Years of life lived with disease and years of potential life lost in children who die of cancer in the United States, 2009

Incidence and survival rates are commonly reported statistics, but these may fail to capture the full impact of childhood cancers. We describe the years of potential life lost (YPLL) and years of life lived with disease (YLLD) in children and adolescents who died of cancer in the United States to estimate the impact of childhood cancer in the United States in 2009. We examined mortality data in 2009 among children and adolescents <20 years old in both the National Vital Statistics System (NVSS) and the Surveillance, Epidemiology, and End Results (SEER) datasets. YPLL and YLLD were calculated for all deaths due to cancer. Histology-specific YPLL and YLLD of central nervous system (CNS) tumors, leukemia, and lymphoma were estimated using SEER. There were 2233 deaths and 153,390.4 YPLL due to neoplasm in 2009. CNS tumors were the largest cause of YPLL (31%) among deaths due to cancer and were the cause of 1.4% of YPLL due to all causes. For specific histologies, the greatest mean YPLL per death was due to atypical teratoid/rhabdoid tumor (78.0 years lost). The histology with the highest mean YLLD per death in children and adolescents who died of cancer was primitive neuroectodermal tumor (4.6 years lived). CNS tumors are the most common solid malignancy in individuals <20 years old and have the highest YPLL cost of all cancers. This offers the first histology-specific description of YPLL in children and adolescents and proposes a new measure of cancer impact, YLLD, in individuals who die of their disease. YPLL and YLLD complement traditional indicators of mortality and help place CNS tumors in the context of other childhood malignancies.


Introduction
Cancer in children and adolescents <20 years old is rare, but is one of the leading causes of death among children in the United States [1,2]. Age-adjusted incidence rates (AAIRs) of cancer diagnosed in individuals <20 years old have increased significantly since 1975, and this increase has particularly been driven by increases in the incidence
Mortality rates of childhood cancer have decreased by over 50% between 1975 and 2014 [3]. Between 2003 and 2009, the 5-year survival rate for children <20 years old diagnosed with any kind of cancer was 83% [3]. Survival rates vary significantly by tumor type. Improvements in survival after diagnosis with CNS tumors have not kept pace with more common cancers of childhood [3,5]. While survival statistics provide important information for cancer surveillance, mortality rates cannot fully describe the burden caused by premature mortality due to childhood cancers. Previous large cohort studies-in particular, the Childhood Cancer Survivor Study [6]have provided description of the impact of childhood cancer on adult survivors. Few studies, however, have examined the impact of childhood death due to cancer. We sought to measure the impact of death due to childhood brain and other CNS tumors by examining years of potential life lost (YPLL), a critical benchmark for measur-ing both the public health and economic impact of diseases. We calculate YPLL of specific CNS tumor histologies and compare these results to other common childhood and adolescent cancers. Furthermore, we describe years of life lived with disease (YLLD) in children and adolescents who die of these tumors in order to better measure the impact of childhood cancer on individuals. These statistics are particularly useful in CNS tumors and other rare cancers where overall incidence and/or survival calculations fail to communicate the impact of these tumors in comparison to more common cancers.

Materials and Methods
This analysis of population-based and de-identified datasets was conducted under a protocol deemed exempt by the University Hospitals Case Medical Center Institutional Review Board (IRB). The methods used in this analysis have been previously used to describe the burden of CNS tumors in the United States [7]. This analysis utilized data from two sources in order to capture overall YPLL for the United States, as well as to assess YPLL and YLLD by specific histologies (Fig. 1  . YPLL was calculated as the difference between age of death and the average life expectancy for a person of the same age, race, and ethnicity in the United States in 2009. Total YPLL was the sum of YPLL among individuals who die of cancer, and mean YPLL was calculated by dividing total YPLL by the number of deaths. The NVSS dataset was also used to calculate mortality rates per 100,000 population for each histology group. In order to assess histology-specific YPLL and YLLD, we utilized the Surveillance, Epidemiology and End Results (SEER) dataset released by the National Cancer Institute [10.] This dataset contains all cancer diagnoses from within 18 SEER-funded registries (representing 26% of the United States population) between 1973-2000 (varying depending on registry) and 2009, as well as active follow-up for outcomes. Starting in 2004 with the passage of the Benign Brain Tumor Cancer Registries Amendment, the SEER registries also began collecting nonmalignant brain tumor cases. Using only persons who died as a result of their cancer, we used month of diagnosis, year of diagnosis, survival months, and age of diagnosis to calculate approximate month and year of death and approximate age at death. Each death record was matched with expected years of potential life based on age at death, gender, race, and ethnicity to estimate potential remaining years of life at time of death. Diagnoses in the SEER dataset included brain and other CNS tumors  Table 1 for the total number of deaths in 2009 due to each of these causes. Mortality and incidence rates per 100,000 population for each histology group were also calculated from the SEER dataset.
