Development of a prognostic nomogram and genetic insights for prostate cancer patients with secondary primary malignancies: A SEER retrospective cohort study and Mendelian randomization analysis

Abstract

Background: Prostate cancer (PCa) patients are at risk of developing second primary malignancies (SPMs), which can significantly shorten their survival. Understanding the risk of SPMs and associated factors is crucial to the optimization of patient follow-up.

Methods: This study focuses on PCa patients who were later diagnosed with SPMs using data from the Surveillance, Epidemiology, and End Results (SEER) database. Variables were carefully selected, and the data were analyzed using machine learning techniques combined with multivariate Cox proportional hazards modeling. Subsequently, a nomogram was generated to predict the 1-, 3-, and 5-year survival rates for SPMs patients. Additionally, a two-sample Mendelian randomization (TSMR) analysis was conducted to investigate the causal relationships between PCa and its top ten SPMs.

Results: Among the variables, age, marital status, SPM site, M stage, American Joint Committee on Cancer (AJCC) stage, PCa surgery, and prostate-specific antigen (PSA) levels were identified as key prognostic factors through least absolute shrinkage and selection operator (LASSO) and backward stepwise regression. Based on these factors, a nomogram was developed to visually represent survival predictions, complemented by a web-based calculator for easy application. This nomogram, which serves as a supplement to traditional AJCC staging, demonstrated strong predictive power for 1-, 3-, and 5-year survival, with area under the curve (AUC) values exceeding 0.85. Additionally, TSMR analysis revealed a causal link between PCa and urothelial carcinoma (UC).

Conclusion: This study developed a nomogram for predicting survival in prostate cancer patients with secondary primary malignancies, enhancing prognosis accuracy. TSMR identified a causal link between PCa and UC.

Graphical abstract

Keywords

nomogram / prostate cancer (PCa) / second primary malignancies (SPMs) / SEER database / survival prediction / two-sample Mendelian randomization (TSMR)

1 INTRODUCTION

Prostate cancer (PCa) is the second most common cause of cancer-related deaths and the most often diagnosed cancer among men, with a 29.4/100 000 incidence rate and a 7.3% mortality rate[1]. The survival time of cancer patients has increased over the past decade as a result of more frequent early diagnoses and improvements in cancer therapies[2]. However, patients remain at risk of developing a second primary malignancy (SPM), which is unrelated to the spread, metastasis, or recurrence of the first cancer, following radiation therapy and curative surgery[3,4]. Approximately 11.3% of patients with PCa have been identified with SPMs[5], and in some countries, the highest incidence of SPMs was observed in patients with PCa, representing 16.9% to 22.5% of all SPMs[6]. Prior research indicated that SPMs in adolescents and young adults have a reduced survival period, with poorer rates compared to those with the same cancer type as their first primary malignancy[7]. It also suggests that understanding the risk of SPM and associated factors is crucial to optimize patient follow-up[8]. Therefore, identifying high-risk patients for developing SPMs in clinical settings could significantly enhance malignancy management strategies.

A risk-predicting model is helpful for analyzing risk factors and predicting hazards. However, it is complicated to apply it in clinical practice. Nomogram interprets the multiple independent risk factors into an intuitive graph. Due to its simplicity, intuitive nature, and practical applicability[9,10], it has gained widespread adoption for prognostication across diverse cancer types[11^-13], including models for predicting SPMs[14,15]. However, traditional observational studies are often limited by confounding factors and reverse causality[16]. To address these limitations, we introduced two-sample Mendelian randomization (TSMR) analysis. TSMR is a genetic approach that uses single-nucleotide polymorphisms (SNPs) as instrumental variables (IVs) to infer causal relationships between traits, and it has emerged as a vital tool for uncovering potential causal links in disease etiology [17].

This research aims to explore the features of SPMs and develop a nomogram to predict patients' survival with SPMs after PCa. A web-based risk prediction calculator has been developed for clinical use. Then, comparative analyses are conducted between the nomogram and traditional American Joint Committee on Cancer (AJCC) staging to validate the predictive efficacy. Additionally, we aim to employ TSMR to assess the causal relationships between PCa and its top ten SPMs. Overall, this study not only enhances risk management for SPMs that occur following a PCa diagnosis by utilizing an intuitive prognostic tool, but also employs TSMR to clarify the causal genetic links between PCa and its top 10 associated SPMs, thereby deepening our understanding of their underlying pathogenesis and advancing personalized treatment strategies for patients with PCa-related SPMs.

2 METHODS

2.1 Data source

This study utilized the Surveillance, Epidemiology, and End Results (SEER) database, utilizing SEER*Stat software (version 8.4.3; Surveillance Research Program, National Cancer Institute) for the extraction of patient data. The SEER Research Plus database covers cancer case information from 2000 to 2019 across 17 registries, including detailed demographic, diagnostic, tumor-specific, therapeutic, and survival information.

