Introduction

High-grade gliomas are the most common primary brain tumor, accounting for more than half of over 20,000 primary brain cancers diagnosed annually in the United States. These cancers are challenging to treat and are associated with relatively short survival [1]. Traditionally, estimates of survival are made at the time of diagnosis, most often utilizing median progression free and overall survival as well as 2- and 5-years survival rates. These estimates provide an overall prediction of disease outcome that are most helpful for cancer surveillance and public health intervention purposes. However, for individual survivors and physicians caring for these patients, these estimates are often not informative, or even misleading. For instance, patients who have survived a certain period of time after diagnosis (e.g. 2 years) likely have different probabilities of surviving the next 1 year from those that were estimated at the time of diagnosis. This is because a large proportion of patients will die within the first 1 or 2 years, and patients who have survived longer than 2 years likely have different survival patterns from those at the time of diagnosis. Conditional survival probabilities address this problem by calculating the survival probabilities based on the patients who have survived beyond a certain period of time, and therefore provide more accurate and dynamic estimates of survival after the initial diagnosis.

The present study, utilizing the Los Angeles County tumor registry data, was designed to answer the following question: What are the conditional survival probabilities of surviving one additional year, given that patients have already survived 1, 2, 3, 4 and 5 years after diagnosis of a high-grade glioma?

Methods

The population-based cancer registry, Cancer Surveillance Program (CSP) is both a member of the statewide population-based surveillance system, the California Cancer Registry (CCR), and part of the surveillance, epidemiology, and end results (SEER) program [2]. It extracts records of cancer patients from hospitals, institutes, clinics and medical laboratories equipped to diagnose cancer in Los Angeles County. The information collected by CSP includes demographic information, tumor characteristics, diagnostic information, extent of disease and limited treatment information, as well as the last follow-up date, vital status and cause of death.

Malignant glioma cases were identified using the International Classification of Diseases for Oncology (ICD-O-3) codes [3]. Cancer histology was characterized by the new World Health Organization (WHO) classification for brain tumors [4]. The selected glioma histology’s included malignant glioma, anaplastic astrocytoma, gemistocytic astrocytoma, glioblastoma, giant cell glioblastoma and gliosarcoma. Anaplastic oligodendroglial tumors were not included. 2,983 cases of high-grade glioma diagnosed in Los Angeles County from 01/01/1990 to 12/31/2000 were retrieved. Patients who were younger than 19 years old at the time of diagnosis or who were diagnosed with malignant gliomas in the brainstem, spinal cord, optic nerve or ventricle were excluded. A total 2,743 adult patients with supratentorial high-grade gliomas, WHO grade III or IV gliomas were included in this study. Cases were followed for survival through June 2004.

Overall survival was calculated as the time from the date of diagnosis (first surgery and pathological diagnosis) to the date of death for any reason, or the date the patient was last known to be alive. In contrast, the conditional survival was calculated as the time from the date of a landmark (e.g. 1 year anniversary post-diagnosis) to the date of death or the date the patient was last known to be alive, conditional on the patient surviving beyond the landmark date. Thus the conditional survival probabilities excluded patients who did not survive beyond the landmark date; stated differently, conditional survival analysis includes only those patients who are known to be alive and therefore still “at risk” of surviving or dying after the landmark date.

Numbers and percentages were used to summarize the study data. In univariate analysis, the log-rank test was used to test the association of overall survival with demographics, marital status, socioeconomic status, tumor characteristics at diagnosis, primary treatment, treatment start time, and type of treating hospital. The relative risk of death as well as associated 95 % confidence intervals were calculated based on the Pike estimate [5], using the observed and expected number of events from log-rank test statistic. Those factors significantly associated with overall survival in the univariate analysis were included in a multivariable analysis using the Cox proportional hazards model.

Kaplan–Meier plots were used to estimate and illustrate the probabilities of overall and conditional survival [6]. Bar charts were used to show the estimated probability of surviving 5 years from the time of diagnosis and conditional on having already survived for 1, or 2 or 3 years by age (≤55 years, 56–70 years and >70 years) and WHO status.

Results

A total of 2,743 malignant glioma patients meeting the inclusion criteria were identified in the CSP database. Baseline demographics and disease characteristics are summarized in Table 1. The median age at diagnosis was 64 years old; slightly more patients were male (55 %) than female (45 %). More patients were diagnosed with WHO grade IV gliomas (76 %) than with WHO grade III gliomas (24 %). After diagnosis, 83 % of patients underwent some form of treatment (i.e. surgery, radiation or chemotherapy). The associations of overall survival with demographics and disease baseline characteristics are shown in Table 2. Age at diagnosis, tumor site, WHO grade, primary treatment, time to start of treatment, as well as whether the patient was treated at a teaching hospital were statistically significantly associated with overall survival in univariate analysis (Table 2) and in multivariable analysis (Table 3, unconditional portion). Overall median survival and its 95 % confidence interval were 6.6 (6.2, 6.9) months with median follow-up of 8.6 years.

