 US standard population, and generated using SEER*Stat Software, Version 6.6.2 (http://www.seer.cancer.gov/seerstat) (²⁸). Rates were suppressed if the numerator was less than 16 observations, consistent with our previous work (¹–¹²).

Trends in age-standardized cancer incidence and US death rates were analyzed using joinpoint regression, which involves fitting a series of joined straight lines on a logarithmic scale to the trends in the annual age-standardized rates (http://www.srab.cancer.gov/joinpoint). We allowed a maximum of three joinpoints in models for the period 1992–2007 (Table 1), four joinpoints in models for the period 1975–2007 (Table 2 and Supplementary Table 2, available online), and up to two joinpoints for the period 1998–2007 for short-term fixed interval incidence (Table 3) and mortality analyses (Table 4). The joinpoint method is described in detail elsewhere (²⁹). We present the long-term (1975–2007 and 1992–2007) trends in incidence using annual percent changes (APCs; ie, the slope of the line segment) based on observed data and APCs adjusted for reporting delays (which affect mostly recent years). Delay-adjustment is a statistical method to correct for unreported (delayed) or updated cancer cases. Delay-adjusted APCs, used in our description of results, are available only for long-term incidence data (Table 1 and Supplementary Table 2, available online) (³⁰). The average APC (AAPC), a summary measure to compare fixed interval trends by race and ethnicity, is estimated as a geometric weighted average of the joinpoint APC trend analysis, with the weights equal to the lengths of each segment during the prespecified fixed interval (http://srab.cancer.gov/joinpoint/aapc.html) (³¹). The APC was suppressed if the numerator was less than 10 cancers for any year, consistent with our previous methods (¹–¹²).

In describing long- and short-term trends with estimates of APC and AAPC, the terms “increase” or “decrease” signify that the slope (APC or AAPC) of the trend was statistically significant (P < .05) using a t test (APC) or Z test (AAPC). For non-statistically significant trends, we used terms such as “level,” “stable,” “non-statistically significant increase,” or “non-statistically significant decrease.” All statistical tests were two-sided.

Trend analysis showed that overall cancer incidence rates for all racial and ethnic groups combined decreased by 0.8% per year during the most recent period, 2003–2007 (Table 1); a statistically significant decrease of 0.6% per year was noted in women, whereas a non-statistically significant decrease of 0.8% per year was noted in men that was influenced by a recent (2005–2007) non-statistically significant increase in prostate cancer incidence. When prostate cancer was excluded from the trend analysis, there was a statistically significant decrease in cancer incidence for all sites combined (data not shown). Incidence for prostate and breast cancers, two of the most frequently diagnosed cancers, showed possible changing trends. Cancer of the prostate showed a non-statistically significant annual increase of 3.0% in 2005–2007, after a statistically significant decrease in 2001–2005. The trend analysis of breast cancer in women showed a decrease from 1999 until 2007. However, inspection of the annual breast cancer incidence rates during this period (data not shown) revealed that, after a sharp decrease in rates in 2002–2003, the lower rates subsequently remained stable. The cancer rates among children (0–19 years of age) showed an increase of 0.6% per year for both the most recent 5-year period (2003–2007) and the entire period (1992–2007).

During the period 2003–2007, incidence rates for five of the 15 most common cancers among men demonstrated a statistically significant decrease: lung and bronchus (lung), colon and rectum (colorectal), oral cavity and pharynx (oral), stomach, and malignant brain tumors. Trends in four cancers among men (melanoma of the skin, kidney and renal pelvis [kidney], pancreas, and liver and intrahepatic bile duct [liver]) showed statistically significant increases during the period 2003–2007, whereas trends for prostate, urinary bladder (bladder), and esophageal cancers and leukemia, myeloma, and non-Hodgkin lymphoma did not demonstrate a statistically significant increase or decrease. Among women, statistically significant increasing trends were noted in three of the four cancers that were increasing in men (kidney, pancreas, melanoma of the skin); leukemia and thyroid cancer also increased. Statistically significant decreasing trends for women included cancers of the breast, lung, colon and rectum, corpus uteri (uterus), cervix uteri (cervix), bladder, and oral cavity. No statistically significant trends in non-Hodgkin lymphoma, malignant brain tumors, and cancer of the ovary were observed.

