• primary CNS lymphoma;
  • population-based;
  • SEER


  1. Top of page
  2. Abstract


The age-adjusted incidence of primary central nervous system lymphoma (PCNSL) has increased since the 1970s, and treatment for this disease has evolved considerably. The objective of this study was to examine time trends in overall survival and disease-specific mortality in a population-based cohort of patients with PCNSL.


We identified patients diagnosed with PCNSL from 1975–1999 in the Surveillance, Epidemiology, and End Results (SEER) cancer registries. To assess time trends, year of diagnosis was classified in 5-year intervals: 1975–1980, 1981–1985, 1986–1990, 1991–1995, and 1996–1999. Overall survival distributions were estimated via Kaplan-Meier methodology and a competing risk analysis was used to assess PCNSL-specific mortality. We used information on underlying cause of death to distinguish likely immunocompetent patients from those whose PCNSL was related to human immunodeficiency virus/acquired immunodeficiency syndrome (HIV/AIDS). We also examined survival stratified by age at diagnosis.


From 1975–1999, 2462 patients were diagnosed with PCNSL in SEER. Median survival was 4 months (95% CI 4, 5) for the entire cohort and 9 months (95% CI 8, 11) for the immunocompetent cohort (n = 1565). In the immunocompetent cohort, 965 of 1323 (73%) deaths were attributed to PCNSL. No significant time trend was observed in either overall or PCNSL-specific survival.


Overall survival for patients with PCNSL has not improved consistently in the past three decades despite important therapeutic advances during this time. Although results from clinical trials suggest progress in the treatment of PCNSL, survival improvements are not reflected in this population-based cohort. Cancer 2005. © 2005 American Cancer Society.

Primary central nervous system lymphoma (PCNSL) is an extranodal form of non-Hodgkin lymphoma (NHL) restricted to the brain, cerebrospinal fluid, or eyes. The age-adjusted incidence of PCNSL in the U.S. has increased substantially since the 1970s, from a rate of 0.16 per 100,000 in 1973–1984 to 0.48 per 100,000 in 1985–1997.1 Much of this increase is attributable to the spread of the human immunodeficiency virus / acquired immunodeficiency syndrome (HIV/AIDS), although the incidence in immunocompetent patients has increased as well.2–4 In both immunocompetent and immunocompromised patients, untreated PCNSL has a poor prognosis, with a median overall survival of approximately 3–4 months.5

The treatment of PCNSL has evolved considerably in recent decades. Surgical resection alone, although essential for diagnosis, has little or no survival benefit and is often accompanied by severe and permanent adverse neurological sequelae.6 Throughout the 1980s, whole-brain radiotherapy (WBRT) was the mainstay of PCNSL treatment, regardless of HIV status. Although WBRT induces a complete or partial response in most patients and extends median survival to 11–14 months,7 a large majority of patients develop recurrent disease within months of initial remission.8 More extensive neuraxis radiation has not been shown to improve survival.8

Given the high rate of disease recurrence among patients treated with radiation therapy, research in the late 1980s and early 1990s focused on the addition of chemotherapy to WBRT. In two large multicenter trials, standard NHL chemotherapy regimens (cyclophosphamide, doxorubicin, vincristine, prednisone [CHOP] and cyclophosphamide, doxorubicin, vincristine, dexamethasone [CHOD]) followed by WBRT failed to improve survival compared with WBRT alone.9, 10 However, during the 1990s multiple prospective studies, including a large Radiation Therapy Oncology Group (RTOG) study, using methotrexate-based regimens in combination with WBRT, showed improvement in survival for HIV-negative patients with a reported median survival of 30–60 months.7, 10–19 However, this improved disease control and prolonged survival was accompanied by a high risk of neurotoxicity, particularly in older patients.

Therefore, recent efforts have increasingly focused on the development of chemotherapy-only strategies to minimize adverse neurological effects while maintaining improved survival. In the absence of WBRT, methotrexate-based regimens have achieved high response rates and progression-free and overall survival without neurotoxicity.14, 19–20 Although chemotherapy was traditionally withheld in patients with HIV/AIDS because of their underlying immunosuppression, several small studies have demonstrated that chemotherapy may also be effective and tolerable in this patient subgroup.21, 22

Significant progress in the treatment of PCNSL is evident from the results of clinical trials. However, it is not clear whether therapeutic advances have led to improved prognosis for patients treated outside of the clinical trial setting. The objective of this study was to describe patterns and time trends in overall survival from PCNSL using population-based data from the Surveillance, Epidemiology, and End Results (SEER) cancer registry program.