YLLD was defined as the difference between age at diagnosis and age of death for those who die of cancer before 20 years of age. Total YLLD was the sum of survival years for each person in the dataset, and mean YLLD was estimated by dividing total YLLD by number of deaths. All analyses were completed using R version 3.1.1 [11] and SEER*Stat 8.1.5 [12]. All age-adjusted rates that were included have been adjusted to the 2000 U.S. standard population.

Results
There were a total of 48,259 deaths in children <20 years old in the United States in 2009, of which 2223 were attributed to neoplasms (Table 1). Within the SEER dataset, there were 547 deaths (~25% of NVSS). These deaths more commonly occurred in males and white non-Hispanics. CNS tumors, leukemia, and lymphoma accounted for~61% of total deaths due to cancer in 2009. The largest proportion of deaths occurred in children ages 15-19 years, largely due to leukemia. For children who died at ages 0-4 years, CNS tumors (including tumors of the brain, spine, or other CNS locations) caused~31% of all deaths as compared to 43.1% of those in children ages 5-9 years (Table 2). Mean YPLL were not significantly different between the NVSS and SEER populations ( Table 2).
The age-adjusted mortality rates for each of the four disease groups (CNS tumors, leukemia, Hodgkin lymphoma, non-Hodgkin lymphoma) were similar in both datasets in 2009 ( Fig. 2A) with CNS tumors having the highest age-adjusted mortality rate (NVSS: CNS = 8.1 per 1,000,000; leukemia = 6.9 per 1,000,000; Hodgkin lymphoma = 0.3 per 1,000,000; non-Hodgkin lymphoma = 0.9 per 1,000,000). CNS tumors also had the highest AAIR in 2009 (AAIR = 5.2, 95% CI = 4.8-5.4), with leukemia having the second highest AAIR (AAIR = 4.5, 95% CI = 4.3-4.8) (Fig. 2B). In both datasets, approximately 30% of deaths were attributed to CNS tumors,~25% to leukemia,~3% to non-Hodgkin lymphoma, and~1% due to Hodgkin lymphoma (Fig. 2C). There were a total of 153,390.4 YPLL due to neoplasms in 2009 (Table 2), representing 4.4% of a total 3,489,798.0 YPLL due to all   causes in individuals <20. Of these,~31% were due to CNS tumors,~26% to leukemia,~3% to non-Hodgkin lymphoma, and <1% to Hodgkin lymphoma. For each death due to disease, mean YPLL was 70.3 for CNS tumors, 69.2 for leukemia, 65.7 for non-Hodgkin lymphoma, and 62.9 for Hodgkin lymphoma (Fig. 2D). CNS tumors caused the highest loss of potential life years, followed by leukemia, non-Hodgkin lymphoma, and Hodgkin lymphoma (Fig. 3A). Among all specific histologies examined, the histologic types with the highest mean YPLL were atypical teratoid/rhabdoid tumors (ATRT, ICD-O-3 histology code: 9508) (mean YPLL = 78.04) and high-grade gliomas (ICD-O-3 histology codes: 9381, 9401, 9440-9442, 9451, 9460 for all sites, 9380 and 9400 only for site code C71.7) (mean YPLL = 70.67) (Fig. 3B). Mean YLLD was not significantly different between CNS tumors and other common childhood can-cers in children and adolescents who die of cancer before 20 years of age (leukemia P = 0.224, non-Hodgkin lymphoma P = 0.308, and Hodgkin lymphoma P = 0.623) ( Table 3). The histologies with the highest mean YLLD were primitive neuroectodermal tumor (PNET) (mean YLLD = 4.59), medulloblastoma (mean YLLD = 3.17), and acute lymphoblastic leukemia (mean YLLD = 3.09). The histology with the lowest mean YLLD was ATRT (mean YLLD = 0.63). Individuals who died of gliomas, non-Hodgkin lymphoma, Hodgkin lymphoma, and myeloid leukemia lived the shortest after diagnosis on average (Fig. 3C). There was a significant difference between the different embryonal subtypes: PNET had the longest mean YLLD, while ATRT had the shortest followed by acute myeloid leukemia (Fig. 3D). Those diagnosed with gliomas and embryonal tumors died at the youngest median ages, while those diagnosed with non-Hodgkin lymphoma    or Hodgkin lymphoma died at the oldest median ages (Fig. 3E). Of all embryonal tumors, ATRT had both the lowest median age