Development and validation of two nomograms for predicting overall survival and Cancer-specific survival in prostate cancer patients with bone metastases: a population-based study

Abstract

Background

Prostate cancer with bone metastasis has significant invasiveness and markedly poorer prognosis. The purpose of this study is to establish two nomograms for predicting the overall survival (OS) and cancer-specific survival (CSS) of prostate cancer patients with bone metastasis.

Methods

From January 2000 to December 2018, a total of 2683 prostate adenocarcinoma with bone metastasis patients were identified from the Surveillance, Epidemiology, and End Results Program (SEER) database. These patients were then divided into a training cohort and a validation cohort, with OS and CSS as the study endpoints. Correlation analyses were employed to assess the relationship between variables. Univariate and multivariate Cox analyses were utilized to ascertain the independent prognostic factors. Calibration curves and the area under the time-dependent receiver operating characteristic curve (time-dependent AUC) were employed to evaluate discrimination and calibration of the nomogram. DCA was applied to examine accuracy and clinical benefits. The clinical utility of the nomogram and the AJCC Stage System was compared using net reclassification improvement (NRI) and integrated discrimination improvement (IDI). Lastly, the risk stratifications of the nomogram and the AJCC Stage System were compared.

Results

There was no collinearity among the variables that were screened. The results of multivariate Cox regression analysis showed that seven variables (age, surgery, brain metastasis, liver metastasis, lung metastasis, Gleason score, marital status) and six variables (age, surgery, lung metastasis, liver metastasis, Gleason score, marital status) were identified to establish the nomogram for OS and CSS, respectively. The calibration curves, time-dependent AUC curves, and DCA revealed that both nomograms had pleasant predictive power. Furthermore, NRI and IDI confirmed that the nomogram outperformed the AJCC Stage System.

Conclusion

Both nomograms had satisfactory accuracy and were validated to assist clinicians in evaluating the prognosis of PABM patients.

Introduction

Prostate cancer is the most common malignancy of the male reproductive system and the second leading cause of cancer-related mortality worldwide, with adenocarcinoma accounting for 90% of cases [1,2,3]. Although the use of PSA testing for early detection has improved survival rates, many patients still cannot undergo such testing and are diagnosed in advanced stages [4]. Over 90% of advanced prostate cancer patients develop bone metastasis (BM) [5], shortening median survival time to approximately 1.5–2 years [6]. Prostate cancer often lacks early symptoms, leading to detection in middle and late stages with multiple organ metastases [7]. The bone is the most common site of metastasis, with 10% of new cases diagnosed with BM, increasing to 80% in advanced stages [8]. Patients with BM face severe economic burden, increased mortality risk, and complications such as bone pain, spinal cord compression, and pathological fractures [9].

Bone metastatic prostate cancer leads to a significantly worsened prognosis, and the efficacy of different treatments varies [10], making personalized prediction for patients with prostate adenocarcinoma bone metastasis (PABM) a major focus of research [11]. However, the traditional TNM staging system does not fully reflect the biological behavior of the tumor and the patient’s prognosis, and the staging and grading criteria are relatively isolated, increasing the difficulty for surgeons to evaluate patient prognosis. Moreover, the staging system does not provide clear guidance on treatment strategies for PABM patients.

Nomograms have gained prominence in the field of oncology due to their proven ability to improve predictive accuracy. In recent years, these intuitive and practical models have found widespread application, particularly in the context of personalized medicine, providing a visually intuitive representation of linear prognosis and quantification of individual patient survival [12, 13].In this particular research endeavor, our primary objective is to utilize the extensive dataset available within the Surveillance, Epidemiology, and End Results (SEER) database, specifically focusing on patients with primary atypical meningioma of the brain (PABM), to develop a highly detailed nomogram aimed at predicting the prognosis of PABM patients.