  1. Top of page
  2. Abstract


Our sample consisted of all patients diagnosed with PCNSL from 1975–1999 reported to the SEER program. SEER is a consortium of population-based cancer registries sponsored by the National Cancer Institute in selected states and regions of the U.S. For all incident cancers in their coverage areas, the SEER registries collect data regarding site and extent of disease, surgery and radiation therapy planned or administered within 4 months of diagnosis, and sociodemographic characteristics, with active follow-up for date and cause of death. The SEER registries maintain high standards of data quality, and the program's overall rate of case ascertainment is 98%.23 We included data from the nine registries that have been included in the SEER program since its inception. These registries include the states of Connecticut, Hawaii, Iowa, New Mexico, and Utah, and the metropolitan areas of Atlanta, Detroit, San Francisco–Oakland, and Seattle.

Incident cases of PCNSL were identified in SEER as carcinomas in anatomic sites of the spinal cord, cranial nerves, and other parts of the central nervous system (CNS) (ICD-O-2 codes C70.0-C72.9) with specified non-Hodgkin lymphoma morphologies (ICD-O-2 codes 9590-9595, 9670-9698, 9711, 9714).


The primary outcome of the analysis was overall survival in months. We also evaluated specific causes of death using information in SEER from state death certificates. Deaths were attributed to PCNSL if the underlying cause of death was recorded as neoplasm of the brain and other nervous system or NHL. Observations were censored in cases where a patient was alive at the time of last follow-up. Because immunocompetent and immunosuppressed patients differ in their therapeutic options and overall survival, we attempted to analyze these groups separately. Although SEER does not collect information regarding disease etiology or noncancer comorbidity, we distinguished HIV/AIDS-related PCNSL cases on the basis of cause of death. Deaths due to HIV/AIDS were identified by the International Classification of Diseases (ICD) underlying cause-of-death code listed on the state death certificate (ICD-8 or ICD-9 codes 42.0–44.9, ICD-10 codes B20.0–B24.0). Patients who died of causes other than HIV/AIDS, and those who were alive at last follow-up, were assumed to have PCNSL unrelated to HIV/AIDS.

Statistical Analysis

Overall survival distributions were estimated using Kaplan–Meier methodology for the entire PCNSL cohort. Survival was calculated as the time from date of diagnosis to date of death or last follow-up. Survival distributions stratified by age, year of diagnosis, and radiation therapy were compared by the log-rank test. To assess potential time trends, we classified year of diagnosis in 5-year intervals: 1975–1980, 1981–1985, 1986–1990, 1991–1995, and 1996–1999. All analyses were repeated, excluding patients with PCNSL whose death was attributed to HIV/AIDS.

To assess patterns of disease-specific mortality, we also estimated the cumulative incidence of PCNSL-related death accounting for death due to other causes as a competing risk.24 We assumed that deaths with an underlying cause listed as NHL or carcinoma of the brain and nervous system were attributable to PCNSL. In competing risk analyses, strata of age and year of diagnosis were compared using a modified chi-square test.25 All analyses were performed using SAS (Cary, NC) or R Foundation for statistical computing (available from URL:


  1. Top of page
  2. Abstract

Between 1975 and 1999 there were 2462 patients identified in the SEER database with a primary diagnosis of PCNSL. Of these, 897 died of HIV/AIDS-related illness. Patient characteristics are shown in Table 1 for the entire sample and classified on the basis of HIV/AIDS-related death. Compared with patients who did not die of HIV/AIDS-related illness, the cohort with death due to HIV/AIDS was more likely to be male (93.7% vs. 56.5%), under age 50 (91.7% vs. 26.3%), black (23.2% vs. 5.9%), and from the San Francisco–Oakland SEER registry (46.4% vs. 19.5%). Use of radiation therapy was high, with 73.8% of patients without HIV/AIDS death and 65.4% of patients who died of HIV/AIDS-related illness receiving any type of radiation.