Since the early 1990s, overall cancer death rates have shown a statistically significant decreasing trend among both men and women; whereas for children, cancer death rates have decreased since the mid-1970s (Table 2). Trends in death rates during the most recent 10- and 5-year periods (1998–2007 and 2003–2007) continued to decrease for seven of the top 15 cancer types in both men and women (colon and rectum, brain, stomach, and kidney cancers, and non-Hodgkin lymphoma, leukemia, and myeloma); for cancers of the lung, prostate, and oral cavity in men; and for breast and bladder cancers in women. In contrast, during the corresponding time intervals, death rates from liver cancer and melanoma of the skin in men and those for liver and pancreatic cancers in women continued to exhibit statistically significant increases. Notably, lung cancer death rates in women revealed a statistically significant decrease during the period 2003–2007, following long-term increases during the period 1975–2003, and cervical cancer death rates stabilized after decreasing for many decades. It also is noteworthy that long-term trends in death rates may mask important changes during the shorter term. For example, the 10-year AAPC (1998–2007) for lung cancer in women showed a small non-statistically significant decrease of 0.2% that was composed of a statistically significant increase of 0.3% from 1995 to 2003, followed by a statistically significant decrease of 0.9% from 2003 to 2007 (Table 2).

Black men had the highest cancer incidence rate for 2003–2007 of any racial and ethnic group (Table 3). Except among Hispanics, the top three cancer sites for men in each population group were, in rank order, prostate, lung, and colorectal cancers; among Hispanics, the colorectal cancer rate was slightly higher than the rate of lung cancer. Among women, white women had the highest overall incidence rates. Breast cancer was the most commonly diagnosed cancer among women regardless of race and ethnicity. Lung and colorectal cancers ranked second and third (respectively) among women of all races combined and for white, black, and AI/AN women. However, these rankings were reversed among API and Hispanic women. For all populations, cancer of the uterus ranked fourth. Beyond the top three cancer sites for men and top four for women, cancer rankings varied by race and ethnicity.

Incidence rates for all cancer sites combined decreased between 1998 and 2007 in both men and women in all populations; although the decrease was non-statistically significant among black or AI/AN women (Table 3). Childhood (ages 0–19 years) cancer incidence increased in all populations, although the increase was non-statistically significant for API and AI/AN children. Prostate cancer incidence showed a statistically significant decrease among AI/AN and Hispanic men. Breast cancer incidence rates decreased in all women, but the decrease was of smaller magnitude and non-statistically significant for black and API women. Among men, lung cancer incidence rates decreased for all populations; among women, no statistically significant change was observed in any racial or ethnic group. Colorectal cancer rates decreased among both men and women in all population groups, but the decrease was non-statistically significant for AI/AN women. Cancer of the uterus increased among black, API, and Hispanic women but not among white women. Incidence rates of esophageal cancer increased among white men but decreased among blacks and Hispanics.

Overall cancer death rates from 1998–2007 decreased for all race, ethnic, and sex groups except AI/AN women, among whom the decrease was non-statistically significant (Table 4). However, the largest average annual percentage decrease occurred in black and Hispanic men, approximately 2.5% per year. During the corresponding time interval, overall cancer death rates also showed a statistically significant decrease of 1%–2% per year for children aged 0–19 years in each racial and ethnic group, except in Hispanics and AI/AN, in whom rates were stable. Similarly, death rates in each racial and ethnic group decreased for each of the three major cancers in men and women (lung, colorectal, prostate, or breast), except the trends for prostate and colon cancers among AI/AN men and for lung cancer among women of all racial groups were non-statistically significant. A statistically significant increase in liver cancer death rates was noted among white men and women and black and Hispanic men, whereas the increase in pancreatic cancer death rates was noted only in white men and women.

