Fax: (214) 648-8786
Prognostic model of event-free survival for patients with androgen-independent prostate carcinoma
Article first published online: 20 APR 2005
Copyright © 2005 American Cancer Society
Volume 103, Issue 11, pages 2280–2286, 1 June 2005
How to Cite
Shulman, M. J. and Benaim, E. A. (2005), Prognostic model of event-free survival for patients with androgen-independent prostate carcinoma. Cancer, 103: 2280–2286. doi: 10.1002/cncr.21054
- Issue published online: 18 MAY 2005
- Article first published online: 20 APR 2005
- Manuscript Accepted: 21 JAN 2005
- Manuscript Revised: 11 JAN 2005
- Manuscript Received: 29 FEB 2004
- androgen independent;
- prostate carcinoma;
- clinical complications;
- prognostic model
The current study was conducted to develop a prognostic model of event-free survival (EFS) in men with androgen-independent prostate carcinoma (AIPC).
Data from 160 patients diagnosed with AIPC between 1989–2002 were reviewed. No patient had received cytotoxic chemotherapy. A univariate Cox proportional hazards model identified significant predictors of EFS. Recursive partitioning analysis divided these significant variables into prognostic risk groups. The final prognostic model was tested with a Cox proportional hazards model.
The final prognostic risk model included the presence of metastatic disease at the time of androgen-independent disease progression (P = 0.040), time to prostate-specific antigen (PSA) recurrence (P = 0.043), and PSA doubling time (P < 0.01). Three highly independent risk groups were identified. The observed median EFSs were 6.1 months (95% confidence interval [95= CI], 3.4–8.8 months), 33.6 months (95= CI, 25.3–41.9 months), and 96.1 months (95= CI, 57.9–134.3 months) for the low-risk, intermediate-risk, and high-risk groups, respectively. Each risk group was found to be independently predictive of EFS (P < 0.01). Patients who died of prostate carcinoma experienced significantly more clinical events than those who died of other causes (P < 0.01).
The prognostic model in the current study stratified patients into three highly significant and independent risk groups for EFS. A detailed PSA history and knowledge of metastatic disease are sufficient to risk-stratify patients with AIPC. One very unique aspect of this model was that it was developed from a patient cohort that never received chemotherapy. Cancer 2005. © 2005 American Cancer Society.
Prostate carcinoma remains the second leading cause of cancer-related deaths in the U.S.1 The current estimated lifetime risk of developing prostate carcinoma for patients of all ages is 17.3%.1 However, the lifetime risk of dying from the disease has been estimated to be 3%, indicating that a significant proportion of men with prostate carcinoma die from other unrelated causes.1 Nevertheless, prostate carcinoma can also be a source of significant morbidity. Most patients with advanced prostate carcinoma are at risk of developing severe and disabling disease-related complications. Some of the most common adverse events associated with advanced prostate carcinoma are bone pain, pathologic fractures, spinal cord compression, and local and/or regional urinary tract obstruction. These complications can cause formidable pain and suffering and most likely result in significant healthcare expenditures. Furthermore, in the past 20 years, death rates from cardiovascular and other ailments have steadily declined; patients are living longer and therefore are more likely to experience one or more of these prostate carcinoma-related complications.
To our knowledge, as a separate outcome, clinical progression in patients with advanced prostate carcinoma has not received much attention. In most clinical trials, it is examined as a secondary endpoint and only rarely has the frequency of the various complications been analyzed in any significant detail.2–4 In an era in which disease progression can be detected long before clinical progression ensues, information regarding the rate and probability of experiencing one or more of these disease-related complications should be helpful. As new therapeutic strategies are developed and tested, information concerning the natural history of disease progression may aid the clinician in determining the need for and timing of therapy in the context of the patient's age, associated comorbidities, and probabilities of progression. In addition, the efficacy of novel therapies could be better judged by designing trials that consider the individual's risk for clinical progression.
In the current study, we examined the frequency and probabilities of developing one or more of these events in a cohort of patients with prostate carcinoma, all of whom had biochemical evidence of disease progression despite androgen-deprivation therapy (ADT).
