FAITH G. DAVIS, PH.D., SALLY FREELS, PH.D., JAMES GRUTSCH, PH.D., SUNA BARLAS, M.S., AND STEVEN BREM, M.D.

Transcription

1 J Neurosurg 88:1 10, 1998 Survival rates in patients with primary malignant brain tumors stratified by patient age and tumor histological type: an analysis based on Surveillance, Epidemiology, and End Results (SEER) data, FAITH G. DAVIS, PH.D., SALLY FREELS, PH.D., JAMES GRUTSCH, PH.D., SUNA BARLAS, M.S., AND STEVEN BREM, M.D. Division of Epidemiology and Biostatistics, School of Public Health, University of Illinois at Chicago, Chicago, Illinois; Division of Research, Chicago Institute of Neurosurgery and Neuroscience, Chicago, Illinois; and Neuro-Oncology Program, H. Lee Mofitt Cancer Center, Tampa, Florida Object. The authors present population-based survival rate estimates for patients with malignant primary brain tumors based on an analysis of 18 years of data obtained from the Surveillance, Epidemiology, and End Results (SEER) program of the National Cancer Institute. Methods. Estimates of survival rates at 2 and 5 years after diagnosis for patients with specific histological tumor types were categorized by patient s age at diagnosis ( 20 years, years, and 65 years or older) and by the time period in which the patients were diagnosed ( , , ). Where appropriate, survival estimates were adjusted for changing patterns in the mean age at diagnosis. Conclusions. The authors observed a pattern of declining survival rates in patients with increasing age of the patient at diagnosis for most histological groups and overall improvements in survival rates of patients across these time periods adjusting for age at diagnosis. There were improvements in 2- and 5-year survival rates over the three time periods for children and adults with medulloblastoma and for adults with astrocytoma and oligodendroglioma. Improvements in survival rates for pediatric patients with medulloblastoma have leveled off in the most recent time period, and gender differences in survival rates for patients with this tumor, which were present in the 1970s, have disappeared. Clinically significant improvements in survival rates were not apparent in patients aged 65 years and older. Changes in diagnostic and treatment procedures since the mid-1970s have resulted in improved survival rates for patients diagnosed as having medulloblastoma, oligodendroglioma, and astrocytoma, controlling for age at diagnosis. Glioblastoma multiforme continues to be the most intractable brain tumor. KEY WORDS outcome malignant tumor brain tumor outcome age histology C URRENT management of primary brain tumors must be based on accurate estimates of survival times, data that have been based largely on clinical series. Survival times would be expected to have improved over time with the introduction of better diagnostic procedures and the improvements in treatment that have taken place in the last 20 years. 1 The advantage of population-based survival data is that they represent the total impact that these developments have on a population and prevent the selection biases associated with referral populations and populations recruited into clinical trials. Population-based survival rates for all patients with malignant primary brain tumors are reported annually; 24 however, a breakdown of survival rate estimates for patients according to histological tumor types is available on a more limited basis. 21,32 This report provides populationbased survival rate estimates by using the most recent data from the Surveillance, Epidemiology, and End Results (SEER) program from 1973 to 1991, which we stratified J. Neurosurg. / Volume 88 / January, 1998 by patient age and histological tumor type. Our objective was to evaluate changes in the length of survival of patients with primary malignant brain tumors that have occurred over the last two decades by separating these patients according to histological tumor type and according to age groups selected to reflect pediatric, adult, and Medicare patient populations. Clinical Material and Methods Description of the Data Set The SEER program of the National Cancer Institute collects incidence and survival data for patients with malignant tumors from selected population-based cancer registries across the United States. Because these data are population based they have the advantage of not being influenced by selection biases associated with treatment and referral patterns. Regions were selected for inclusion 1