The distribution of malignant, benign, and borderline brain and ONS tumors during the period 2004–2007 is shown for adults in Table 5 and for children in Table 6. Nonmalignant tumors were about twice as common as malignant tumors among adults (aged ≥20 years). Women had an overall brain tumor incidence rate of 26.55 per 100 000 persons; men had a corresponding rate of 22.37. Tumors of neuroepithelial tissue were the most common histological group of malignant brain tumor, occurring more frequently in men than women. Glioblastoma, which occurred 1.6 times more frequently in men, was the most common subtype of neuroepithelial tumor. The most common type of nonmalignant adult brain tumor was meningioma, which was 2.3 times more common in women than in men, with an incidence rate of 12.42 per 100 000 vs 5.46 per 100 000 persons, respectively. The rate of meningioma in women was by far the highest incidence rate for any type of brain tumor in either sex. In contrast to neuroepithelial tumors, 94.9% of which were malignant, only 2.1% of meningiomas were malignant.

Incidence rates among children aged 0–19 years were much lower than in adults (48.47 per 1 000 000 children vs 24.55 per 100 000 adults), but the tumors were much more likely to be malignant in children: 65.2% vs 33.7% malignant in adults. Boys had only slightly more tumors of neuroepithelial tissue than girls (35.71 in boys vs 32.62 in girls per 1 000 000 children), yet adult men had incidence rates of neuroepithelial tumors 1.4 times higher than women. Tumors of the meninges were more likely to be malignant in children when compared with adults and occurred in boys and girls with similar frequency. Tumors of neuroepithelial tissue were more likely to be malignant in adults, whereas germ cell tumors were more likely to be malignant in children. Tumors of the nerve sheath were rarely malignant, and lymphomas of the brain were relatively rare in both adults and children.

Whites had the highest incidence rates of brain and ONS tumors (19.0 per 100 000 persons), followed by Hispanics (17.8 per 100 000 persons) and blacks (17.7 per 100 000 persons) (Table 7). AI/ANs had lower incidence rates of brain and ONS tumors (15.3 per 100 000 persons), and APIs had the lowest incidence rates (13.5 per 100 000 persons). Glioblastoma was the most common type of malignant brain tumor, with whites having the highest incidence rates, followed by Hispanics, blacks, and AI/ANs; with APIs having roughly one-half the rate of these tumors when compared with whites. The two most common types of nonmalignant brain tumors were higher among blacks compared with whites. Meningioma was the most common brain tumor, with black women having the highest incidence rates overall (9.7 per 100 000 persons), and black men having the highest incidence rates among men. Blacks and Hispanics had the highest incidence rates of tumors of the sellar region; however, acoustic neuromas occurred more than twice as often among whites compared with blacks.

Childhood brain and ONS tumor counts and incidence rates using ICCC-3 definitions are presented in Table 8. This classification system is used widely with childhood brain and ONS tumors but does not allow for comparison with adults. Childhood cancer incidence rates presented in Table 8 may differ from those presented elsewhere in this article because of the different classification systems used to produce the tables.

Childhood brain and ONS tumors demonstrate unique age-specific incidence patterns by sex and type of tumor classification (Supplementary Table 3, available online). The incidence rate remained below six per 100 000 persons until age 25–29 years, after which the incidence increased steadily until age 84 years. By sex, incidence rates were higher in boys through age 10–14 years, after which the incidence rates became higher in women, primarily because of the large increase in meningiomas in this group (Supplementary Tables 4 and 5, available online). Incidence rates were higher in men or similar to women for all age groups for most tumors of neuroepithelial tissue, nerve sheath tumors, germ cell tumors, and lymphomas; however, women had a much higher incidence rate of meningioma than men, beginning with age group 20–24 years. Incidence rates for tumors of the sellar region were higher in women until age 45–49 years; however, this pattern was reversed at 50 years and older. Acoustic neuromas began increasing from age 40 years, peaked at ages 65–69 years, after which the incidence rates decreased.

Trends of malignant neuroepithelial tumors by histological group are shown in Figure 1 and Supplementary Table 6 (available online). The incidence of these tumors in men and women combined increased at a rate of 1.9% per year from 1980 to 1987 and decreased at a rate of 0.4% per year from 1987 to 2007, resulting in a minimal net change from 1980 to 2007. However, trends in incidence rates differed markedly among histological groups within this category. Marked changes in incidence rates for primary brain lymphomas also were observed (data not shown).