For the TSMR analysis, genome-wide association study (GWAS) summary statistics for PCa were obtained from a meta-analysis of European ancestry cohorts, comprising 10 792 cases and 410 350 controls[18]. For further GWAS details regarding the data sources, please refer to Table S1.

2.2 Inclusion and exclusion criteria

The cohort comprised individuals diagnosed with PCa between January 1, 2010, and December 31, 2018, based on specific International Classification of Diseases for Oncology, Third Edition (ICD-O-3) morphology codes (8000/3, 8010/3, 8140/3, 8255/3, 8480/3, 8481/3, and 8490/3). To enhance the analytical accuracy and clarity and to reduce potential misunderstandings, this study adopted the Warren criterion to define SPMs as neoplasms histologically distinct from the initial primary cancer (IPC), diagnosed in a separate calendar year to avoid misclassification due to metastasis or recurrence[19]. This careful selection process aimed to ensure the precise identification of both initial PCa cases and subsequent SPMs, facilitating an accurate analysis of subsequent malignancies.

From an initial data set of 382 798 PCa cases in the SEER database, 22 265 patients diagnosed with SPMs were identified. The study inclusion criteria were as follows: (1) The age of the patient must be over 18 years. (2) The tumor must be verified as malignant. (3) The patient's comprehensive survival data is available. (4) The patient's medical records must document exactly two malignancies. The exclusion criteria were as follows: (1) If the ICD-O-3 histology codes for SPMs and IPCs were identical. (2) If the records were first sourced from autopsies or death certificates. (3) If the patient lacked essential data or had a recorded survival time of zero months. After careful screening, 6363 eligible patients were randomly assigned to training (4455) and validation (1908) sets, following a 7:3 ratio. Using random splitting, significant prognostic factors were discovered in the training set and used to generate the nomogram. The validation set was employed for internal validation to confirm the predictive accuracy and efficacy of the model.

2.3 Clinical characteristics and outcome definition

Baseline characteristics, laboratory tests, and clinicopathological data were extracted, including age at diagnosis of PCa, interval between IPC and SPMs (years), marital status at diagnosis of SPMs (married, previously married, never married), race (white, black, others), SPMs site (urinary bladder, lung and bronchus, melanoma of the skin, kidney and renal pelvis, NHL—nodal, liver, thyroid, NHL—extranodal, pancreas, larynx, Others), clinical T stage of PCa (T1, T2, T3, T4), clinical N stage of PCa (N0, N1), clinical M stage of PCa (M0, M1), AJCC stage of SPMs (0a/0is, I, II, III, IV), PCa surgery (No/unknown, Yes), PCa radiation therapy (No/unknown, Yes), PCa chemotherapy (No/unknown, Yes), SPMs radiation therapy (No/unknown, Yes), SPMs chemotherapy (No/unknown, Yes), International Society of Urological Pathology grading (ISUP1 ~ 5), PSA levels (≤ 4 ng/mL, 4–10 ng/mL, 10–20 ng/mL, > 20 ng/mL), survival months, and vital status. In our study, age refers to the age of the patient at the time of their initial PCa diagnosis. The TNM staging refers specifically to the pathological stage. The ISUP grade group is based on biopsy Gleason scores. The primary endpoint was overall survival (OS), defined as the time from diagnosing SPMs to death from any cause or for survivors to the last follow-up. The final follow-up date was December 31, 2019.

2.4 Statistical analysis

2.4.1 Statistical methods and model development

Continuous and categorical variables were summarized using median with interquartile range (IQR) and frequencies (%). Differences in quantitative variables not conforming to a normal distribution between training and validation cohorts were assessed using the Mann–Whitney U test, and differences in categorical variables were evaluated using χ² tests. Variable selection employed least absolute shrinkage and selection operator (LASSO) regression with a one-standard error criterion to minimize overfitting, followed by backward stepwise regression based on the Akaike Information Criterion (AIC) for final variable inclusion in the multivariate Cox proportional hazards model. This systematic approach aimed to facilitate rigorous feature selection and avoided multicollinearity, as evidenced by variance inflation factor (VIF) assessments (VIF > 10 indicating significant multicollinearity was not observed)[20]. To preserve statistical integrity and avoid model overfitting, the covariates of the nomogram were also carefully matched with Harrell's criterion, which states that the number of outcome events should exceed the number of covariates by at least 10 times[10]. A nomogram for predicting 1-, 3-, and 5-year OS rates was developed, enabled by the Cox regression model. Furthermore, a separate web-based calculator was designed to enable easy and practical application of the nomogram in clinical settings.