MATERIALS AND METHODS
Available for retrospective review were the medical records of 375 patients receiving ADT for advanced prostate carcinoma at the Dallas Veterans Affairs Hospital (DVAH) between 1977 and 2002. Indications for ADT included: 1) primary therapy for locally advanced and/or metastatic disease at the time of presentation, 2) prostate-specific antigen (PSA) and/or clinical progression after local therapy, and 3) an increase in PSA occurring during watchful waiting. Patients were followed at the DVAH every 3 months with a PSA level. Radionuclide bone imaging, plain film radiography, and computed tomography (CT) scans were used to identify bone metastases.
Androgen-independent progression (AIP) in men undergoing ADT for prostate carcinoma was defined as two consecutive increases in PSA above the nadir value, each of which was greater than 25% of the nadir value. The main endpoint was event-free survival (EFS). Survival duration was defined as the time from the first PSA increase during ADT to the first clinical event. Clinical events were divided into three categories: bone, locoregional, and neurologic events. All clinical events were secondary to prostate carcinoma. A bone event was defined as new bone pain, an increasing narcotic requirement for bone pain, palliative external beam radiation therapy (EBRT) or strontium therapy, or pathologic fracture. A locoregional event was defined as acute urinary retention (AUR), clot retention from macroscopic (gross) hematuria (GH), nephrostomy tube placement to relieve ureteral obstruction, transurethral resection of the prostate (TURP) for bladder outlet obstruction or refractory GH, or postobstructive renal failure. A neurologic event was defined as acute spinal cord compression or the need for palliative EBRT.
PSA doubling time (PSADT) was calculated using the following equation5: PSADT = [log (2) × t] ÷ [log (final PSA) – log (initial PSA)], in which “log” is the natural logarithm function and “t” is the time from the initial to the final PSA level. The PSA level measured on the date of AIP was defined as the initial PSA. The final PSA value was the level measured at the time of the last follow-up.
A univariate Cox proportional hazards model was used to identify all statistically significant predictors of EFS. P values < 0.05 were defined as being statistically significant. Only those variables found to be significant on univariate analysis were included in a multivariate proportional hazards model. However, the multivariate model had only 31 events, which limited its prognostic power. Therefore, a recursive partitioning (RP) analysis was performed to identify significant prognostic groupings of the variables that were significant on univariate analysis. The best split at each stage was selected using a log-rank statistic for EFS. The final RP model was tested with a Cox proportional hazards model. Actuarial survival analysis was performed using the Kaplan–Meier method. The chi-square test, Mann–Whitney U test, and student t test (two-tailed) were used when appropriate to identify statistically significant differences between the groups.
The Institutional Review Board at the DVAH approved the current study.
One hundred seventy patients were diagnosed with androgen-independent prostate carcinoma (AIPC) between 1989–2002. Three patients received cytotoxic chemotherapy and seven patients had incomplete medical records. After excluding these 10 patients, 160 men with AIPC remained to constitute the basis of this study. Table 1 displays patient, disease, and treatment characteristics. The median time from the initiation of ADT to AIP was 18.5 months (range, 0.6–74.9 months). Nineteen patients without metastatic disease at the beginning of ADT demonstrated radiographic evidence of new bone metastases at the time of AIP. The median PSADT after AIP was 6.5 months (range, 0.7–90.0 months). The mean patient follow-up from the time of AIP was 33.8 months (95% confidence interval [95% CI], 29.7–37.8 months). Seventy-nine patients died during follow-up, 58 of whom (73.4%) died of prostate carcinoma. The median overall survival was 46.1 months (95% CI, 34.7–57.6 months). The median disease-specific survival was 67.6 months (95% CI, 51.0–84.3 months). At the time of the initiation of ADT, 56 patients (35.0%) were asymptomatic, 14 patients (8.8%) had bone pain, 19 patients (11.9%) had symptoms from locally advanced disease, and the symptom status of 71 patients (44.3%) was unknown. Of the 160 patients diagnosed with AIPC, 78 patients (48.8%) did not experience any disease-related clinical events during follow-up. Of the 82 patients who experienced ≥ 1 adverse disease-related clinical event (51.3%), the median EFS was 39.9 months (95% CI, 29.9–50.0 months). There was no difference with regard to symptom-free survival between white and black patients (P = 0.9541, log-rank test). Table 2 summarizes the total number and type of bone, locoregional, and neurologic events documented during the follow-up period. The mean number of bone, locoregional, and neurologic events per patient were 0.8 (95% CI, 0.6–1.0 events), 0.5 (95% CI, 0.3–0.6 events), and 0.10 (95% CI, 0.04–0.16 events), respectively. Those patients who died of prostate carcinoma experienced a significantly higher absolute number of clinical events compared with those patients who died of other causes (Table 3). However, a significant proportion of patients (21.6%) who died from other causes experienced complications that were directly related to prostate carcinoma progression. Moreover, when stratified by clinical event category, patients who died of prostate carcinoma were found to experience a significantly greater number of mean bone and locoregional events during the follow-up period (Table 4). The difference in the number of neurologic events noted between these two groups approached statistical significance.