Table 1. PCNSL Patient Characteristics (SEER 1975-1999)
 All patients N = 2462 No. (%)Excluding HIV deaths N = 1565 No. (%)HIV deaths N = 897 No. (%)
Year of diagnosis
1975-1980122 (5.4)122 (8.2)0 (0.0)
1981-1985182 (8.0)182 (12.3)0 (0.0)
1986-1990552 (24.2)386 (26.1)166 (20.8)
1991-1995846 (37.1)400 (27.0)446 (55.9)
1996-1999577 (25.3)391 (26.4)186 (23.3)
20-29171 (7.0)59 (3.8)112 (12.5)
30-39609 (24.7)170 (10.9)439 (48.9)
40-49454 (18.4)182 (11.6)272 (30.3)
50-59311 (12.6)246 (15.7)65 (7.3)
60-69419 (17.0)410 (26.2)9 (1.0)
70-79394 (16.0)394 (25.2)0 (0.0)
80+104 (4.2)104 (6.7)0 (0.0)
Male1725 (70.1)885 (56.5)840 (93.7)
Female737 (29.9)680 (43.5)57 (6.3)
White2028 (82.4)1373 (87.7)655 (73.0)
Black300 (12.2)92 (5.9)208 (23.2)
Other134 (5.4)100 (6.4)34 (3.8)
SEER registry   
San Francisco721 (29.3)305 (19.5)416 (46.4)
Conn.296 (12.0)233 (14.9)63 (7.0)
Detroit387 (15.7)293 (18.7)94 (10.5)
Hawaii98 (4.0)78 (5.0)20 (2.2)
Iowa215 (8.7)193 (12.3)22 (2.5)
New Mexico97 (3.9)76 (4.9)21 (2.3)
Seattle357 (14.5)216 (13.8)141 (15.7)
Utah69 (2.8)60 (3.8)9 (1.0)
Atlanta222 (9.0)111 (7.1)111 (12.4)
NHL grade
Low74 (3.0)72 (4.6)2 (0.2)
Intermediate974 (39.6)757 (48.4)217 (24.2)
High265 (10.8)161 (10.3)104 (11.6)
Unclassified1149 (46.7)575 (36.7)574 (64.0)
None689 (28.0)386 (24.7)303 (33.8)
Any1742 (70.8)1155 (73.8)587 (65.4)
Unknown/Missing31 (1.3)24 (1.5)7 (0.8)
Vital status
Alive242 (9.8)242 (15.5)
Dead2220 (90.2)1323 (84.5)897 (100)

For the entire cohort, median overall survival was 4 months (95% confidence interval [CI], 4, 5). Excluding cases with HIV/AIDS-related deaths, median survival was 9 months (95% CI, 8, 11) (Fig. 1). In the 242 patients alive at last follow-up, median follow-up was 50 months. The overall survival distributions include all causes of death. There were 965 deaths attributed to PCNSL (i.e., recorded as NHL, brain and nervous system), 897 deaths due to HIV/AIDS, and 358 deaths due to other causes. There were 965 deaths attributed to PCNSL, 897 deaths due to HIV/AIDS, and 358 deaths due to other causes including diseases of the heart (n = 49); in situ, benign, or unknown behavior neoplasm (n = 49); non-PCNSL malignant neoplasm (n = 36); pneumonia and influenza (n = 20); other infectious and parasitic diseases (n = 21); cerebrovascular diseases (n = 17); and other causes (n = 56). The 36 deaths attributed to non-PCNSL cancers were recorded as acute lymphocytic leukemia (n = 3); acute myeloid leukemia (n = 1); Hodgkin lymphoma (n = 2); myeloma (n = 1); carcinoma of the breast (n = 4); cervix (n = 1); esophagus (n = 1); lung (n = 5); pancreas (n = 1); prostate (n = 2); ureter (n = 1); bladder (n = 1); and miscellaneous malignant cancer (n = 14). A death certificate was not available in 76 patients.

thumbnail image

Figure 1. Overall survival of the entire cohort (N = 2462) and the cohort excluding HIV-related death (N = 1565).

Download figure to PowerPoint

The cumulative incidence of PCNSL-specific mortality, accounting for deaths due to HIV/AIDS and deaths due to other causes as competing risks, was estimated for the entire cohort (Fig. 2). The sum of the three incidence curves is equivalent to one minus the overall survival curve. The cumulative incidence curve for HIV-related mortality is very steep, indicating that these patients die soon after PCNSL diagnosis.

thumbnail image

Figure 2. Cumulative incidence of mortality due to PCNSL, HIV/AIDS, and other causes (N = 2462).

Download figure to PowerPoint

Overall survival estimates stratified by patient characteristics are shown in Table 2. For the entire sample, median survival decreased dramatically after 1985, from 14 months in the cohort diagnosed 1981–1985, to 3–5 months in later cohorts. However, this is largely because of HIV/AIDS-related death, which was only recorded for cases diagnosed after 1985. Excluding deaths due to HIV/AIDS, there is no significant trend in survival among cases diagnosed in 1986–1999 (Fig. 3). Similarly, no time trends were observed when assessing disease-specific mortality.