Method

Identification of patient population

The data for this study were obtained from the Surveillance, Epidemiology, and End Results (SEER) database, which covers approximately 28% of the U.S. population. The SEER database released its data in April 2022, based on cancer incidence in 18 registries across the United States between 2000 and 2018, submitted in Nov 2021. Data abstraction was performed using the SEER*Stat software version 8.4.0. This study included all patients diagnosed with bone metastasis in prostate adenocarcinoma between 2010 and 2018, and the pathologic tumor stage was recorded according to the American Joint Committee on Cancer (AJCC) 7th TNM staging system for prostate cancer. The inclusion criteria: (1) patients diagnosed with prostate adenocarcinoma (PA) (Site recode of ICD-O-3/WHO2008:C61.9 and Histologic Type ICD-O-3:8140); (2) bone metastases (SEER Combined Mets at DX-bone); (3) basic demographic variables, including age, race and gender; (4) complete survival data, follow-up data and specific causes of death; (5) tumor characteristics, including histological information and type, TNM stage; (6) therapeutic measures that whether received surgery, chemotherapy and radiotherapy; (7) known PSA values, Gleason scores at diagnosis; and (8) known metastasis status. The exclusion criteria were as follows: (1) patients were diagnosed only by autopsy or death certificate; (2) patients diagnosed without histological confirmation; (3) patients had more than one primary tumor; (4) patients’ follow-up information was incomplete; (5) patients’ baseline demographic data were unknown or incomplete; (6) patients’ clinicopathological parameter data were unknown or incomplete. Figure 1 provides a detailed illustration of the process for selecting eligible study participants.

Fig. 1

Flow diagram illustrating recruitment of patients

Definition of outcomes

In our study, we wanted to predict 3-year and 5-year overall survival and cancer-specific survival (CSS) in patients with bone metastases from prostate cancer using a column chart that we created using clinical and pathological information. Cancer-specific survival (CSS) is a measure of the likelihood of survival from a specific type of cancer. It is calculated by excluding deaths from other causes and only counting deaths from cancer [14].

Identification potential covariates

Based on the published literature and the variables included in the SEER database that were applicable to this study, we included in this study lung, liver, and brain metastatic PSA levels and whether or not they were treated with chemotherapy or radiotherapy and surgery, patient age and ethnicity, and marital status, as well as pathological T or N stage and Gleason score. To ensure the reliability and interpretability of the nomogram we aimed to develop, we conducted Spearman’s correlation analysis to examine potential correlations between these selected variables.

Statistical analysis

We utilized R software (version 4.1.0) for all statistical analyses and developed three models: the Cox-AJCC model, the multifactor Cox model, and the competitive risk model. First, raw data were preprocessed and transformed into factors for subsequent analysis. Pearson correlation analysis assessed potential correlations among variables. The data were then randomly split into training and validation cohorts at a 7:3 ratio. Univariate Cox analysis identified independent variables using P values (P < 0.1).

To compare significant factors and calculate hazard ratios (HRs) and 95% confidence intervals with a standard of P < 0.05, multivariate Cox regression analysis was conducted on variables with differences in univariate Cox regression analysis. Next, a nomogram predicting 3-year, and 5-year overall survival (OS) and cancer-specific survival (CSS) rates were developed using Cox regression based on the training cohort. To evaluate the clinical benefits and utility of the Cox-AJCC model and the multifactor Cox model, and to select the optimal predictive model, the net reclassification index (NRI) and integrated discrimination improvement (IDI) methods were employed. These two complementary validation methods serve different purposes: NRI primarily compares the predictive ability of the old model with the new one by considering improvement at a specific cutoff point, while IDI primarily investigates the overall improvement of the model by examining the model’s overall enhanced performance [15]. For CSS, a competing risk nomogram was created using the Fine-Gray proportional hazards model.