Relative survival (¹⁴) for brain and ONS tumors is strongly related to age at diagnosis, histological type, and era of diagnosis, as demonstrated in Figure 2. Five-year survival for all malignant tumors of neuroepithelial tissue as well as for most histological types and age groups has increased over time. Five-year survival among children and adolescents with all malignant tumors of neuroepithelial tissue combined increased from 62.9% for those diagnosed in 1980–1989 to 75.3% in 2000–2006 (Supplementary Table 7, available online). Favorable survival and trends also were observed for those aged 20–39 years, for whom 5-year survival increased from 54.1% in 1980–1989 to 65.1% in 2000–2006. However, among those diagnosed at age 40–64 years, survival increased from only 16.1% to 26.6%; among those diagnosed at age 65 years or older, the 5-year survival was under 5%, even in the most recent period (2003–2007). Among the most common tumors, the 5-year survival for pilocytic astrocytoma increased from 90.1% in 1980–1989 to 96.4% in 2000–2006 among children aged 0–19 years. Survival was nearly as high in those aged 20–39 years and increased markedly in those aged 40 years or older, from 46.1% in 1980–1989 to 83.5% in 2000–2006. In contrast, there was relatively little improvement in survival for astrocytomas during this time interval. Irregular trends and low survival for glioblastoma were observed in the two youngest age groups (0–19 and 20–39 years), with 5-year survival exceeding 20% only in the most recent time period. In the two older age groups (40–64 and ≥65 years), glioblastoma survival exceeded 5% at 5 years only in those aged 40–64 years diagnosed in 2000–2006. Five-year survival for oligodendroglioma and anaplastic oligodendroglioma in those aged 0–19 years increased from 70.2% to 90.8% in 2000–2006. Survival of greater than 75% was observed for those aged 20–39 years, in each time period. Although survival increased over time for those aged 40–64 years and those 65 years and older, it remained generally lower than in the younger age groups. In contrast, survival trends for neuroepithelial tumors in the grouping “embryonal, primitive, and medulloblastoma” differed by age group, with marked increases for those in the 0–19 and 20–39 year age ranges but marked decreases for those aged 40 years and older. Survival for malignant gliomas, not otherwise specified, showed greater improvement over time than for mixed gliomas.

Death rates for malignant brain tumors during 1999–2007 were stable in children (aged 0–19 years) but decreased in adults (aged ≥20 years) at a rate of 1.2% per year. Death rates for benign brain tumors decreased in both children (−2.5% per year) and adults (−2.2% per year). Trends in mortality for malignant vs non-malignant brain tumors could not be examined for earlier time periods because of inconsistencies in ICD coding over time.

High-quality cancer surveillance data now cover 93% of the US population for incidence and the entire population for mortality; however, certain limitations in data sources, data collection, and analyses may have influenced the findings of this report. First, state and national population estimates are provided annually by the Census Bureau to estimate intercensal populations. Differences between the numerator (incidence data) and denominator (US Census population data) can occur in the designation of race and/or ethnicity, place of residency, age, single vs multiple races, and the like. Every effort is made to ensure that the definition of the numerator and denominator are the same. Intercensal population estimates based on numbers updated by birth and death data are more subject to error than the estimates based on the actual count. Although these population estimates are believed to be the most accurate available, errors in the estimates may increase as time passes from the original recording of Census data. The NCI developed modifications to these Census estimates to attempt to account for changes in 2005 county-level populations because of displacement of people after Hurricanes Katrina and Rita in the most-affected counties of Louisiana, Mississippi, Alabama, and Texas. Censal and other data are used to classify the incidence cases, and census definitions are used to determine residency for the incidence cases. Race and ethnicity, however, generally are self-reported, but for the incidence cases, this information may come from a wider group of sources (patient, relative, nurse, doctor, coroner, funeral director). To enhance race and ethnicity, determination for the incidence cases, special studies and algorithms are used. For example, a match of incidence cases to IHS rosters is undertaken to correct the possible underreporting of AI/ANs, and NAACCR has developed guidelines and algorithms for enhancing Hispanic-Latino and API identification. Consistency over time in definitions for both census and incidence data is an issue, and efforts have been made to bridge single race and multiple race reporting (more information available on http://www.cdc.gov/nchs/nvss/bridged_race.htm). Second, joinpoint models were used to describe long-term (1992–2007) and short-term (1998–2007) trends. The AAPC, a summary measure of a trend over a prespecified fixed interval based on an underlying joinpoint model, was used to describe all trend data. The joinpoint model is preferable to single linear regression when a sufficient number of years are available for analysis because it enables identification of recent changes in magnitude and direction of trends. However, it may mask the underlying data and give an impression of a continuous increase or decrease over time when this is not the case. In addition, although methods have been adapted recently to adjust for delayed reporting of aggregated data similar to earlier published methods used for incidence from the nine oldest SEER registries (³⁰), methods have not been tested on data from registries outside of SEER and were employed in our analysis only for SEER 9 and SEER 13. Delayed reporting may affect the most recent joinpoint segments, overestimating recent decreases and underestimating recent increases.