2 F. G. Davis, et al. TABLE 1 Distribution of malignant primary brain tumors and death by patient age and tumor type: SEER Patient Age at Diagnosis (yrs) All Ages at Diagnosis No. of No. of No. of No. of No. of No. of No. of No. of Tumor Type Tumors Deaths Tumors Deaths Tumors Deaths Tumors Deaths GBM ,894 4,442 3,479 3,331 8,581 7,933 astrocytoma 1, ,388 3,018 1,705 1,611 7,253 4,952 pilocytic astrocytoma medulloblastoma ependymoma oligodendroglioma mixed glioma other gliomas ,668 1,296 pineal tumor lymphoma other NOS totals 3,433 1,382 12,054 8,973 6,460 6,127 21,947 16,482 in the SEER registry for their ability to operate a population-based cancer registry and for their diverse population subgroups. The first year in which data were available was 1973; by 1991 the data reflect patients from five states (Connecticut, Hawaii, Iowa, New Mexico, and Utah) and four metropolitan areas (Detroit, Atlanta, San Francisco Oakland, and Seattle Puget Sound). These geographic areas constitute 9.5% of the total United States population. 10 To perform our analysis, we requested and received the SEER public use data set. This data set includes selected items on individuals (tumor description, information on the first therapeutic intervention, follow-up results, and demographic data) that allow the estimation of incidence and survival rates for patients with all cancers, without containing any personal identifiers. Because these data are limited to the first treatment of the disease, analyses by treatment were not attempted. Data Selection For this study, 24,553 patients diagnosed between 1973 and 1991 with brain and central nervous system (CNS) tumors were selected by using the following International Classification of Diseases for Oncology (ICDO) topography codes: all of the brain, ; other and unspecified parts of the nervous system, ; pituitary, 194.3; and pineal region, Because our primary focus was on survival of patients with primary malignant tumors, we excluded 388 patients (1.6%) with uncertain or benign behavior codes. We also excluded 1293 patients (5.3%) with a history of any previous cancer and 346 patients (1.4%) who developed subsequent cancer because these patients overall survival is likely to be compromised by additional disease. This selection process provided a study sample of 22,526 patients who were diagnosed with primary malignant brain tumors between 1973 and 1991 and who had no previous or subsequent cancer diagnosis. Tumor Classification Tumors were classified by histological type based on ICDO morphology codes: glioblastoma multiforme (GBM; ), meningioma ( ), astrocytoma ( ), medulloblastoma ( ), nerve sheath tumors ( ), hemangioma ( ), ependymoma ( ), oligodendroglioma ( ), mixed glioma (9382), other gliomas ( ), craniopharyngioma (9350), pineal tumor ( ), lymphoma ( , , and ), not otherwise specified (NOS) ( ), and other (all remaining ICDO morphology codes). 29 The SEER program collects data only on primary malignant tumors; therefore, we excluded three histological groups meningiomas, nerve sheath tumors, and hemangiomas that would not be well represented because they contain a high proportion of benign tumors. This final exclusion (579 cases) resulted in a total sample of 21,947 malignant cases diagnosed over the 18-year period. Patient Age Groups The study population was further divided into children ( 20 years at diagnosis), younger adults (21 64 years), and older adults ( 65 years), reflecting the fact that brain tumors in children have distinct histological characteristics with a unique response to treatment and that clinical studies indicate patient survival rates vary by age at diagnosis. Adults were separated into two age groups because mortality and incidence rates in older adults have increased more rapidly than in younger adults. 7,9 The older adult group also corresponds to the age at which patients become eligible for universal health insurance, or Medicare, in the United States, a factor that is thought to have influenced the increasing rates in this group. Time Periods To facilitate the assessment of potential changes in survival rates over time, three time periods were defined according to the year of diagnosis: 1973 to 1980, 1981 to 1985, and 1986 to The first covers a time period in which diagnostic procedures were changing. The last two periods divide the decade fairly evenly, with current sur- 2 J. Neurosurg. / Volume 88 / January, 1998

3 Brain cancer survival over time vival rates reflected by the last time period. Advances in diagnostic procedures in the mid- to late 1970s and the early 1980s specifically the introduction of the computerized tomography and magnetic resonance imaging may have affected postdiagnosis survival by allowing the diagnosis of cases at an earlier stage of the disease process. 25 As such, for each tumor type the mean age of patients at diagnosis was estimated within each age group for each time period. For those patient age tumor type groups that demonstrated a statistically significant change in the mean patient age at diagnosis over time, survival estimates adjusted for age at diagnosis were computed and reported. Data Analysis The distribution of survival time from diagnosis to death was estimated using life tables with 1-month intervals. 15 Two- and 5-year rates were estimated overall, by gender and by age of the patient at diagnosis, for each histological group that had at least 20 observed deaths (this restriction prevented the reporting of imprecise estimates and limited the results presented for lymphoma and pineal tumors, the least common tumors). Within each histological tumor type and 5-year patient age group, the Cox proportional hazards regression model was used to compare survival rates between the three time periods. 