Table 2. Overall Survival
 All patients (N= 2462)Excluding HIV deaths (n = 1565)
No.Median(95% CI)No.Median(95% CI)
  • a

    Statistically significant (P < 0.01) difference in survival distributions by the log-rank test.

Year of diagnosis      
1975-19801227(5, 11)a1227(5, 11)
1981-198518214(9, 19)18214(9, 19)
1986-19905525(4, 6)3869(7, 12)
1991-19958463(3, 4)4007(6, 10)
1996-19995775(4, 6)39112(9, 15)
≤ 6015743(3, 4)a68514(11, 18)a
> 608887(6, 8)8807(6, 8)
Age group 10-yr intervals      
20-291713(2, 4)5910(3, 13)a
30-396093(2, 3)1705(4, 6)
40-494543(3, 4)18231(17, 47)
50-5931111(8, 15)24623(14, 30)
60-6941911(8, 15)41011(9, 16)
70-793946(5, 7)3946(5, 7)
80+1044(3, 5)1044(3, 5)
Male17253(3, 4)a8858(7, 9)
Female7379(7, 12)68012(9, 14)
White20285(4, 5)a13739(8, 11)
Black3003(2, 3)929(4, 25)
Other1347(5, 11)10013(7, 18)
SEER registry      
Atlanta2223(2, 4)a1117(4, 21)
Conn.2965(4, 8)2338(6, 12)
Detroit3875(4, 6)2938(6, 12)
Hawaii987(5, 11)7811(7, 19)
Iowa2158(6, 12)19311(7, 14)
New Mexico976(5, 7)766(5, 12)
San Francisco7213(3, 3)3058(6, 12)
Seattle3575(4, 6)21613(9, 18)
Utah699(6, 26)6014(7, 33)
Low7458(26, 81)a7258(32, 84)a
Intermediate9747(6, 8)75711(9, 14)
High2654(3, 5)1617(5, 10)
Unclassified11493(3, 3)5756(5, 8)
Any17426(6, 7)a115513(11, 14)a
No6891(1, 2)3863(2, 3)
Unknown306(3, 17)247(3, 31)
thumbnail image

Figure 3. Overall survival excluding HIV/AIDS-related deaths, stratified by time period (N = 1565).

Download figure to PowerPoint

Excluding deaths due to HIV/AIDS, median survival declines with age at diagnosis, from a high of 31 months among patients age 40–49 years, to a low of 4 months among patients over the age of 80 (Fig. 4). In this analysis, the 229 patients under age 40 appear to have unexpectedly poor survival (median of 10 months for patients age 20–29 and 5 months for those in the 30–39 age group). However, this group is disproportionately male (82.1%), suggesting that many in the group actually had HIV/AIDS-related PCNSL and their underlying cause of death (78.6% died) was not correctly attributed to HIV/AIDS. The impact of this potential misclassification is minimal; excluding all patients under age 40 increased median survival from 9 months to 10 months.

thumbnail image

Figure 4. Overall survival excluding HIV/AIDS-related deaths, stratified by age (N = 1565).

Download figure to PowerPoint

The majority of patients had intermediate grade (48%) or unclassified grade (37%) PCNSL. Although only 5% of patients were classified as having low-grade PCNSL, they had a significantly longer median survival (58 mos, 95% CI, 32, 84 mos), suggesting that low-grade PCNSL may be a different biologic entity.

Approximately 74% of the cohort received radiation therapy (Table 1). There were slight differences by race (66% of blacks received radiation therapy compared with 75% of whites), tumor grade (86% of low grade, 77% of high grade, and 70% unclassified grade), and age (75% of patients ≤ age 60 versus 73% of patients older than 60). Table 3 shows the use of radiation therapy over time, stratified by age at diagnosis. Patients were more likely to receive radiation therapy in the 1980s than in the 1990s. A decline in the use of radiation therapy was seen in both patients ≤ 60 years of age and in patients older than 60.

Table 3. Patients Who Received Radiation by Year of Diagnosis and Age in the Immunocompetent Cohort (N= 1565)
Year of diagnosisAge ≤ 60Age > 60
No. (%)No. (%)
1975-198050 (88)41 (64)
1981-198563 (88)91 (83)
1986-1990131 (78)168 (80)
1991-1995134 (71)160 (78)
1996-1999108 (70)143 (63)


  1. Top of page
  2. Abstract

Despite important treatment advances, our analysis of a population-based cohort did not identify significant time trends in survival from PCNSL. Even after excluding individuals who died of HIV/AIDS, median survival from the time of PCNSL diagnosis was 9 months. This outcome is poorer than survival observed in clinical trials and hospital-based cohort studies, perhaps reflecting systematic differences between study subjects and patients in the general population.7, 10–19 Clinical trial participants and patients treated at specialized centers may have better performance status and other observable and nonobservable advantageous prognostic characteristics compared with patients treated in a community setting.