2.4.2 Model validation and comparative analysis

The discriminative ability of the nomogram was comprehensively evaluated using the concordance index (C-index) and receiver operating characteristic (ROC) curves. The area under the curve (AUC) values for 1-, 3-, and 5-year survival predictions provided precise accuracy metrics for the prognostic estimates, with the time-dependent AUC for 5-year continuous survival prediction further validated by 95% confidence intervals obtained from 1000 bootstrap resamplings. Calibration plots further complemented these assessments to verify the precision of the nomogram's survival forecasts. Comparative performance analyses against the AJCC staging system included time-dependent C-index and decision curve analysis (DCA) supplemented by net reclassification improvement (NRI) and integrated discrimination improvement (IDI) metrics to quantitatively assess the improvement of the nomogram over traditional AJCC staging.

2.4.3 TSMR analysis of PCa and its top 10 SPMs

This study employed a TSMR approach to explore the causal relationship between PCa and its 10 SPMs. PCa was defined as the exposure, and the 10 SPMs as the outcome. The MR analysis was based on three assumptions: (1) the relevance assumption, requiring strong association between IVs and exposure; (2) the independence assumption, ensuring IVs are not influenced by confounders; and (3) the exclusion restriction assumption, stipulating that IVs affect the outcome only through the exposure[21]. IVs were selected based on genome-wide significance (p < 5 × 10⁻⁸) and refined using linkage disequilibrium (LD) thresholds (R² < 0.001, clumping distance = 10 000 kb). SNPs with F-statistics < 10 were excluded to minimize weak instrument bias.

\begin{array}{}& F=\frac{R^2(N-2)}{(1-R^2)},\end{array}

where N and R² are the sample size and the variance explained by IVs, respectively.

Causal effects were primarily estimated using the inverse variance weighted (IVW) method, with additional methods, including weighted median, MR-Egger, simple mode, and weighted mode, applied for robustness[17]. Heterogeneity among IVs was assessed using Cochran's Q test, with p < 0.05 indicating significant heterogeneity. If heterogeneity was present, a random-effects IVW model was used; otherwise, a fixed-effects IVW model was applied. A leave-one-out sensitivity analysis was performed to systematically exclude each SNP, ensuring that the results were not influenced by outlying IVs[17].

3 RESULTS

3.1 Patient demographics and disease characteristics

A total of 6363 patients were identified as having SPMs and were randomly divided into a training cohort and a validation cohort at a ratio of 7:3 (Figure 1). The median follow-up was 22 (IQR: 9, 45) months in the training cohort and 21 (IQR: 8, 43) months in the validation cohort. The three most prevalent sites of SPMs were the urinary bladder, lung and bronchus, and melanoma of the skin, while the top three histological types of SPMs were transitional cell carcinoma, squamous cell carcinoma, and malignant melanoma (Figure S1). According to the AJCC TNM staging, most primary tumors in SPM patients were classified as T1 (n = 2966, 46.6%), with N0 (n = 6164, 96.9%) and M0 (n = 6203, 97.5%) stages predominating. The PSA level was predominantly within the range of 4–10 (n = 3910, 61.4%). Most of the patient's Gleason scores were categorized in the ISUP1 (n = 2506, 39.4%). For PCa treatment, 2344 (36.8%), 2886 (45.4%), and 38 (0.6%) patients had PCa surgery, PCa radiotherapy, and PCa chemotherapy, respectively. For SPMs treatment, 1281 (20.1%) patients had SPMs radiotherapy, and 2066 (32.5%) patients had SPMs chemotherapy. There were no significant differences in the site of SPMs, SPMs radiation, SPMs chemotherapy, AJCC stage of SPMs, ISUP, age at diagnosis of PCa, race, clinical T stage of PCa, clinical N stage of PCa, clinical M stage of PCa, PSA value, PCa surgery, PCa radiation, PCa chemotherapy, interval between diagnoses, and marital status at diagnosis of SPMs (Table 1).

3.2 Nomogram variable screening

In the training cohort, all variables were initially included in a LASSO regression model to reduce multicollinearity and to identify prognostic variables. By employing this model with 10-fold cross-validation and selecting the model within a standard error of the minimum (Lambda.1SE), 10 prognostic factors were subsequently identified (Figure S2). Then, a backward regression analysis was performed to remove ISUP grades, SPM chemotherapy, and SPMs radiation, utilizing the AIC minimum principle. Moreover, through univariate Cox regression analysis and Kaplan–Meier (KM) analysis, seven variables were identified as independent prognostic factors for SPMs, demonstrating significant associations with OS (Figure 2A–F). Consequently, age at diagnosis of PCa, marital status at diagnosis of SPMs, SPMs site, clinical M stage of PCa, AJCC stage of SPMs, PCa surgery, and PSA levels were selected. A multivariate Cox regression model was constructed using the seven selected prognostic factors (Table 2).