5 Crude and adjusted survival rates are reported separately by time period only when the survival rate estimates show statistically significant differences over time, using a procedure that allows for multiple comparisons (p 0.05/3 for simultaneous confidence intervals [CIs]). Because patients with GBMs constitute the largest number of cases and exhibit the poorest survival time, estimates of overall survival rates for all patients with CNS tumors are primarily driven by this histological type. To evaluate whether patients with tumors of the histological types used in this analysis had a different survival pattern from patients with GBM, a Cox proportional hazards regression model was used. The survival rates of patients harboring each histological type reported show statistically significant differences compared with survival rates of patients with GBM. To provide for a comparison with the clinical literature, the median length of survival for patients harboring each histological type, controlling for patient age and year of diagnosis, was estimated using a proportional hazards regression model. We estimated survival as a function of patient age and year of diagnosis within each patient age tumor type group and used linear interpolation to estimate the median survival from each model, given the average year of diagnosis within each age group and the average age at diagnosis within each age group. The median survival time is the point in time when the survival rate is 0.5, with 50% of the patients having died and the other 50% remaining alive. Median estimates are based on the most recent decade of cases ( ) to reflect the therapeutic improvements that have been made over time and yet allow an adequate sample size to provide reasonably robust estimates. The median survival times were not reported in groups in which fewer than 20 deaths were observed and they could not be computed when the survival rate was greater that 0.5 at the last time interval J. Neurosurg. / Volume 88 / January, 1998 TABLE 2 Mean age of patient at diagnosis stratified by time period and age group for types of tumors that show statistically significant differences over time Mean Age of Patient Age at Diagnosis (yrs) Change Group in Mean Tumor Type (yrs) Age (yrs) decreasing age at diagnosis over time GBM medulloblastoma astrocytoma oligodendroglioma mixed glioma other glioma NOS other increasing age at diagnosis over time GBM astrocytoma other glioma NOS observed. This affected estimates in the youngest age group, in which the survival rate for patients with astrocytoma was 0.68 at 110 months, that for patients with medulloblastoma was 0.55 at 92 months, and that for patients with other brain diagnoses was 0.60 at 109 months. This also affected estimates in the 21- to 64-year age group, in which the survival rate for patients with ependymoma was 0.57 at 102 months. We estimated survival curves over 10 years for the most recent decade of cases ( ) for those histological tumor groups with an adequate sample size, using a Cox regression model. Survival curves were compared between male and female patients, adjusting for age at diagnosis; within tumor histological type and patient age groups; and also within year of diagnosis groups when appropriate. Results Table 1 displays the distribution of tumors and observed deaths across 12 types of tumors within each of three patient age groups ( 20 years, years, and 65 years). The distribution of tumor type varies across patient age. Astrocytomas tend to predominate in younger patients, whereas GBMs tend to predominate in older patients. Medulloblastomas constitute almost one-sixth of all tumors in children, but are rare in adults. Table 2 shows those patient age tumor types that exhibited statistically significant differences in mean patient age at diagnosis over time. In patients younger than 65 years of age at diagnosis, there was a pattern of decreasing age at diagnosis over time, whereas in patients 65 years of age or older there was a pattern of increasing age at diagnosis over time. The largest differences in the mean age at diagnosis were in the nonspecific histological type categories. For these patient age tumor groupings, subsequent tables (Tables 3 and 4) provide estimates adjusted for these age differences. This removes the effect of age differences at diagnosis in the survival estimates. 3

4 F. G. Davis, et al. TABLE 3 Two-year survival rates for patients with malignant primary brain tumors stratified by patient age at diagnosis and tumor type: SEER Patient Age at Diagnosis (yrs) Year of All Ages Tumor Type Diagnosis* Survival Rate (CI) Survival Rate (CI) Survival Rate (CI) Survival Rate (CI) GBM ( ) 0.03 ( ) 0.03 ( ) ( ) 0.13 ( ) 0.05 ( ) 0.05 ( ) ( ) 0.06 ( ) 0.06 ( ) astrocytoma ( ) 0.03 ( ) 0.43 ( ) ( ) 0.37 ( ) 0.06 ( ) 0.42 ( ) ( ) 0.07 ( ) 0.46 ( ) pilocytic astrocytoma ( ) 0.78 ( ) not applicable 0.87 ( ) medulloblastoma ( ) 0.57 ( ) 0.65 ( ) ( ) 0.88 ( ) not applicable 0.78 ( ) ( ) 0.78 ( ) 0.66 ( ) oligodendroglioma ( ) 0.63 ( ) ( ) 0.40 ( ) 0.42 ( ) 0.70 ( ) ( ) 0.81 ( ) ependymoma ( ) ( ) 0.82 ( ) 0.54 ( ) 0.76 ( ) ( ) mixed glioma ( ) 0.72 ( ) 0.42 ( ) 0.69 ( ) other gliomas ( ) ( ) 0.43 ( ) 0.04 ( ) 0.30 ( ) ( ) NOS ( ) 0.34 ( ) 0.11 ( ) 0.23 ( ) other ( ) 0.62 ( ) 0.28 ( ) 0.61 ( ) overall ( ) 0.31 ( ) 0.04 ( ) 0.25 ( ) ( ) 0.36 ( ) 0.05 ( ) 0.31 ( ) ( ) 0.41 ( ) 0.05 ( ) 0.33 ( ) * A summary estimate is reported for the entire time period, except when the time period was a statistically significant indicator of survival. Rates are adjusted for mean patient age at diagnosis. Tables 3 and 4 show 2- and 5-year survival rates and 95% simultaneous CIs within each age group for each tumor type, as well as overall by patient age group and by time period. When statistically significant differences were not present across time intervals, we report a single estimate and a 95% CI for the entire period. For example, 78% of the children with astrocytoma survived 2 years and no statistically significant differences in survival rates were observed over the three time periods. When a statistically significant difference was present across time, we report three estimates, one for each time period. For example, the 2-year survival rates for adults with astrocytoma improved over time, with a statistically significant difference apparent from the 1970s to the 1980s. The survival rate of 32% (95% CI ) in 1973 through 1980 increased to 37% (95% CI ) in 1981 through This difference is statistically significant as evidenced by the CIs: the upper boundary of 0.33 ( ) and the lower boundary of 0.36 ( ) do not overlap. Although the survival rate of 39% (95% CI ) in the latest time interval ( ) is an increase over the previous time period, this is not a statistically significant difference, as evidenced by the CIs: the upper boundary of 0.38 ( ) overlaps the lower boundary of 0.38 ( ). These estimates are adjusted for patient s age at diagnosis and are indicated in the tables to distinguish them from estimates that did not require adjustment for age at diagnosis. There is a statistically significant difference between the survival rates for the GBM group and the survival rates for each other histological group reported (data not shown). There was a strong gradient in survival rates by patient age, as demonstrated in Fig. 1 for 5-year rates. Those patients diagnosed at younger ages had at least threefold higher survival rates than those diagnosed at older ages. There were statistically significant improvements in survival rates by time period for patients with astrocytoma, medulloblastoma, and oligodendroglioma (Tables 3 and 4). Improvements in the survival rates for patients with medulloblastoma took place in the early 1973 to 1980 time period and seem to have leveled off in the most recent time period (Fig. 2). There also were statistically significant improvements in survival rates for adults with astrocytoma and those with oligodendroglioma between the 1970s and 1980s (Fig. 3). These improvements are not statistically significant between the 1981 and 1985 period and the 1986 and 1991 period. Patients with GBM consistently exhibited the poorest survival rates across all three age groups. Although modest, the twofold improvement at 2 years from 0.03 to 0.06 over time is encouraging. Improvements in 5-year sur- 4 J. Neurosurg. / Volume 88 / January, 1998

5 Brain cancer survival over time TABLE 4 Five-year survival rates for patients with malignant primary brain tumors stratified by patient age at diagnosis and tumor type: SEER Patient Age at Diagnosis (yrs) Year of All Ages Tumor Type Diagnosis* Survival Rate (CI) Survival Rate (CI) Survival Rate (CI) Survival Rate (CI) GBM ( ) 0.01 ( ) 0.01 ( ) ( ) 0.04 ( ) 0.01 ( ) 0.01 ( ) ( ) 0.02 ( ) 0.01 ( ) astrocytoma ( ) 0.01 ( ) 0.31 ( ) ( ) 0.24 ( ) 0.02 ( ) 0.29 ( ) ( ) 0.02 ( ) 0.34 ( ) pilocytic astrocytoma ( ) 0.77 ( ) not applicable 0.84 ( ) medulloblastoma ( ) 0.37 ( ) not applicable 0.40 ( ) ( ) 0.61 ( ) 0.60 ( ) ( ) 0.70 ( ) 0.60 ( ) oligodendroglioma ( ) 0.45 ( ) ( ) 0.26 ( ) 0.13 ( ) 0.51 ( ) ( ) 0.65 ( ) ependymoma ( ) ( ) 0.70 ( ) 0.42 ( ) 0.61 ( ) ( ) mixed glioma ( ) 0.52 ( ) 0.15 ( ) 0.51 ( ) other gliomas ( ) ( ) 0.30 ( ) 0.01 ( ) 0.23 ( ) ( ) NOS ( ) 0.28 ( ) 0.06 ( ) 0.17 ( ) other ( ) 0.50 ( ) 0.16 ( ) 0.50 ( ) overall ( ) 0.17 ( ) 0.02 ( ) 0.14 ( ) ( ) 0.21 ( ) 0.02 ( ) 0.18 ( ) ( ) 0.24 ( ) 0.02 ( ) 0.20 ( ) * A summary estimate is reported for the entire time period, except when the time period was a statistically significant indicator of survival. Rates are adjusted for mean patient age at diagnosis. vival rates for patients with GBM are statistically significant; however, the rate increases are trivial. The statistical significance is an artifact of the large number of cases included and the changes in 5-year rates do not seem to have clinical meaning. There were no apparent improvements in survival over the three time intervals among other histological groups: pilocytic astrocytoma, mixed gliomas, other gliomas, other CNS tumors, and unspecified CNS tumors (Tables 3 and 4). Table 5 shows estimates of median survival times within patient age tumor groups stratified by mean ages at diagnosis of 8.7 years for children, 46.2 years for younger adults, and 73.2 years for older adults and a mean year of diagnosis of A pattern of declining median duration