The discriminative performance of the nomograms was assessed through the computation of the area under the curve (AUC) values, which provide a comprehensive evaluation across various thresholds [16]. Furthermore, the predictive accuracy of our nomograms was examined in both the testing and validation cohorts. This evaluation included the assessment of calibration using calibration curves, the utilization of receiver operating characteristic (ROC) analysis, and the application of decision curve analysis (DCA) [17, 18].we also employed Fine-Gray analysis to explore the competing risks associated with our study outcomes, and the cumulative incidence function (CIF) was used to assess the probability of a specific event at a given time point.

Result

Demographic characteristics

From the SEER database spanning 2010–2018, we identified 2683 patients suffering from PABM. For OS analysis, they were randomly divided into training (n = 1879) and validation groups (n = 804). Demographic and clinicopathologic characteristics of all patients are summarized in Table 1, revealing no statistically significant differences between the training and validation cohorts in OS analysis. The majority were under 70 years of age (56.5%), white (77.3%), and married (64.5%). Most had PSA > 20 ng/mL (70.3%) and a Gleason score of 9 (58.0%). Predominant T and N stages were T1 (35.1%), T2 (36.1%), and N0 (68.1%), respectively. The majority of patients had no distant metastases, 97.7% received radiation therapy, and rarely opted for chemotherapy (16.5%) or surgery (11.7%).

Table 1 Demographic and clinical characteristics of patients with PABM in OS group

The analysis of CSS included 1120 patients, with 793 in the training cohort and 562 in the validation cohort. Although the majority of CSS patients were diagnosed between 2010 and 2014 (70.5%), other demographic characteristics mirrored those in the OS cohort. Table 2 provides a detailed summary of baseline clinical-pathological characteristics. Subgroup analysis of Cumulative Incidence Subgroup analyses of cumulative incidence function (CIF) data showed a higher incidence of CSS in American Indian/Alaska Native patients (Fig. 2B), patients with advanced Gleason scores, and patients with stages T and N (Fig. 2C,D,E). The incidence of CSS was elevated in patients who received chemotherapy (Fig. 2G), lacked radiotherapy (Fig. 2F), and had organ metastases (Fig. 2I, J, K). Surgical intervention significantly increased CIF in PABM patients (Fig. 2H). Interestingly, marriage significantly decreased CIF (Fig. 2M), and both high and low PSA levels were associated with increased CIF (Fig. 2L). Patients over 70 years of age also exhibited higher CIF, but the results were not significant (Fig. 2A).

Table 2 Demographic and clinical characteristics of patients with PABM in CSS group

Fig. 2

Cumulative incidence predictions of CSS in gastric cancer with liver metastasis. (A) Age (B) Race (C) Gleason Score (D) T Stage (E) N Stage (F) Radiotherapy (G) Chemotherapy (H) Surgery (I) brain (J) liver (K) lung (L) PSA (M) Marital Status

Correlations among variables

Before performing Cox regression analysis, we first checked for covariance between the variables of interest using Spearman’s correlation analysis. The results are shown in Fig. 3. We found that T staging and N staging had the strongest correlation of all variables, with a correlation coefficient of 0.28. We also observed a relatively significant positive correlation between chemotherapy and age, with a correlation coefficient of 0.18. In addition, both lung and liver metastases showed a correlation coefficient of 0.17 with PSA levels and marital status, respectively. This is closely followed by Gleason score and N staging and PSA levels, with correlation coefficients of 0.16 and 0.15 respectively. Of note, there was a mild negative association between age and marital status, with a correlation coefficient of − 0.12.Nevertheless, most of the variables showed relatively low correlations, with values ranging from − 0.1 to 0.1, suggesting that there was no significant correlation for the included variables.