|Median age at diagnosis, (yrs) (range)||69.9 (50.2–87.2)|
|Race: % white, % black, and % Hispanic||55.1, 41.5, and 3.4|
|Median PSA, (ng/mL), (range)|
|At diagnosis||54.0 (2.4–8439.0)|
|At initiation of ADT||56.0 (1.8–8439.0)|
|Nadir while receiving ADT||0.6 (< 0.2–103)|
|Gleason score, no. (%)|
|Clinical T classification at diagnosis, no. (%)|
|N classification at diagnosis, no. (%)|
|M classification at diagnosis, no. (%)|
|M classification at the time of AIP, no. (%)|
|Initial management of prostate carcinoma|
|Primary ADT||87 (54.4)|
|Watchful waiting||12 (7.5)|
|Type of ADT|
|LHRH* agonist monotherapy||109 (68.2)|
|Combined androgen blockade||37 (23.1)|
|Type and no. of clinical events per patient||No. of patients (%)|
|Bone pain events|
|Increasing narcotic requirement|
|Palliative EBRT or strontium therapy|
|Acute urinary retention|
|Nephrostomy tube placement|
|≥ 2||1 (0.6)|
|≥ 2||1 (0.6)|
|Postobstructive renal failure|
|Spinal cord compression|
|EBRT for neurologic compromise|
|Total no. of clinical events||Cause of death, no. of patients (%)||P value|
|Prostate ca||Other cause|
|0||7 (9.5)||11 (14.9)||0.002|
|1||10 (13.5)||3 (4.1)|
|2||9 (12.2)||1 (1.3)|
|≥ 3||32 (43.2)||1 (1.3)|
|Variable||Cause of death, no. of patients (%)||P value|
|Prostate ca||Other cause|
|Mean no. of clinical events per patient||2.8||0.6||<0.01|
|Mean no. of bone events per patient||1.6||0.2||<0.01|
|Mean no. of loco regional events per patient||0.9||0.3||0.009|
|Mean no. of neurologic events per patient||0.3||0.1||0.058|
The following variables were tested in the univariate Cox proportional hazards analysis for EFS: age at diagnosis, race, PSA at diagnosis, biopsy Gleason score, percentage of Gleason score 4 or 5 biopsy specimens, clinical TNM stage, type of initial prostate carcinoma management, nadir PSA level after radical retropubic prostatectomy (RRP) or EBRT, indication for ADT, type of ADT administered, PSA at the initiation of ADT, the presence of disease-related symptoms at the initiation of ADT, the presence of metastatic disease at the initiation of ADT, age at the time of AIP, the nadir PSA level during ADT, the time from the initiation of ADT to AIP (hereafter referred to as time to PSA recurrence), the presence of metastatic disease at the time of AIP, and PSADT. The results of this univariate analysis are summarized in Table 5. All significant variables regarding univariate analysis were then entered into a RP analysis to identify statistically significant prognostic risk groupings. Of the 160 patients entered into the RP analysis, 4 patients had missing PSADT values, 34 patients had missing metastatic disease at time of AIP data, and 3 patients had missing time to PSA recurrence data. These 41 patients were excluded from RP analysis, leaving 119 patients to be dividing among the final RP prognostic groups. The most significant prognostic splits, in order from the highest to lowest difference in actuarial EFS, were PSADT, the presence of metastatic disease at the time of AIP, and time to PSA recurrence. Four terminal nodes were identified, each of which could not be further split into two statistically different subgroups (Fig. 1). Terminal nodes I and III were not found to be statistically different (P = 0.132), and were