of survival is apparent across the three patient age groups for all the tumor types reported. Children with medulloblastoma and adults with oligodendroglioma have the most favorable median survival times. Figure 4 shows entire survival curves for selected patient groups by tumor type based on data from 1981 through The long-term prognosis for patients with pilocytic astrocytomas is clearly the best; the long-term prognoses for patients with medulloblastoma, oligodendroglioma, or ependymoma appear similar to each other and are intermediate between those for patients with pilocytic astrocytoma and those with astrocytoma. In contrast, few patients with GBM survive beyond 2 years. J. Neurosurg. / Volume 88 / January, 1998 Comparing survival curves across gender, a statistically significant higher risk was found for selected groups (patient age and time of diagnosis) of males with astrocytoma and medulloblastoma. For patients with astrocytoma this increased risk appeared in men aged 21 to 64 years who were diagnosed between 1981 and 1985 (RR = 1.26, p = ) and 1986 and 1991 (RR = 1.22, p = ) and in men aged 65 years and older diagnosed from 1986 to 1991 (RR = 1.21, p = ). A higher risk in male than female patients was observed for patients 20 years of age or younger who were diagnosed with medulloblastoma between 1973 and 1980 (RR = 2.04, p = ), but this was not apparent in patients who were diagnosed in more recent years. A statistically significant higher risk was found for female patients 20 years of age or younger who had tumors in the other glioma category over the entire time period (RR = 1.32, p = ). Discussion These results indicate that overall survival rates and rates for patients with three specific histological types of malignant primary brain tumors astrocytoma, medulloblastoma, and oligodendroglioma have improved among the SEER registry areas over the last two decades and that this improvement is not due to a shift to younger 5

6 F. G. Davis, et al. FIG. 1. Bar graph displaying 5-year survival rates for patients with primary brain cancers according to patient age and tumor type. Raw data were obtained from the SEER program and cover the time period 1973 to age of the patient at diagnosis. Because these data are population based, this pattern suggests that patients are being diagnosed at an earlier stage of disease or that improvements in treatment have, at least partially, been transferred from research settings to the general population. Improvements noted in 5-year survival rates for patients with medulloblastoma are an extension of those reported by Polednak and Flannery 32 and are similar to populationbased estimates in Australia. 9 Although the survival rates improved over all three time periods, the greatest improvements are apparent in the earliest time period ( ). Recent leveling off has been reported in some regions 9 but not in others. 36 Improvements in survival rates have been reported in the Canadian population (for patients 65 years of age), 22 and for children in Italian 26 and Swedish 19 populations. In the United States, data on all brain and nervous system tumors in children (0 14 years) show 5-year survival rates that increased rapidly from 35% in the 1960 to 1963 period to 55% in the 1974 to 1976 period. This overall survival rate in children appears to have stabilized at 56% in the 1980 to 1985 period 6 and at 62% in the 1986 to 1991 period according to data reported here. If the newer diagnostic techniques introduced in the 1970s identify tumors at an earlier age, survival rates could be artificially increased. This potential lead time bias was partially addressed by controlling for the mean patient age at diagnosis across the three time intervals for those tumor types and age groups in which changes were apparent. It appears that less invasive diagnostic procedures are being used to detect some types of brain tumors at an earlier age. If diagnosis at an earlier age also reflects diagnosis at an earlier stage of disease, the survival improvements reported here may reflect treatment taking place at an earlier stage of disease or improvements in the therapy itself. In contrast, it appears that these diagnostic tools are being used increasingly among older patients because the mean age at diagnosis among the elderly is rising, although survival rates are not showing a corresponding improvement. The pattern of decreasing survival rates with increasing age of the patient across many tumor types has been noted in numerous clinical 3,4,18,30,35,38 and population-based studies. 14,20,21,23,34,36,37 Responses to both radiotherapy 13 and chemotherapy 11 have been correlated with age of patients and are thought, at least partially, to explain the importance of age in the prognosis for patients with some tumor types. In these data, the pattern of improvements in survival rates varied by tumor type and by patient age. Patients with medulloblastoma showed survival improvements in the pediatric population, whereas improvements for patients with astrocytoma or oligodendroglioma were limited to the adult population. Little to no improvement was apparent based on tumor type in the older adult, or Medicare, population. However, the crude survival rates reported here do not adjust for competing causes of death or life expectancy, refinements that may be important to the interpretation of rates in the older population. The absence of variation in survival rates of patients when compared by gender for most of the tumor type 6 J. Neurosurg. / Volume 88 / January, 1998