Independent of patient characteristics, our results suggest that therapeutic advances may not be widely disseminated among community practitioners. Despite the known risks of WBRT, especially among older patients, and the availability of methotrexate-based chemotherapy, over 62% of PCNSL patients age 60 and older diagnosed from 1995–1999 in the cohort received some form of radiation therapy.

Because HIV/AIDS was associated with a rise in PCNSL incidence in the 1980s and 1990s and independently increases mortality risk, we attempted to distinguish HIV/AIDS-related PCNSL from other cases on the basis of underlying cause of death. However, some cases were likely misclassified as a result of erroneous cause-of-death attribution.26–29 Although all patients alive at last follow-up (n = 242) were classified here as immunocompetent, some probably had HIV/AIDS-related PCNSL. In their SEER-based analysis of PCNSL incidence, Olson et al.1 attempted to identify a cohort of HIV/AIDS-free PCNSL cases by excluding cases among subgroups at increased risk of HIV infection: single, never-married men and women and those with ‘unknown’ marital status, and cases from the San Francisco–Oakland SEER registry. Because we focused on survival rather than incidence, and more than 90% of all cases died within the follow-up period, distinction of HIV/AIDS-related PCNSL on the basis of cause-of-death, while imperfect, was preferable for our analysis.

The observational nature of the SEER data prohibits causal inference with regard to the relationship between PCNSL treatment and outcomes. Although receipt of radiation therapy appears to be associated with better survival, this is probably a result of selection bias, assuming that patients with a better prognosis are more likely to withstand aggressive therapy. Moreover, we had no information about the receipt of chemotherapy, and therefore we cannot make any inferences about the effect of chemotherapy on survival, nor can we consider it a confounding factor. Similarly, we had no information regarding some important clinical and sociodemographic factors that affect all-cause survival, such as comorbidity and socioeconomic status. Our analysis of disease-specific survival in the presence of competing mortality risks yielded results similar to analysis of all-cause survival, suggesting that the omission of these and other potential confounders does not substantially bias our assessment of time trends in PCNSL survival.

Although we found that PCNSL survival has not increased significantly in the past three decades, we did not examine other important outcomes such as quality of life. In clinical trials, substitution of chemotherapy for WBRT is associated with reduction in neurologic toxicity. Where such treatment advances have been disseminated in the community, patients may benefit in ways that are not reflected in the metric of survival.

Ours is the second population-based study to suggest that recent therapeutic advances have not resulted in widespread improvement in patient survival. A study by Shenkier et al.30 looking at 122 successive patients treated in the British Columbia Cancer Agency described no increase in survival with evolution of therapy from WBRT to methotrexate-based chemotherapy. These two studies raise several important questions that merit further investigation. It is possible that the promising results of Phase II studies are largely the result of favorable patient selection; however, several of these studies have included patients with poor prognostic factors and the RTOG study reported by DeAngelis et al.13 was a large community-based study. Therefore, it seems more likely that despite multiple Phase II studies demonstrating improved outcomes, a new standard of care within community practice has not been adopted. PCNSL is a rare tumor and community oncologists, radiation oncologists, and neurosurgeons may only see a handful of cases over the course of their practice. Furthermore, while numerous studies have demonstrated the efficacy of high-dose methotrexate in this patient population, multiple different single-agent and multi-agent chemotherapy regimens have been used; no Phase III study has been conducted to demonstrate clear superiority of a particular regimen. Finally, there is a common perception that high-dose methotrexate is too toxic to safely administer to older patients, many of whom may continue to be treated with radiotherapy alone despite the risk of neurotoxicity and despite the inclusion of elderly patients in many Phase II studies. Therefore, to improve the outcome of all patients diagnosed with PCNSL, we need to better understand patterns of care particularly in older patients. Given the relative rarity of this tumor, newly diagnosed PCNSL patients should ideally be referred to tertiary centers with an interest in PCNSL for inclusion in clinical trials. Results of such trials can then be used to develop evidence-based treatment recommendations and guidelines for community practitioners.


  1. Top of page
  2. Abstract