of diagnosis and death (Fig. 3F). Both YPLL and YLLD varied with age at diagnosis for all tumor types (Table 4). For children diagnosed with cancer between 0 and 4 years, embryonal tumors were the largest cause of YPLL (Fig. 4A). For children 5-14, gliomas were the largest contributor to YPLL with myeloid leukemia causing the highest YPLL in those diagnosed at ages 15-19. Embryonal tumors and acute lymphoblastic leukemia caused nearly equal amounts of YLLD in children ≤4 years at time of diagnosis (Fig. 4B). In children 5-14 years, lymphoid leukemia was the biggest source of YLLD. Once children were between 15 and 19 years, gliomas were the largest cause of YLLD with almost no YLLD due to embryonal tumors.

Discussion
CNS tumors are the second most common malignancy in children and are the most common cause of cancerrelated death in 0-19 year olds [13]. However, these tumors often fail to receive the attention or research funding commensurate with their impact [14]. We sought to examine YPLL to complement and counterbalance measures of impact based only on incidence and total mortality. YPLL measures the total number of years cut short due to disease and, therefore, gives a more comprehensive assessment of diseases that kill early or are incur-able [15]. The second instrument used in this report, YLLD, measures the total years from diagnosis to death for those who die of malignancy before 20 years of age. YLLD presents another method to measure the impact of disease and may provide a potential benchmark to gauge clinical success by using registry data for tumors that are rapidly fatal.
Prior reports have examined YPLL due to cancer in the general population [16] and among childhood CNS tumors by site of tumor [7]. Similar to our results, Thuppal et al. found 697 total deaths due to brain and CNS tumors in children 0-19 in 2001, compared to the 678 total deaths due to brain and CNS tumors in children 0-19 found in this analysis. Unlike prior reports, our study is the first description of YPLL in childhood tumors defined by histological subtype. An understanding of survival statistics by histological subtype is essential to the analysis of childhood CNS tumors, which otherwise represent a diverse population of tumors with widely variable outcomes. An analysis by histological subtype is made possible by combining two large national registry datasets, SEER and NVSS. Although these two data sources cover different and overlapping geographic areas, a comparison of relative age-adjusted mortality, disease-attributed death, and mean YPLL in 2009, as well as population demographics, demonstrates concordant results and confirms the validity of the SEER sample. In addition, SEER incidence data in 2009 and median age at diagnosis reported here remain remarkably similar to data presented from    6%], respectively). After considering specific histology classifications, gliomas (including high-grade glioma, low-grade glioma, and ependymoma) are responsible for the greatest loss of potential life years in children and adolescents-almost 60% greater than acute lymphoblastic leukemia. Among gliomas, high-grade glioma is responsible for approximately 75% of YPLL. Although high-grade gliomas are uncommon compared to acute lymphoblastic leukemia, this analysis reflects the loss of life due to this disease and suggests the potential for future clinical improvement. Furthermore, YPLL measures the impact of disease on society and complements crude mortality indicators by demonstrating the importance of childhood brain tumors to overall life lost due to cancer. Childhood brain and CNS tumors are the source of 31.1% of all YPLL due to childhood cancer, and the source of 1.4% of all YPLL due to all causes in 2009. In comparison, the causes of death that contributed the largest amount of YPLL were death due to perinatal causes (1,015,428.7 YPLL, or 29.1%), accidents (692,637.5 YPLL, or 20.0%), and congenital causes (494,527.9 YPLL or 14.2%).