7 Brain cancer survival over time FIG. 2. Bar graphs showing survival rates for pediatric (0 20 years of age) and adult (21 64 years of age) patients with medulloblastoma. Raw data were obtained from the SEER program and cover the time period 1973 to **Rates are adjusted for mean age at diagnosis. groupings in these data is consistent with other population-based studies of malignant primary brain tumors. 20,33,34 The magnitude of the statistically significant differences found when comparing groups by gender for astrocytoma and other gliomas appears to be modest. Whether this discrepancy reflects differences in biological processes or differences in health care may require further investigation. Gender differences in survival have previously been reported for nonmalignant brain tumors. 33 The improvement in survival rates for patients with medulloblastoma is substantial, with 5-year survival rates increasing 20% from the 1970s to the 1980s for all age categories. In the early 1970s, patients with this type of tumor survived a median of 8 to 15 months after the onset of symptoms, 40 a much lower figure than the median survival time (74 months) observed here for adult patients in J. Neurosurg. / Volume 88 / January, 1998 FIG. 3. Bar graphs depicting survival rates for adult patients (21 64 years of age) with brain cancers. Raw data were obtained from the SEER program and cover the time period 1973 to **Rates are adjusted for mean age at diagnosis. the 1980s. Over the past decade, numerous clinical trials have shown improvement in survival rates for children with medulloblastoma who have been treated with craniospinal radiation and, more recently, for patients treated with a combination of radiation therapy and chemotherapy. Other population-based studies have reported improved survival in children with medulloblastoma, 26 although in one study that controlled for therapy the decade of treatment was not an independent prognostic factor. 13 Two- and 5-year survival rates for children with medulloblastoma leveled off in the 1980s, which may 7

8 F. G. Davis, et al. FIG. 4. Graph showing survival curves for patients with cancers diagnosed from 1981 to Raw data were obtained from the SEER program. indicate that optimum therapeutic practice has reached the general population and no further gains can be expected until new treatments are devised. This leveling of survival rates is an extension of that reported in a previous analysis of SEER data, which was restricted to patients between 0 and 14 years and to the time period 1973 to Because medulloblastomas are a common pediatric brain tumor, this prognostic improvement is important to the overall survival pattern of patients with pediatric brain tumors. Although improvements in survival rates for patients with ependymoma are not statistically significant, they are encouraging. The location of ependymomas within the brain makes them difficult to remove surgically, and a good survival outcome depends on achieving a complete resection of tumor. 28 During the past 20 years surgical advances have enabled neurosurgeons to achieve a total removal of tumor in an ever increasing percentage of patients. Improvements in survival rates over time are apparent among younger adults for two major gliomas astrocytomas and oligodendrogliomas that are primarily treated with surgical resection. However, some studies have not shown this improvement in survival, 36 and the lack of improvement in survival rates among pediatric patients with astrocytoma is consistent with that reported in the United Kingdom between 1954 and The improvement noted here takes into account the decreasing mean patient age at diagnosis, but does not directly address the stage of disease, technical improvements in surgery, and/or changes in tumor classification. The diagnostic criteria for lower grade astrocytomas have been more clearly outlined in the last decade and cases that were previously undiagnosed or classified as unknown histological composition may now be classified as astrocytomas. Patients with lower grade tumors have a better prognosis than those with higher grade tumors; therefore, their inclusion may artificially improve survival estimates. Although grade is not consistently recorded for astrocytic tumors, the 5-year survival rates for patients with specified pilocytic astrocytomas are consistent with at least one recently published clinical series. 