3.3 Nomogram construction and validation

A nomogram was constructed for SPMs according to the variables screened (Figure 3). The total score is determined by adding 0 to 100 points for each predictor to evaluate the patient's survival chance. Most of the patients in this study had overall risk scores between 220 and 350. In the training set, AUC values for 1-, 3-, and 5-year OS predictions were AUC (95% CI) = 0.86 (0.84–0.87), AUC (95% CI) = 0.86 (0.84–0.87), and AUC (95% CI) = 0.84 (0.84–0.87), respectively (Figure 4A). In the validation set, AUCs for 1-, 3-, and 5-year OS predictions were: AUC (95% CI) = 0.87 (0.85–0.89), AUC (95% CI) = 0.85 (0.83–0.88), and AUC (95% CI) = 0.85 (0.83–0.88). (Figure 4B). Additionally, time-dependent AUC curves for both sets were generated (Figure 4I,J), illustrating the ability of the model to discriminate effectively over time. C-indexes, determined by following 1000 bootstrap resamplings, were utilized to assess the efficacy of the nomogram. The training and validation sets yielded C-indexes of 0.81 and 0.82 for OS predictions, respectively. This evidence further substantiates the distinguished discriminative capacity of the model. To confirm the accuracy of our model, we employed calibration plots to examine the correspondence between predicted results and real-world observations (Figure 4C–H). The results revealed that the predictions of 1-, 3-, and 5-year survival rates across the training and validation cohorts demonstrated satisfactory consistency with the actual scenarios. A web-based dynamic nomogram was created based on the nomogram to aid in generalization and clinical application. It may forecast the OS of patients with SPMs following PCa by precisely entering data on various independent prognostic markers.

3.4 Clinical value of the nomogram compared with the AJCC stage

To evaluate prognostic accuracy, NRI and IDI were compared between the nomogram and traditional AJCC staging. In the training cohort, NRI values for 1-, 3-, and 5-year OS predictions were NRI (95% CI) = 0.10 (0.06–0.14), NRI (95% CI) = 0.33 (0.22–0.40), and NRI (95% CI) = 0.35 (0.15–0.40), respectively. IDI values for the same timeframes were IDI (95% CI, p) = 0.09 (0.08–0.10, p < 0.001) for 1-year, IDI (95% CI, p) = 0.12 (0.11–0.13, p < 0.001) for 3-year, and IDI (95% CI, p) = 0.12 (0.11–0.13, p < 0.001) for 5-year OS predictions. These findings, corroborated in the validation cohort, demonstrate the enhanced prognostic accuracy of the nomogram over SPM AJCC staging alone (Table 3). The clinical benefits of the nomogram were compared with those of the SPMs AJCC staging. According to DCA curves, the nomogram could predict the 1-, 3-, and 5-year OS more accurately than the AJCC staging for nearly all threshold probabilities in the validation and training groups, as well as with the treat-all-patients scheme and the treat-none scheme (Figure 5A–F). Furthermore, through a comparison of the time-based C-index between the nomogram and the AJCC staging system, the nomogram consistently exhibited a higher C-index than the AJCC staging across all time points, which indicates that the nomogram has a greater capacity for discrimination and can more precisely forecast patient outcomes (Figure 5G,H).

3.5 Risk stratification based on the nomogram

To categorize patients with SPMs into different risk groups for more accurate prognostic prediction, each patient received a total score calculated by the nomogram. Based on the optimal cut-off points, patients were divided into two risk categories: low risk and high risk. It demonstrates that the Kaplan-Meier OS curve of the nomogram possessed notable discriminatory power in both risk categories. In contrast, the AJCC staging system had limited capacity to accurately identify high-risk patients in the training and validation cohorts (Figure 2G,H).