Table 1 Baseline characteristics
Full size table
Table 2 Association of baseline characteristics with overall survival—univariate analysis
Full size table
Table 3 Association of baseline characteristics with survival—multivariable analysis
Full size table

Trend of survival

The probability of overall survival (unconditional) at 1 year after diagnosis was 0.31 ± 0.01 and probability of surviving 2, 3, 4, and 5 years from diagnosis was 0.13 ± 0.01, 0.10 ± 0.01, 0.08 ± 0.01, and 0.06 ± 0.01, respectively (Fig. 1). In the conditional survival analysis, as patients remained alive longer (e.g. 1 and 2 years after diagnosis), the (conditional) probability of surviving additional years increased (Fig. 2a–d). The probabilities of survival at 1 year after diagnosis, and for one additional year conditional on having already survived 1, 2, 3, 4, and 5 post-diagnosis years are listed in Table 4. This increase was greatest during the first 2 post-diagnosis years, from 0.43 ± 0.02 conditional on surviving 1 year post diagnosis, to 0.72 ± 0.02 conditional on surviving 2 years after diagnosis, and leveled out as patients survived longer, i.e., 0.80 ± 0.03, 0.84 ± 0.03 and 0.91 ± 0.02 conditional on surviving 3, 4, and 5 years after diagnosis respectively.

Fig. 1
figure 1

Kaplan-Meier plots of overall (unconditional) survival from diagnosis of all patients with malignant gliomas (dashed red line) and by WHO grade: grade III (solid green line) and grade IV (solid blue line)

Full size image
Fig. 2
figure 2

Kaplan-Meier plots of conditional survival of patients with malignant gliomas (all patients: dashed red line) and by WHO Grade (Grade III: solid green line. Grade IV: solid blue line). a Survival conditional on having survived one year post diagnosis. b Survival conditional on having survived 2 years post diagnosis. c Survival conditional on having survived three years post diagnosis. d: Survival conditional on having survived 4 years post diagnosis

Full size image
Table 4 Comparing unconditional probability of survival at 1 year with conditional probabilities of surviving one additional year
Full size table

The prognosis for patients with WHO grade III tumors was better than those with WHO IV gliomas, estimated by both unconditional and conditional survival probabilities—however, this difference, which was both substantial and statistically significant at the time of diagnosis and conditional on surviving 1 and 2 years, was no longer substantial or statistically significant once patients had survived 4 or 5 years; absolute differences in the probability of surviving one additional year were 0.47–0.26 = 0.21, 0.40, and 0.30 conditional on surviving 0, 1, and 2 years, but were 0.05, and 0.08, conditional on surviving 4 and 5 years, respectively. In both groups of patients, the conditional probability of one additional year of survival increased as patients survived longer (Table 4, Figs. 1, 2a–d). This trend was also seen in the probability of surviving 5 additional years survival conditioned on having survived for 1, 2, or 3 years (Fig. 3a).

Fig. 3
figure 3

Unconditional and conditional probabilities of surviving 5 additional years according to age (a) and WHO grade (b). Blue bars represent (unconditional) probability of surviving 5 years after diagnosis. Green, yellow, and red bars represent the probability of surviving an additional 5 years, conditional on already having survived 1, 2 or 3 years post diagnosis, respectively

Full size image

A similar pattern was seen when patients were grouped by age: (1) younger patients manifested better one-year survival probabilities, (2) the conditional probabilities of one additional year of survival increased each year post-diagnosis, and (3) the differences among the age groups decreased as the time post-diagnosis increases (Fig. 3b).

Discussion

High-grade glioma is a relatively rare cancer with very poor survival. According to the Central Brain Tumor Registry in the United States during the years of 1998 to 2002, the estimated age-adjusted incidence of high-grade gliomas is 6.7 and 4.6 patients per 100,000 persons per year, for WHO grade III and grade IV gliomas respectively (Central Brain Tumor Registry in the United States, 2006). Gliomas account for 1.4 % of all new cancer cases and 2.3 % of all cancer deaths in the United States [2]. In general, non-Hispanic whites have the highest incidence of glioma and mortality rates from glioma among all the ethnic groups [2]. Among all the newly diagnosed cases of primary brain cancer in the United States, approximately 50 % are histologically classified as glioma and 50 % of all gliomas are glioblastoma (WHO grade IV) [7].

In the present study, data were analyzed from a large cohort of 2,743 adult patients with supratentorial high-grade glioma diagnosed in Los Angeles County between the years of 1990 and 2000, the largest reported study of conditional survival analyses in patients with high-grade glioma. In this study conditional probabilities to predict patients’ survival were used. The study showed that as patients survived longer after initial diagnosis, their probability of subsequent survival markedly increased.