combined. The final prognostic model therefore included three risk groups. The low-risk group (terminal node IV) was defined only by a PSADT > 10 months. The intermediate-risk group was defined by a PSADT ≤ 10 months and a time to PSA recurrence > 13 months (terminal nodes I and III). The presence or absence of metastatic disease at the time of AIP became insignificant in this risk group when terminal nodes I and III were combined. Finally, the high-risk group included patients with a PSADT ≤ 10, measurable metastatic disease at the time of AIP, and a time to PSA recurrence ≤ 13 months (terminal node II). The low-risk group (hazard ratio [HR] referent [P < 0.01]), intermediate-risk group (HR of 3.6, 95% CI, 1.9–6.9 [P < 0.01]), and high-risk group (HR of 9.6, 95% CI, 9.6–63.8 [P < 0.01]) each were found to be independently predictive of EFS on the Cox proportional hazards analysis. EFS curves for the final prognostic model are shown in Figure 2. The median EFSs were 6.1 months (95% CI, 3.4–8.8 months), 33.6 months (95% CI, 25.3–41.9 months), and 96.1 months (95% CI, 57.9–134.3 months) for the low-risk, intermediate-risk, and high-risk groups, respectively.
|Variable||P value||Hazards ratio||95% CI|
|Cancer-related symptoms at the initiation of ADT||0.006||2.398||1.282–4.484|
|Nadir PSA on ADT||<0.01||1.039||1.019–1.059|
|Time to PSA recurrence||0.043||0.979||0.958–0.999|
|Metastatic disease at the time of AIP||0.040||0.826||1.029–3.242|
The mean number of bone events per patient during the follow-up period was 0.3, 0.7, and 1.4 for the low-risk, intermediate-risk, and high-risk groups, respectively (P < 0.019). There were no statistically significant differences noted with regard to the number of locoregional and neurologic complications by risk groups.
In the current study, we report what to our knowledge is the first prognostic model for the development of prostate carcinoma-related complications in men with AIPC. Prior to this analysis, patients with AIPC were considered to be homogeneous with respect to their risk of clinical progression. However, the results of the current study clearly demonstrated that the risk of clinical progression in AIPC is not uniform and can be predicted from readily available clinical variables. Of these variables, PSADT proved to be the most significant predictor of EFS in men with AIPC. An optimal PSADT split of 10 months was found to result in the largest difference in EFS. In fact, the subgroup of men with AIPC and a PSADT > 10 months had no other variables that were found to be predictive of EFS. This low-risk group had a remarkable median EFS of 96.1 months and experienced an initial symptom-free period that was 1.9 and 8.2 times longer than the intermediate-risk and high-risk groups, respectively (P < 0.01). Patients in the intermediate-risk group also enjoyed a prolonged EFS. They had a median EFS of 33.6 months and experienced an event-free interval that was 4.4 times longer than that of the high-risk group (P < 0.01). Lastly, patients in the high-risk group were reported to have a median EFS of 6.1 months and were characterized by a rapid PSADT, metastatic disease, and a brief time to biochemical failure while receiving ADT. Each risk group was found to be independently predictive of EFS.