3.6 Exploring causal links between PCa and top 10 SPMs via TSMR

Based on SEER data identifying the top 10 high-frequency SPMs in PCa, we next sought to elucidate the causal directions between PCa and these SPMs using MR. Our analysis demonstrated a significant causal effect of PCa on urothelial carcinoma (UC) risk, namely for bladder cancer (BCa) and upper tract urothelial carcinoma (UTUC). For BCa, the random-effects IVW analysis indicated that PCa may promote BCa progression (OR [95% CI, p] = 1.16 [1.05–1.28, p = 3.9 × 10⁻³]). Similarly, for UTUC, the fixed-effects IVW analysis suggested a causal effect of PCa on UTUC progression (OR [95% CI, p] = 1.46 [1.16–1.85, p = 1.5 × 10⁻³]). In the MR analysis assessing the effect of PCa on BCa, the MR-Egger regression intercept revealed no evidence of horizontal pleiotropy (p = 0.90, n_SNP = 33), although heterogeneity was present (p = 0.02, n_SNP = 33). For the PCa–UTUC analysis, the MR-Egger regression intercept similarly indicated no evidence of horizontal pleiotropy (p = 0.93, n_SNP = 32), and no significant heterogeneity was detected (p = 0.35, n_SNP = 32). Notably, no causal relationships were observed between PCa and the other SPMs. Collectively, these findings support a causal role for PCa in increasing UC risk. For further details, please refer to Figure 6 and the Supporting Tables.

4 DISCUSSION

With enhanced survival rates for PCa patients, an increased likelihood of experiencing recurrences, metastases, SPMs, and multiple primary malignant neoplasms is expected, consequently diminishing patient survival outcomes[22,23]. Moreover, SPMs notably present a diagnostic conundrum in clinical settings, as they can impact the identical organ yet are histologically divergent from the initial tumor, qualifying neither as metastatic nor recurrent entities. Hence, our selection process excluded individuals with IPC and SPMs that were histologically analogous and diagnosed within the same calendar year to preclude any potential biases[19]. Our investigation identified that 2.32% of individuals with PCa were diagnosed with SPMs, lower than the previous finding[5,24]. This may be due to rigorous screening protocols, which minimize the likelihood of SPM misdiagnosis and facilitate the delivery of targeted therapeutic interventions.

This study systematically analyzed the distribution of SPMs in PCa patients, identifying 57 distinct SPMs. The results showed that the top 10 SPMs had an incidence rate exceeding 2%. The three most common locations for SPMs in this study were the bladder, lungs and bronchi, and cutaneous melanoma. While previous studies have reported that lung cancer, colorectal cancer, cutaneous melanoma, non-Hodgkin lymphoma, and BCa are common SPMs[5,24], our findings highlight BCa as particularly prominent in PCa patients, making it a major component of SPMs. Moreover, the increasing incidence of kidney and renal pelvis SPMs further underscores the pivotal role of urinary system cancers in the spectrum of SPMs observed in PCa patients.

To further validate this unique phenomenon, we employed a TSMR approach based on a triangulation strategy, which is increasingly advocated in causal inference research, to analyze PCa and its top ten SPMs[25]. TSMR uses SNPs as IVs, effectively controlling for environmental factors, socioeconomic backgrounds, and treatment interventions, thus providing more robust evidence for causal inference. The analysis surprisingly revealed that PCa may promote the occurrence of BCa and UTUC, a novel finding that is the first to establish a causal link between PCa and UC, strongly supporting the observations we made in the SEER database. We explored possible pathogenic mechanisms for this finding: on one hand, PCa and UC may share some common carcinogenic pathways[26]; on the other hand, the multifocal nature of UC suggests that urinary stasis may play a role in its pathological process[27]. Chronic inflammation[28] and genetic mutations[29] could further drive this progression. In summary, this study not only reveals the distribution characteristics of SPMs in PCa patients but also provides a new perspective on the molecular mechanisms underlying the co-occurrence of PCa and UC, offering important insights into the development of precision treatment strategies. In contrast, no significant causal associations were detected between PCa and the other top 10 SPMs. Furthermore, the lower heritability observed in certain malignancies (e.g., lung and liver cancers) implies that environmental factors may play a more critical role in their development[30,31].

This research successfully utilized the SEER database to construct and validate a nomogram predicting the 1-, 3-, and 5-year survival rates of PCa patients with SPMs. Through precise variable selection and multivariate Cox regression modeling, we identified significant impacts on the survival rates of SPM patients from factors such as age, marital status at diagnosis of SPMs, SPMs site, M stage, AJCC stage, PCa surgery, and PSA levels. Since SPM includes many hematologic malignancies or lymphomas, which are non-solid tumors and cannot be surgically treated, we did not include surgical information for SPM in the model. By integrating the aforementioned seven demographic and clinicopathological characteristics, the nomogram is a quantitative model that surpasses the AJCC staging system in prognostic assessment and clinical decision-making[10,32]. Although the AJCC staging has long been considered the preferred standard for predicting the OS of SPM patients, observed prognostic disparities among patients with the same stage reveal prognostic heterogeneity attributed to age, marital status, and clinicopathological features not covered by AJCC staging.