Fig. 3

The results of correlation analysis between all included variables. CT: chemotherapy; RT: radiotherapy

Evaluation of nomogram variables

Cox regression analysis revealed a model including factors such as age, surgical intervention, brain, liver and lung metastases, Gleason score and marital status that had the most significant P value in the training set. In total, seven variables showed a significant correlation with overall survival (OS) in univariate analysis. Further multivariate Cox regression analysis revealed that factors such as age ≥ 70 years (HR = 1.404, 95% CI = 1.233–1.598, P < 0.001), prostatectomy (HR = 0.356, 95% CI = 0.180–0.703, P = 0. 003), having a Gleason score ≥ 9 (HR = 2.774, 95% CI = 1.629–4.723, P < 0.001) and being married (HR = 1.390, 95% CI = 1.218–1.587, P < 0.001) emerged as independent prognostic factors for PABM patients. Table 3 provides a more complete breakdown of these details.

Table 3 Univariate and multivariate Cox proportional hazards regression analysis of patients with PABM in the OS group

Regarding the CSS-based classification, the specifics of the CSS group members diagnosed with PABM are described in Table 4. CSS-related prognostic factors for patients diagnosed with PABM were identified by univariate Cox regression analysis, adjusting for age, surgical procedures, liver or lung metastases, Gleason score and marital status. A subsequent multivariate Cox regression analysis was performed to identify the independent prognostic variables for PABM patients. These observations are shown in Table 4.

Table 4 Results of univariate and multivariate analyses by Fine-Gray proportional subdistribution hazards model

Nomogram development and validation

To predict 3-year and 5-year CSS and OS for patients diagnosed with PA and bone metastases, we formulated two different nomograms using independent prognostic determinants derived from multivariate Cox and CSS analyses (Fig. 4A,B). These well-established nomograms can be used to assess the potential influence of different factors that either augment or directly induce mortality in PA patients by aggregating the individual variables. The calibration curves of our nomograms showed remarkable agreement between expected and actual outcomes in both training and validation subsets (Fig. 5).

Fig. 4

Constructed nomograms for prognostic prediction of overall survival and cancer specific survival

Fig. 5

Calibration curves. A 3-year and 5-year likelihoods of OS and CSS in the training dataset. B 3-year and 5-year likelihoods of OS and CSS in the validation dataset

The AUC curves for the Cox and Competing risk models by time period were presented in Fig. 6A and B, respectively. Based on our analysis, the nomogram demonstrated good discriminatory power in predicting OS and CSS within 5 years, as shown by AUC values greater than 0.65. Specifically, in the training cohort, the Cox model exhibited AUC values of 0.666 and 0.701 at 3 and 5 years, respectively (Fig. 6C). The corresponding AUC values in the validation cohort were 0.676 and 0.690 (Fig. 6D). Similarly, the competing risk model showed AUC values of 0.661 and 0.702 at 3 and 5 years, respectively, and the corresponding values in the validation group were 0.681 and 0.724 (Fig. 6E,F).

Fig. 6

Time-dependent AUC and receiver operating characteristic (ROC) curves of OS and CSS. A,B Time-dependent AUC of using the nomogram to OS and CSS probability within 5 years in the training cohort and validation cohort. The blue line represents AUC = 0.65. And the shading area between blue dotted curves represents 95% credible intervals. C,D ROC curves corresponding to 3-year and 5-year OS in the training and validation cohort, respectively. E,F ROC curves corresponding to 3-year and 5-year CSS in the training and validation cohort, respectively

Decision curve analyses were performed on both the training and validation samples and the results are shown in Fig. 7. The decision curves for 3-year and 5-year survival showed optimal net benefits, highlighting the improved predictive accuracy of the established nomogram.