The traditional estimates of survival are the survival rates which vary greatly depending upon prognostic characteristics and thus are challenging to apply for individual patients [8–11]. Conditional survival probabilities are estimated among a specific cohort of patients by excluding those who did not survive to the start of the landmark or time point of interest. Therefore, this conditioned cohort study represents a subpopulation of patients, in contrast to the analysis beginning at the time of diagnosis when all patients were included. In the present study, the estimated chance of 5-year survival after diagnosis is 6 % (136 out of 2,743 of patients survived 5 years after diagnosis—Fig. 1), a figure often cited in the literature and based on university brain tumor treatment centers [12–14]. However, when analyzed by conditional survival probabilities and conditional median survival times, the data demonstrate that as patients survive longer after diagnosis, their probabilities of surviving an additional one year increases (ranging from 43 % conditioned on 1-year survival to 91 % conditioned on 5-year survival) (Table 4), irrespective of original glioma grade.

The survival rate among the study population dropped steeply after diagnosis and overall median survival is approximately 6.6 months (Fig. 1). The very short median survival, nearly 40 % of that reported in clinical trials in patients with glioblastoma (median survival 14.6 months), reflects perhaps more accurately survival in a large urban community treating all patients without the restrictions imposed by trial inclusion or exclusion criteria [15] and in a wide variety of hospitals. Polley et al. [16] recently reported the overall and conditional probabilities of survival on 498 patients treated on 6 Phase II trials between 1975 and 2002; although 128 of these 498 received temozolomide, their outcome was not significantly different from patients on the pre-temozolomide protocols. Overall in the Polley series, the 4 year survival was 7 % (95 % CI: 5, 10 %); this was doubled what was observed in the group of WHO grade IV patients in this manuscript which was 3 % (95 % CI: 2, 4 %). As with the series reported in this manuscript, the conditional probabilities increased as the time post diagnosis increased (Fig. 4). Interestingly however, the (unconditional) survival during the first year was nearly doubled in the Polley series (58 % (95 % CI: 54, 63 %) vs. 26 % (95 % CI: 24 %, 28 %)). Selecting patients by way of entry into clinical trials, usually by specifying independence in activities of daily living (i.e. a Karnofsky performance status >60) and age (<71 years), results in what might be expected as optimal survival rates based on contemporary treatment paradigms. Rarely reported are trials designed for elderly patients with high-grade gliomas or for patients with compromised neurological performance. Los Angeles County, the site of the present study, is a large ethnically diverse community with a significant proportion of socioeconomically challenged patients that negatively impacts survival in essentially all cancer diagnoses.

Fig. 4
figure 4

One year survival probabilities (represented as percents) and 95 % confidence intervals for glioblastoma (WHO grade IV). Patients in the Los Angeles County CSP cancer registry series (in red squares) are compared to patients enrolled on 6 Phase II clinical trials reported by Polley et al. [16] (in blue circles). Unconditional probabilities are represented at 0 years after diagnosis—i.e. Survival since diagnosis; conditional probabilities are provided for having survived 1, 2, 3 and 4 years post

Full size image

Age is a well recognized negative prognostic factor for high-grade glioma and this study recapitulates its prognostic significance (Fig. 3b) [17–19]. However, the study data also suggests that when a high-grade glioma was diagnosed in the older age cohort, defined as patient’s ≥56 years of age, the probabilities of surviving one additional year increase substantially as patients survive longer (Table 4).

In conclusion, the present study examines the outcome of 2,743 patients with high-grade glioma, all identified in a population-based database, making it an ideal platform for performing conditional survival analyses. Although the CSP, the source where the data in this study was retrieved, is a California based cancer registry, the results are generalizable to patients with high-grade glioma in the United States. However, as a population-based cancer registry, CSP description of disease treatment is not exact. For example, it does not differentiate chemotherapy regimen, extent of surgery resection, types of salvage therapy, baseline performance status, etc. In addition, the data collection in the present study involves patients diagnosed between 1990 and 2000; the vast majority of these patients received all of their therapy prior to 2004 (when the data were abstracted). This is before temozolomide (Temodar [TMZ], Merck Pharmaceutical, Whitehouse Station, NJ), together with concurrent radiotherapy, became the standard of care for patients with newly diagnosed glioblastoma [7]. Therefore, this study does not reflect the impact of newer treatment, similar to other large comprehensive datasets [20]. This study suggests that conditional survival probability may be helpful to predict prognosis for patients with not only high-grade glioma, but other diseases with changing hazards over time. In summary, estimated by conditional survival method, patients diagnosed with high-grade glioma appear to have a large gain in their conditional survival probability over time.