Mean YPLL offers another view of the impact of cancer-related deaths by measuring the number of years of life lost due to an individual cancer death. In this measure, as well, the impact of brain and other CNS tumors is greater than leukemia or lymphoma in children <20 years old and reflects the young age at which death occurs in these children. YPLL has been used to measure predicted clinical impact of new therapeutic modalities [18], and mean YPLL is considered an important factor in the allocation of research funds [14]. For childhood CNS tumors, these measures demonstrate the relative impact of these tumors on society.
YLLD is a measure of cancer's impact on children and their families, reflecting the time spent between diagnosis and death for those who eventually die of their cancer. While it is impossible to predict the quality of life or extent of disability during this period, YLLD represents the area under the survival curve for the portion of indi-  viduals who die of their disease and presents another measure of clinical progress for tumors that are fatal. YLLD can be affected by available therapeutic options as well as disease severity. Total YLLD due to brain and other CNS tumors (gliomas and embryonal tumors) outpaces that due to leukemia (lymphoid and myeloid) in children <10 years old, although acute lymphoblastic leukemia remains the single largest contributor to YLLD in this age group. In our study, mean YLLD was greatest among children and adolescents with medulloblastoma, PNET, and acute lymphoblastic leukemia, which may reflect longer treatment courses or multiple treatment options that may prolong survival in these specific diseases. Mean YLLD was smallest among individuals with ATRT (0.63 years) and those with high-grade glioma (including diffuse intrinsic pontine glioma) between 5 and 9 years old (1.18 years), reflecting the shortened life expectancies for these diagnoses and the paucity of effective therapies for relapsed/refractory disease. To use YLLD as a potential benchmark of clinical response, more research is needed. Future studies should investigate the effect of treatments and demographic factors on YLLD to validate this measure. In addition, the end of life is a highly individual time during which some families may wish to extend life and others may focus on treating symptoms. Measuring quality of life for patients and their families during this period will be essential to understand the impact and importance of YLLD. Our study is subject to certain limitations. While registry data from large national databases (SEER and NVSS) offer the advantage of covering a substantial proportion of the United States population, we are limited to the data captured in these registries. The effect of therapies and the quality of life during treatment are not available in these data and should be investigated using different data sources. Our analysis also investigates a single time point (the year 2009) as characteristic of the current impact of death due to childhood cancer. Statistics on death certificates often lag behind the present day; 2009 is the most current year for which life expectancy tables are available from the NCHS. In general, 2009 appears to be representative of recent epidemiology studies of pediatric and adolescent cancers, although the incidence of new CNS tumors exceeded leukemia in that year. Because this analysis focuses on deaths occurring in 2009, rather than new cancer incidence, CNS tumor incidence will not affect YPLL or YLLD statistics reported here.
While our analytic methods focus on childhood deaths due to CNS tumors, we may fail to capture deaths that occur after 20 years of age or in long-term survivors of cancer [19]. This may result in the under-counting of both YPLL and YLLD for pediatric cancer.
Although CNS tumors were first recorded in the SEER database in 1973, SEER sites were added until the year 2000, and nonmalignant CNS tumors began to be included starting 1 January 2004 [20]. This may limit the YPLL and YLLD in SEER sites added recently and will decrease the apparent impact of histologically benign tumors, where each subject could only contribute a maximum of 5 years to either analysis. This may contribute to the interpretation of YLLD for low-grade glioma which is only nominally greater than that for highgrade glioma (1.47 vs. 1.45 years) in our analysis. However, the diminished YLLD for low-grade glioma also reflects that perioperative and early mortality may be a significant factor affecting overall mortality for some low-grade gliomas.

Conclusion
CNS tumors are the second most common malignancy in children but have the highest cost in YPLL. Among specific histologies, high-grade gliomas-including diffuse intrinsic pontine glioma-have the highest YPLL, almost 17% greater than the malignancy responsible for the next greatest YPLL, acute lymphoblastic leukemia. ATRT is the tumor responsible for the shortest mean YLLD, reflecting its rapid progression from diagnosis to death. This study reports on measures of cancer impact that are complementary to traditional indicators such as total mortality, and places pediatric and adolescent CNS tumors in the context of other common pediatric and adolescent malignancies. The epidemiologic data contained in this report contribute to expanding our understanding of the far-reaching implications of CNS tumors on the pediatric and adolescent population in the United States and, hopefully, may stimulate research into reducing the impact of these devastating tumors.