Fig. 7

Decision curve analysis of the nomogram in the estimation of OS and CSS of patients with PABM. A Training cohort. B Validation cohort. The None line illustrates the net benefit when assuming no patients die; the All line represents the net benefit if all patients die

Comparison of clinical performance between the nomograms and the AJCC staging system

To further explore the clinical applicability of the nomograms proposed here, we used the NRI, IDI and C-index metrics for a comparative assessment of the precision of the nomogram against the AJCC staging methodology. Within the training subgroup, our nomogram registered a C-index of 0.6333 (95% CI = 0.5888–0.6390) in contrast to the AJCC system, which registered 0.5413 (95% CI = 0.5167–0.5676). Additionally, the NRI values for 3-year and 5-year OS were 0.3697 (95% CI = 0.2597–0.5146) and 0.4465 (95% CI = 0.2952–0.5801), respectively. Meanwhile, the IDI values for 3-year and 5-year OS were 0.05 (95% CI = 0.030–0.074, P < 0.001) and 0.053 (95% CI = 0.030–0.080, P < 0.001), respectively (as shown in Table 5). Analogous results were observed in the validation subset (as shown in Table 5), reinforcing the notion that our nomogram has improved prognostic ability when compared to the AJCC staging system.

Table 5 Comparison of different models for estimating the overall survival of PABM patients

Risk classification for prostate cancer patients with bone metastases

We stratified the risk based on the total scores calculated from the nomogram, which enabled the classification of each PA patient into two risk groups (low-risk and high-risk) based on the median value. Kaplan-Meier survival analysis revealed that the low-risk group had significantly better OS and CSS than the high-risk group (Fig. 8C–F). Furthermore, the nomogram exhibited excellent discriminatory power between the two risk groups, while the AJCC Stage System had limited capability to distinguish between them (Fig. 8A,B). These findings were replicated in the validation cohort.

Fig. 8

Kaplan–Meier OS and CSS curves of PABM patients with different risks stratified by the nomogram. A,B PABM patients in the training and validation cohort at different stages are classified according to the AJCC staging system. C,D PABM patients in the training and validation cohort at different stages are classified according to the cox model nomogram. E,F PABM patients in the training and validation cohort at different stages are classified according to the competing risk model nomogram

Discussion

Prostate cancer is one of the most common cancers in men, and its incidence is increasing every year [19]. Bones are the most common site of distant metastasis, and elderly men are often diagnosed with advanced prostate cancer due to bone metastases, which often indicate a poor prognosis [20, 21]. Adenocarcinoma is the most common pathological type of prostate cancer, accounting for over 90%, therefore, in this study, we chose adenocarcinoma as the subject of our research. Early detection and limited treatment options for PABM require urgent solutions. In our study, we constructed two predictive models to predict the prognosis of PABM patients. The validation of our nomograms revealed that they exhibit strong discriminative performance and calibration. Furthermore, our risk stratification approach successfully categorized PABM patients into high- and low-risk groups, demonstrating a significant difference between the two groups.

It has been reported that several factors affect the survival of patients with prostate adenocarcinoma bone metastasis (PABM), including age, Gleason score, PSA level, clinical stage, and treatment modality [22, 23]. Therefore, we attempted to incorporate as much information as possible from the SEER database into Cox and competing risk models to predict the prognosis of PABM patients. According to Zhang, Z. et al. study [24], age significantly affects the prognosis of bone-metastatic prostate cancer patients. In our study, we also found that advanced age was closely associated with a poorer prognosis in PABM patients, as demonstrated by both the CSS and OS models. The Gleason score is an important indicator for evaluating the degree of differentiation of prostate cancer. A higher score often indicates a lower degree of tumor differentiation, stronger invasiveness, and a poorer prognosis [22, 25]. In addition to the Gleason score, PSA value and clinical T/N stage are generally considered strong prognostic factors for PCa and are often combined to assess the risk of death in PCa patients with bone metastasis [26]. Our predictive model achieved results that are generally consistent with previous studies. Furthermore, with regard to PABM patients should undergo surgery, and what type of surgery to choose, has always been a controversial topic in metastatic prostate cancer. We have found that for PABM patients, aggressive surgical treatment can actually shorten their survival time. However, in the study by Qin XJ et al. [27], tumor cytoreduction surgery such as TURP can reduce PSA levels during follow-up, reduce tumor burden, alleviate urinary obstruction and pain, thus relieving disease-related suffering. However, surgical complications may worsen the prognosis, so the patient’s overall condition should be thoroughly evaluated before surgery to truly improve the patient’s condition [28].