In comparison with the AJCC staging system, our study comprehensively evaluated the predictive efficacy of the nomogram versus the AJCC staging system using three parameters: DCA, NRI, and IDI, filling a void in existing research. By calculating the changes in the proportion of correctly and incorrectly classified predictions, NRI quantified the improvement of the new model over the baseline model in differentiating patients across various risk levels, thus assessing the actual enhancement in model performance. IDI refined the assessment of predictive performance by comparing the average differences in predicted probabilities of patient events between the two models, showcasing the new model's progress in enhancing predictive accuracy and facilitating personalized risk assessments for patients[33,34]. The results indicated that our nomogram surpassed the AJCC staging system in predictive capability through NRI and IDI evaluations. Additionally, DCA assessed the net benefit of the model across different risk thresholds, highlighting its practical value in clinical decision-making and aiding physicians in making data-driven treatment choices under specific circumstances[35,36]. Our research demonstrated that, compared to the traditional staging system, the nomogram exhibits high clinical utility and applicability in predicting survival. However, we do not consider the model as a direct replacement for the AJCC staging system; rather, we view it as a complementary tool that offers more refined risk predictions for enhancing clinical decision-making.

PCa is more common in older men, with incidence rates increasing with age, particularly in those aged 75 and above[1,37]. Previous studies have confirmed the significant clinical implications of variables, including age, race, marital status, PSA content, and treatment history for prognosis[38–41]. This proves that the variables selected in our nomogram are valid and clinically relevant. Although elderly patients may introduce various confounding factors, we have excluded the confounding factors of metastatic or recurrent tumors, thus not excluding elderly patients from the model.

Chemotherapy and radiotherapy are common and effective methods for treating a wide range of tumor types[42,43]. However, some radiotherapy records in the SEER database lacked detailed treatment information. To maximize data utilization and ensure consistency, we converted these records—and the diverse types of PCa surgeries—into binary variables (no/unknown vs. yes). In our data set, 55% of patients did not undergo surgery, 30% underwent radical prostatectomy, and the remaining patients underwent various other procedures. While this binary categorization increased sample sizes and improved model stability, it may have introduced some selection bias, as the “chemotherapy” and “radiotherapy” variables capture only overall treatment status rather than specific effects. Although selection bias is a frequent concern in retrospective analyses, our study aimed to develop a prognostic model to identify independent predictive factors for patient survival, rather than to compare the effectiveness of different treatment methods.

Compared to previous research such as Liu et al.[24], our study presents a more advanced prognostic model by integrating comprehensive demographic, pathological, and treatment data, significantly enhancing prediction accuracy and clinical applicability. We validated our model using ROC curves, calibration plots, the C-index, and time-dependent ROC curves, and further demonstrated its superiority with NRI and IDI analyses that refine risk stratification. Furthermore, we explored the causal relationship between PCa and the top 10 SPMs, revealing patterns of SPM occurrence following PCa from both causal inference and epidemiological perspectives. This robust approach not only surpasses earlier studies but also provides valuable insights into the genetic links between diseases, thereby supporting more informed long-term treatment planning and improved patient management.

Despite these findings, there are several limitations to this research. First, the SEER database lacks precise month-of-diagnosis data, so we excluded patients with same-year SPMs and IPC diagnoses to avoid misclassification. Additionally, although androgen deprivation therapy (ADT) is a cornerstone in the treatment of PCa, its omission from this study requires further clarification[44]. The SEER database has inherent limitations regarding the granularity of treatment details it provides[45]. Specifically, it lacks comprehensive information on specific systemic therapies, including the administration, dosage, and duration of ADT. This limitation precludes the inclusion of ADT-specific data in our analysis, as the database does not consistently capture the use of hormonal treatments such as LHRH agonists, LHRH antagonists, anti-androgens, and CYP17 inhibitors[46]. Finally, the model' s performance declined at very high-risk thresholds—likely due to a small sample size—suggesting that future studies should expand high-risk cohorts and explore more advanced modeling techniques.

Future research should prioritize multicenter validation and prospective cohort studies to confirm the clinical utility of the nomogram developed in this study. By incorporating data sets from multiple centers that include comprehensive clinical and lifestyle information, future investigations can provide more robust and generalizable evidence, ultimately improving the prognostic management of SPM patients after PCa.

5 CONCLUSION

With improved survival rates among PCa patients, the risk of developing SPMs has become a critical concern. Using the SEER database, we developed a nomogram to predict survival in PCa patients with SPMs through rigorous variable selection and data processing. Rather than replacing traditional AJCC staging, our model serves as a complementary tool, enhancing prognostic precision. In parallel, our TSMR analysis identified a strong causal link between PCa and UC, offering key genetic insights into their shared etiology. Together, these findings advance risk stratification and provide a foundation for precision treatment strategies in PCa management.

References

[1] Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209–249.

[2] Zheng X, Li X, Wang M, Shen J, Sisti G, He Z, et al. Second primary malignancies among cancer patients. Ann Transl Med. 2020;8:638.