18 The inability to separate out patients with low-grade astrocytomas and those with anaplastic astrocytomas with any confidence in this data set is a clear limitation. Patients with oligodendroglioma are generally thought to have a better prognosis than patients with astrocytoma, 2 a pattern that is not apparent in these data. This suggests that known classification problems and/or diagnostic biases may be present. It is also possible that therapeutic advances in the treatment of oligodendrogliomas have not fully reached the general pediatric population. The improvement noted at 5 years for patients with GBM is a statistical artifact reflecting the large number of patients with this tumor type and has little clinical relevance. Clinical trials do not show improvements in longterm survival for patients with GBM, and very few of these patients survive 5 years. Although the modest twofold improvement over time (from 0.03 to 0.06) at 2 years is encouraging, the reason for this is unclear, and innovative and effective new therapies are needed for these intractable tumors. The lack of improvement in survival rates for older adults suggests that the pattern of increasing mortality rates over time 7 reflects increasing incidence rates that are not distorted by varying survival rates among this age group. The reason for the increase in mortality rates has been a source of controversy and likely involves improvements in both diagnostic and reporting procedures, as well as a real increase in incidence. 8,25,31 The increasing mean patient age at diagnosis is not followed by a corresponding improvement in survival among the older age groups. These data on survival rates are important because they are population-based estimates rather than estimates based 8 J. Neurosurg. / Volume 88 / January, 1998

9 Brain cancer survival over time TABLE 5 Estimated median survival time in patients with malignant primary brain tumors: SEER * Median Survival Time (mos) Age 20 Years Age Years Age 65 Years Tumor Type at Diagnosis at Diagnosis at Diagnosis GBM astrocytoma NC pilocytic astrocytoma (18) (15) (5) medulloblastoma NC 74.0 (0) oligodendroglioma (14) ependymoma 96.5 NC (14) mixed glioma (14) other gliomas NOS other NC * Estimates were made using a mean age at diagnosis of 8.7 years for children, 46.2 years for young adults, and 73.2 years for older adults, and a mean year of diagnosis of In categories in which less than 20 deaths were recorded, the number of deaths is shown in parentheses. NC = median could not be computed. on referral populations, and they reflect the combined results of all medical practices in the SEER regions, controlling for changes in the mean patient age at diagnosis. These results reflect the medical experience of approximately 10% of the United States population, which has been oversampled for ethnic and rural populations. 10 In contrast, rates reported from surgical series or clinical trials reflect only what can be achieved under ideal therapeutic conditions with patients willing to undergo aggressive treatment; such results are not likely to be representative of the general population and may represent a biased patient population. 17,39 Furthermore, the results of the current study span an important 18-year period in which diagnostic and therapeutic practices changed significantly. 1,12 These data are limited to primary malignant brain tumors. The exclusion of benign tumor types, the most common being meningiomas and acoustic neuromas, depresses the overall estimate of survival for patients with primary brain tumors. For example, population-based 5- year survival rates for all patients with benign and malignant tumors in Victoria, Australia are 37% in males and 52% in females 33 in contrast to estimates of 25% in males and 26% in females based on these data restricted to malignant tumors. Conclusions Clinically significant improvements in survival of patients with brain tumors have taken place in the last two decades overall and for patient subpopulations with medulloblastoma. Improvements in survival rates for patients with astrocytoma or oligodendroglioma were also apparent among adults. These improvements are not explained by a shift in the patient age at diagnosis over time and suggest that therapeutic advances made in the research setting are being applied in the community setting. Modest gender differences in survival were apparent for patients with astrocytomas and other gliomas. A pattern J. Neurosurg. / Volume 88 / January, 1998 of declining survival rates with increasing patient age at diagnosis was evident for all those with nonpediatric tumor types. Although these results are encouraging, further gains in survival rates at the population level can be expected over time only if therapeutic improvements continue. Glioblastoma multiforme, the most common primary brain tumor, remains intractable, although 2-year survival rates for patients with these tumors have improved twofold. References 1. Black PM: Brain tumors. (First of two parts.) N Engl J Med 324: , Bruner JM: Neuropathology of malignant gliomas. Semin Oncol 21: , Burger PC, Green SB: Patient age, histologic features, and length of survival in patients with glioblastoma multiforme. Cancer 59: , Chandler KL, Prados MD, Malec M, et al: Long-term survival in patients with glioblastoma multiforme. Neurosurgery 32: , Cox DR: Regression models and life-tables. J R Stat Soc B 34: , Crist WM, Kun RE: Common solid tumors of childhood. N Engl J Med 