PSADT is emerging as a surrogate marker for prostate carcinoma progression and survival after the treatment of both clinically localized and advanced prostate carcinoma. PSADT has been shown to be a significant predictor of metastasis-free survival in men with an isolated PSA recurrence after radical prostatectomy.6, 7 In addition, PSADT may serve as a possible surrogate for prostate carcinoma-specific survival after EBRT for localized disease.8 PSADT in AIPC may be an equally valuable marker of treatment response, disease progression, and survival.9–12 Akimoto et al. showed that PSADT was the most statistically significant predictor of cause-specific survival in 56 patients with AIPC.11 The data from the current study add to the growing body of evidence that PSADT may be a valuable surrogate marker for the risk of disease progression in men with AIPC.
Overall, patients who died of prostate carcinoma experienced 4.7 times the number of clinical events than those who died of other causes (P < 0.01). Of all clinical events, bone pain was the most common disease-specific complication in the patient population in the current study. In fact, bone pain is widely recognized to be the most common symptom in patients with advanced prostate carcinoma and is a source of severe and debilitating complications.13 The mean number of bone events per patient during the follow-up period was 0.8. Patients who died of prostate carcinoma experienced bone events at a rate that was eight times greater than those who died of other causes (P < 0.01). Patients in the high-risk group experienced 2.0 and 4.6 times the total number of bone events during follow-up of patients in the intermediate-risk and low-risk groups, respectively (P = 0.019 and P = 0.005, respectively). Approximately 18.7% of the study cohort had documented bone pain from metastatic prostate carcinoma and required an increase in palliative narcotics. Approximately 8.7% of patients experienced a pathologic fracture.
Locoregional symptoms were the second most common complication of AIPC in the population in the current study. The mean number of local events was 0.46 per patient during the follow-up period. Patients who died of prostate carcinoma experienced local and/or regional events at a rate that was three times that of men who died of other causes (P = 0.009). Approximately 15% of patients had 1 or more episodes of AUR. The placement of a percutaneous nephrostomy tube, which has been associated historically with a poor outcome,14 was required to relieve urinary tract obstruction in 6.9% of our patients. Approximately 10.0% of patients in the current study required TURP for AUR or to control GH. Renal failure was reported to occur in 5.6% of patients. Lastly, the mean number of neurologic events was reported to be 0.10 per patient during the follow-up period. Spinal cord compression was reported to occur in 8 patients (5%). The incidence of locoregional and neurologic complications in the cohort of patients in the current study was found to compare favorably with reports in the literature.13, 15
The strength of our proposed prognostic model is that it identifies three distinct and independent risk groups for the clinical progression of AIPC using three readily available clinical variables. Risk grouping patients with the new diagnosis of AIPC may enable physicians to better discriminate between patients at high-risk, intermediate-risk, or low-risk of symptomatic progression. Because there currently is no cure for AIPC, this information could then be used to identify patients at higher risk and direct them toward new prospective trials of therapeutic agents designed to prevent, delay, and/or palliate symptoms as well as to improve the quality of life in men with AIPC. Numerous studies have supported the use of palliative therapies as efficacious in reducing bone pain and skeletal events, decreasing analgesic requirements, increasing time to pain progression, and improving quality of life.2–4, 16, 17 An additional appeal of the current study is that our prognostic model was developed from a cytotoxic chemotherapy-naïve patient population. Because this prognostic model was not confounded by the use of cytotoxic adjuvant therapies, it may reflect the true risk of symptomatic disease progression more accurately in patients with a new diagnosis of AIPC.
The current study was limited by its retrospective study design. As a consequence, the patient population was heterogeneous. Because clinical data were collected retrospectively, careful control and monitoring of data acquisition was not possible. However, the heterogeneity of the current study data may give our prognostic groups more strength to predict EFS in the real-world population of men with AIPC who are treated by practicing clinicians.
Not all men with AIPC in the PSA era have equal risks of disease-related symptomatic progression. A prognostic model of EFS stratified men with AIPC into three highly significant and independent risk groups. One very unique aspect of this model was that it was developed from a patient cohort that had never received chemotherapy. Therefore, this model may describe the true natural history of symptomatic progression in AIPC more accurately than previous models.