Chemotherapy has been proven to be important in the treatment of bone metastatic prostate cancer, according to several clinical trials [29, 30]. Docetaxel is one of the most commonly used chemotherapeutic drugs for bone metastatic prostate cancer, which can prolong the survival of patients [29]. However, our competing risk model showed that chemotherapy increased the risk of death in patients with prostate cancer and bone metastasis. This may be closely related to the incomplete information about chemotherapy in the SEER database. In fact, if patients are fit enough, chemotherapy is recommended along with castration in newly diagnosed M1 phase [30].

Radiotherapy is also an important treatment for prostate cancer and plays an important role in metastatic disease. In our study, radiotherapy significantly reduced the Cumulative incidence predictions of CSS in PABM patients, thereby prolonging the tumor-specific survival time. Similarly, the SEER database’s limitations in documenting radiotherapy information may affect the precision of our conclusions. Parker CC’s three-phase clinical study confirmed that radiotherapy can significantly improve the survival rate of low metastatic load PC patients, but the radiation dose still needs further research to clarify and formulate standardized radiotherapy guidelines [31].

In addition to the aforementioned factors, marital status has been recognized as potential predictors of survival in prostate cancer patients [32, 33]. The effect of marital status on survival has also been previously reported [34]. Our findings are consistent with this, indicating that married patients tend to have a better survival advantage than single patients [35], likely due to the provision of explicit treatment and emotional support [36]. This was consistent with our study. Consistent with prior research findings [37], our study reveals that unmarried PCa patients are more likely to develop progressive disease. This increased risk could potentially be linked to changes in male hormones following marriage, as it is widely recognized that androgens play a critical role in the development of PCa [38].

Besides bone metastasis, we also investigated liver, brain, and lung metastasis, even though patients diagnosed with these distant sites accounted for only a small portion of the SEER database. In this study, we found that lung metastasis (5.59%) was the most common distant metastatic site, followed by liver (2.98%) and brain (1.01%), and the probability of combined metastases was consistent with previous reports. When distant metastasis was detected, PCa cells had already spread to multiple organs, which is consistent with the results of several previous studies [39, 40]. This also indicates that patients with liver, brain, or lung metastases were more likely to have bone metastasis, and multiple site metastases together influenced the prognosis of PABM patients. In our CSS model, lung, liver, and brain metastases significantly increased the CIF value, resulting in an increased risk of competing death events for PABM patients.

Despite the valuable insights provided, our study has some limitations. Firstly, being a retrospective study that utilizes the SEER database, it may be susceptible to inherent biases such as selection bias and limited generalizability to Asian patients due to the predominantly white population in the database. Furthermore, the lack of information on systemic treatment, specific chemotherapy regimens, radiation dose, endocrine therapy, and genetic markers may impact the accuracy of our models and necessitates further research. Secondly, our study only accounts for BM at the time of initial diagnosis, failing to account for the possibility of BM occurring later in the disease course. Lastly, our nomograms may be subject to overfit bias due to the exclusive use of the SEER dataset for both training and validation cohorts, highlighting the need for external validation with diverse clinicopathological data. To address these limitations, future studies should consider well-designed prospective analyses that incorporate more comprehensive data, including genetic markers, and construct predictive models that encompass both clinicopathological information and genetic markers.

Conclusions

We’ve created a precise predictive model for prostate cancer patients with bone metastasis, using common clinical indicators to accurately address patient inquiries. Our model empowers clinicians and patients with clearer prognoses, enhancing treatment decision-making. We also developed prognostic tools and a risk system for prostate adenocarcinoma bone metastasis, laying the groundwork for personalized treatment. However, further research is needed to optimize treatment for these patients.