Third, US Department of Veterans Affairs (VA) hospitals traditionally have been a major source of data for cancers diagnosed among veterans, representing approximately 3%–8% of cancer diagnoses among men. A 2007 policy change regarding the transfer of VA cancer data to state central cancer registries has resulted in incomplete reporting of VA hospital cases in some but not all state registries. This change has affected reporting from the third quarter of the 2004 diagnosis year through the current time period. As a result, cancer incidence rates among men for 2005–2007 are thought to be underestimated by 0.8%–2% for all cancers combined, according to independent statistical analyses conducted by the CDC and SEER. The level of underreporting varied from 0.5% to 4% according to cancer site, race, and age group (¹⁴,¹⁰⁴). The amount of underestimation also may vary by local VA facility reporting patterns and the VA’s contribution to the total number of cancers. Progress in collecting VA data has been made in many states with the enactment of special data-sharing agreements with the VA. Over time, as cancer registries receive these missing VA cases, national cancer incidence estimates will be more complete and accurate.

Fourth, as routinely noted in the Annual Reports to the Nation (¹–¹²), the broad racial and ethnic groups categorized for our analyses may mask variations in the cancer burden by country of origin; for example, Chinese and Vietnamese in the API group (¹⁰⁵) and Cubans and Mexicans in the Hispanic group (⁹,¹⁰⁶), or by other unique characteristics of high- or low-risk populations (¹⁰⁷–¹¹⁰). Also, cancer rates for populations may be limited by difficulties in ascertaining race and ethnicity information from medical records, death certificates, and census reports (²⁵).

The observed decreases in overall cancer incidence and death rates in nearly all racial and ethnic groups are highly encouraging. This progress could be accelerated by comprehensively applying existing cancer control knowledge of cancer prevention, early detection, and treatment to public health and clinical practices. Unfortunately, at this point in time, not all cancer sites are amenable to cancer control practices, and innovative methods to study these cancers and rare tumors must be developed. For example, the relative rarity of brain tumors, including many histological subtypes, has required investigators to establish consortia and pooled studies, especially for studies of genetic risk factors and gene–environment interaction (⁷⁰,¹¹¹–¹¹⁴). Many advances are being made in the molecular characterization of brain and ONS tumors and many other types of cancer. Tumor biospecimen banking linked with treatment and outcome information will be particularly important in studying the prognostic and predictive value of such markers and in developing targeted therapies (¹¹⁵–¹¹⁷) to improve effectiveness, lessen toxicity, and measure response to therapy more quickly. It is too early to assess the impact of some treatment advances or the progress in targeted therapy that is expected to emerge in future years.

The US population aged 65 years and older is expected to double in size by 2030 (about 71 million persons) compared with the number reported in the 2000 census (¹¹⁸). Improvements in health and welfare also mean that individuals are expected to live longer, often with a range of health conditions that include the diagnosis of cancer. Even with declining cancer incidence rates, the absolute number of individuals diagnosed with cancer will continue to increase because of these population changes, leading to increased demand for cancer-related medical services through the spectrum of diagnosis, active treatment, and posttreatment medical management. Effective management of the cancer burden will require the application of sound cancer control strategies in prevention, detection, treatment, and survivorship, as well as resources to provide good quality of care. Continued utilization of quality population-based data systems and translation of evidence-based clinical and basic research findings to public health practices are essential to the development of public policies for cancer.