[3] Vogt A, Schmid S, Heinimann K, Frick H, Herrmann C, Cerny T, et al. Multiple primary tumours: challenges and approaches, a review. ESMO Open. 2017;2:e000172.

[4] Travis LB, Wahnefried WD, Allan JM, Wood ME, Ng AK. Aetiology, genetics and prevention of secondary neoplasms in adult cancer survivors. Nat Rev Clin Oncol. 2013;10:289–301.

[5] Chattopadhyay S, Zheng G, Hemminki O, Försti A, Sundquist K, Hemminki K. Prostate cancer survivors: risk and mortality in second primary cancers. Cancer Med. 2018;7:5752–5759.

[6] Chen T, Fallah M, Jansen L, Castro FA, Krilavicuite A, Katalinic A, et al. Distribution and risk of the second discordant primary cancers combined after a specific first primary cancer in German and Swedish cancer registries. Cancer Lett. 2015;369:152–166.

[7] Keegan THM, Bleyer A, Rosenberg AS, Li Q, Goldfarb M. Second primary malignant neoplasms and survival in adolescent and young adult cancer survivors. JAMA Oncol. 2017;3:1554–1557.

[8] Ghiasvand R, Green AC, Veierød MB, Robsahm TE. Incidence and factors associated with second primary invasive melanoma in Norway. JAMA Dermatol. 2024;160:402.

[9] Yan Y, Liu H, Mao K, Zhang M, Zhou Q, Yu W, et al. Novel nomograms to predict lymph node metastasis and liver metastasis in patients with early colon carcinoma. J Transl Med. 2019;17:193.

[10] Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16:e173–e180.

[11] Zhang Y, Hong Y, Zhuang D, He X, Lin M. Bladder cancer survival nomogram: development and validation of a prediction tool, using the SEER and TCGA databases. Medicine. 2019;98:e17725.

[12] Wang C, Yang C, Wang W, Xia B, Li K, Sun F, et al. A prognostic nomogram for cervical cancer after surgery from SEER database. J Cancer. 2018;9:3923–3928.

[13] Feng SS, Li H, Fan F, Li J, Cao H, Xia ZW, et al. Clinical characteristics and disease-specific prognostic nomogram for primary gliosarcoma: a SEER population-based analysis. Sci Rep. 2019;9:10744.

[14] Ding Q, Wang K, Li Y, Peng P, Zhang D, Chang D, et al. Clinical characteristics and survival analysis of patients with second primary malignancies after hepatocellular carcinoma liver transplantation: a SEER-based analysis. Am J Clin Oncol. 2023;46:284–292.

[15] Deng R, Zhou J, Qiu J, Cai L, Gong K. Clinical characteristics analysis and prognostic nomogram for predicting survival in patients with second primary prostate cancer: a population study based on SEER database. J Cancer Res Clin Oncol. 2023;149:11791–11806.

[16] Davey Smith G, Hemani G. Mendelian randomization: genetic anchors for causal inference in epidemiological studies. Hum Mol Gen. 2014;23:R89–R98.

[17] Skrivankova VW, Richmond RC, Woolf BAR, Davies NM, Swanson SA, VanderWeele TJ, et al. Strengthening the reporting of observational studies in epidemiology using Mendelian randomisation (STROBE-MR): explanation and elaboration. BMJ. 2021;375:n2233.

[18] Rashkin SR, Graff RE, Kachuri L, Thai KK, Alexeeff SE, Blatchins MA, et al. Pan-cancer study detects genetic risk variants and shared genetic basis in two large cohorts. Nat Commun. 2020;11:4423.

[19] Warren S, Gates O. Multiple primary malignant tumors: a survey of the literature and statistical study. Am J Cancer. 1932;16:1358–1414.

[20] Elliott M. Regression methods in biostatistics: Linear, logistic, survival, and repeated measures models, Edited by Vittinghoff, E., Glidden, D. V., Shiboski, S. C., and Mcculloch, C. E. Biometrics. 2006;62:1271–1272.

[21] Burgess S, Butterworth A, Thompson SG. Mendelian randomization analysis with multiple genetic variants using summarized data. Genet Epidemiol. 2013;37:658–665.

[22] Campbell, Jr. LV Multiple primary malignant neoplasms. Arch Surg. 1969;99:401–405.

[23] Sartor O, De Bono JS. Metastatic prostate cancer. N Engl J Med. 2018;378(7):645–657.

[24] Liu Y, Zhang P, Zhang Y, Zheng L, Xu W, Hou D, et al. Clinical characteristics and overall survival nomogram of second primary malignancies after prostate cancer, a SEER population-based study. Sci Rep. 2021;11:1293.