324: , Davis DL, Ahlbom A, Hoel D, et al: Is brain cancer mortality increasing in industrial countries? Am J Ind Med 19: , Desmeules M, Mikkelsen T, Mao Y: Increasing incidence of primary malignant brain tumors: influence of diagnostic methods. J Natl Cancer Inst 84: , Giles GG, Gonzales MF: Epidemiology of brain tumors and factors in prognosis, in Kaye AH, Laws ER Jr (eds): Brain Tumors. An Encyclopedic Approach. New York: Churchill Livingstone, 1995, pp Gloeckler Ries LA, Miller BA, Hankey BF, et al (eds): SEER Cancer Statistics Review: Table and Graphs. NIH Pub. No Bethesda, MD: National Institutes of Health, Grant R, Liang BC, Page MA, et al: Age influences chemotherapy response in astrocytoma. Neurology 45: , Greig NH, Ries LG, Yancik R, et al: Increasing annual incidence of primary malignant brain tumors in the elderly. J Natl Cancer Inst 82: , Hershatter BW, Halperin EC, Cox EB: Medulloblastoma: the Duke University Medical Center experience. Int J Radiat Oncol Biol Phys 12: , Kallio M, Sankila R, Jääskeläinen J, et al: A population-based study on the incidence and survival rates of 3857 glioma patients diagnosed from 1953 to Cancer 68: , Kaplan EL, Meier P: Nonparametric estimation from incomplete observations. J Am Stat Assoc 53: , Kibirige MS, Birch JM, Campbell RHA, et al: Review of astrocytoma in childhood. Pediatr Hematol Oncol 6: , Kirby S, Brothers M, Irish W, et al: Evaluating glioma therapies: modeling treatments and predicting outcomes. J Natl Cancer Inst 87: , Kreth FW, Faist M, Warnke PC, et al: Interstitial radiosurgery of low-grade gliomas. J Neurosurg 82: , Lannering B, Marky I, Nordberg C: Brain tumors in childhood and adolescence in West Sweden Epidemiology and survival. Cancer 66: , Levi F, La Vecchia C, Te VC: Descriptive epidemiology of malignant brain tumors in the Swiss Canton of Vaud. Neuroepidemiology 9: ,

10 F. G. Davis, et al. 21. Mahaley MS Jr, Mettlin C, Natarajan N, et al: National survey of patterns of care for brain-tumor patients. J Neurosurg 71: , Mao Y, Desmeules M, Semenciw RM, et al: Increasing brain cancer rates in Canada. Can Med Assoc J 145: , McLendon RE, Robinson JS Jr, Chambers DB, et al: The glioblastoma multiforme in Georgia, Cancer 56: , Miller BA, Gloeckler Ries LA, Hankey BF, et al (eds): Cancer Statistics Review: NIH Publication No Bethesda, MD: National Cancer Institute, Modan B, Wagener DK, Feldman JJ, et al: Increased mortality from brain tumors: a combined outcome of diagnostic technology and change of attitude toward the elderly. Am J Epidemiol 135: , Mosso ML, Colombo R, Giordano L, et al: Childhood cancer registry of the province of Torino, Italy. Survival, incidence, and mortality over 20 years. Cancer 69: , Novakovic B: U.S. childhood cancer survival, Med Pediatr Oncol 23: , Oi S, Raimondi AJ: [Brain tumors in infants and children factors affecting prognosis. Part 1: Ependymoma.] No Shinkei Geka 8: , 1980 (Jpn) 29. Percy C, Van Holten V, Muir C (eds): ICD-O. International Classification of Diseases for Oncology, ed 2. Geneva: World Health Organization, Phuphanich S, Ferrall S, Greenberg H: Long-term survival in malignant glioma. Prognostic factors. J Flor Med Assoc 80: , Polednak AP: Re: increased mortality from brain tumors: a combined outcome of diagnostic technology and change of attitude toward the elderly. Am J Epidemiol 140: , 1994 (Letter) 32. Polednak AP, Flannery JT: Brain, other central nervous system, and eye cancer. Cancer 75: , Preston-Martin S, Staples M, Farrugia H, et al: Primary tumors of the brain, cranial nerves and cranial meninges in Victoria, Australia, : patterns of incidence and survival. Neuroepidemiology 12: , Sant M, Crosignani P, Bordo BM, et al: Incidence and survival of brain tumors: a population-based study. Tumori 74: , Shaw EG, Scheithauer BW, O Fallon JR, et al: Mixed oligoastrocytomas: a survival and prognostic factor analysis. Neurosurgery 34: , Shugg D, Allen BJ, Blizzard L, et al: Brain cancer incidence, mortality and case survival: observations from two Australian cancer registries. Int J Cancer 59: , Takeuchi K, Sano K, Nomura K: An epidemiological study of brain tumors in the elderly. Neurol Res 13:21 24, Walker MD, Green SB, Byar DP, et al: Randomized comparisons of radiotherapy and nitrosoureas for the treatment of malignant glioma after surgery. N Engl J Med 303: , Winger MJ, Macdonald DR, Schold SC Jr, et al: Selection bias in clinical trials of anaplastic glioma. Ann Neurol 26: , Zulch KJ: Brain Tumors: Their Biology and Pathology. Berlin: Springer-Verlag, 1986, pp Manuscript received July 18, Accepted in final form July 15, This work was conducted under contract to the Central Brain Tumor Registry of the United States and was supported by the Pediatric Brain Tumor Foundation of the United States through the Ride for Kids Fundraising Program, which is sponsored by the American Honda Motor Company, Motorcycle Division. Address reprint requests to: Faith G. Davis, Ph.D., Division of Epidemiology and Biostatistics (MC 922), School of Public Health, 2121 West Taylor Street, Room 505, University of Illinois at Chicago, Chicago, Illinois J. Neurosurg. / Volume 88 / January, 1998