[25] Lawlor DA, Tilling K, Davey Smith G. Triangulation in aetiological epidemiology. Int J Epidemiol. 2016;45:1866–1886.

[26] Singh A, Kinoshita Y, Rovito Jr. PM, Landas S, Silberstein J, Nsouli I, et al. Higher than expected association of clinical prostate and bladder cancers. J Urol. 2008;179:S2–S5.

[27] Kirby RS, Lowe D, Bultitude MI, Shuttleworth KED. Intra-prostatic urinary reflux: an aetiological factor in abacterial prostatitis. Br J Urol. 1982;54:729–731.

[28] Platz EA, De Marzo AM. Epidemiology of inflammation and prostate cancer. J Urol. 2004;171:S36–S40.

[29] Shariat SF, Tokunaga H, Zhou J, Kim J, Ayala GE, Benedict WF, et al. p53, p21, pRB, and p16 expression predict clinical outcome in cystectomy with bladder cancer. J Clin Oncol. 2004;22:1014–1024.

[30] Hopkinson NS. Smoking and lung cancer-70 long years on. BMJ. 2024;384:q443.

[31] Singal AG, Kanwal F, Llovet JM. Global trends in hepatocellular carcinoma epidemiology: implications for screening, prevention and therapy. Nat Rev Clin Oncol. 2023;20:864–884.

[32] Pu N, Li J, Xu Y, Lee W, Fang Y, Han X, et al. Comparison of prognostic prediction between nomogram based on lymph node ratio and AJCC 8th staging system for patients with resected pancreatic head carcinoma: a SEER analysis. Cancer Manag Res. 2018;10:227–238.

[33] Uno H, Tian L, Cai T, Kohane IS, Wei LJ. A unified inference procedure for a class of measures to assess improvement in risk prediction systems with survival data. Stat Med. 2013;32:2430–2442.

[34] Pencina MJ, D'Agostino Sr. RB, Steyerberg EW. Extensions of net reclassification improvement calculations to measure usefulness of new biomarkers. Stat Med. 2011;30:11–21.

[35] Vickers AJ, Elkin EB. Decision curve analysis: a novel method for evaluating prediction models. Med Decis Making. 2006;26:565–574.

[36] Fitzgerald M, Saville BR, Lewis RJ. Decision curve analysis. JAMA. 2015;313:409–410.

[37] Droz JP, Albrand G, Gillessen S, Hughes S, Mottet N, Oudard S, et al. Management of prostate cancer in elderly patients: recommendations of a Task Force of the International Society of Geriatric Oncology. Eur Urol. 2017;72:521–531.

[38] Bergengren O, Pekala KR, Matsoukas K, Fainberg J, Mungovan SF, Bratt O, et al. 2022 update on prostate cancer epidemiology and risk factors—a systematic review. Eur Urol. 2023;84:191–206.

[39] Chen ZH, Yang KB, Zhang Y, Wu CF, Wen DW, Lv JW, et al. Assessment of modifiable factors for the association of marital status with cancer-specific survival. JAMA Netw Open. 2021;4:e2111813.

[40] Nam RK, Toi A, Klotz LH, Trachtenberg J, Jewett MAS, Appu S, et al. Assessing individual risk for prostate cancer. J Clin Oncol. 2007;25:3582–3588.

[41] Stephenson AJ, Kattan MW, Eastham JA, Bianco Jr. FJ, Yossepowitch O, Vickers AJ, et al. Prostate cancer-specific mortality after radical prostatectomy for patients treated in the prostate-specific antigen era. J Clin Oncol. 2009;27:4300–4305.

[42] Chabner BA, Roberts TG. Chemotherapy and the war on cancer. Nat Rev Cancer. 2005;5:65–72.

[43] Baumann M, Krause M, Overgaard J, Debus J, Bentzen SM, Daartz J, et al. Radiation oncology in the era of precision medicine. Nat Rev Cancer. 2016;16:234–249.

[44] Hellerstedt BA, Pienta KJ. The current state of hormonal therapy for prostate cancer. CA Cancer J Clin. 2002;52:154–179.

[45] Howlader NNA, Krapcho M, Garshell D, Miller D, Kosary F. SEER cancer statistics review, 1975-2017. Bethesda, MD: National Cancer Institute.

[46] Duggan MA, Anderson WF, Altekruse S, Penberthy L, Sherman ME. The surveillance, epidemiology, and end results (SEER) program and pathology: toward strengthening the critical relationship. Am J Surg Pathol. 2016;40(12):e94–e102.

RIGHTS & PERMISSIONS

2025 The Author(s). UroPrecision published by John Wiley & Sons Australia, Ltd on behalf of Higher Education Press.