Impact of Dual Thyroid and Prostate Cancer Diagnosis on Patient Survival: Insights From a Large Cohort Study

Introduction

Thyroid cancer (TC) and prostate cancer (PC) are among the most prevalent malignancies. TC ranks as the seventh most common cancer globally [1] and generally has a favorable prognosis, with a 5-year survival rate of 98% in the United States [2]. PC is the second most common cancer in men worldwide, representing 14.2% of new male cancer cases [1]; its 5-year survival rate is 97% [2]. Despite this favorable prognosis, the high incidence of PC causes a substantial mortality burden, ranking fifth in global male cancer mortality [1].

Although TC and PC differ in pathogenesis, clinical manifestations, and treatment, their occurrence and progression are both influenced by endocrine factors. Thyroid-stimulating hormone suppression therapy and androgen deprivation therapy – routine treatments for differentiated thyroid cancer and PC, respectively – are closely related to cancer prognosis [3–6]. Notably, lower thyroid-stimulating hormone levels and elevated thyroid hormone levels have been associated with an increased risk of PC [7,8], suggesting biological interplay between the 2 cancers.

With improvements in cancer screening and clinical techniques, the recorded incidences of these 2 cancers have increased, and survival times have lengthened. Accordingly, the number of patients diagnosed with both TC and PC has also risen. Previous retrospective studies have shown that patients with TC often have an increased risk of developing a second primary cancer [9–11]; conversely, cancers such as PC increase the risk of developing TC [12,13]. Studies specifically involving patients with TC as 1 of multiple primary cancers have revealed heterogeneous prognostic patterns. For instance, patients with both breast cancer and TC demonstrated a lower mortality risk compared with those only exhibiting breast cancer or TC [14,15]. Similarly, a history of TC was associated with a better prognosis in patients with liver cancer [16], but it conferred no significant survival difference among individuals with lung cancer [17]. Currently, the prognosis of patients with both TC and PC is uncertain.

Given the increasing co-occurrence of these 2 cancers, their potential endocrine interplay, and the critical lack of prognostic data, their mutual impact on survival must be evaluated. Here, we analyzed the long-term survival outcomes of patients with TC and PC using the Surveillance, Epidemiology, and End Results (SEER) database.

Material and Methods

ETHICAL CONSIDERATIONS:

This study was conducted using de-identified, publicly accessible data from the SEER database. Because the research involved no direct patient contact and all data were anonymized, Institutional Review Board approval was not mandatory, and the requirement for informed consent was waived.

DATA SOURCE AND POPULATION:

Population data for this retrospective cohort study were obtained from the SEER 17 Registries dataset (November 2023 submission, 2000–2021). This dataset covers 17 registry areas, representing approximately 26.5% of the United States population. Given that prostate cancer is a sex-specific malignancy, this study focused on male patients. Men with records of both TC (International Classification of Diseases [ICD]-O-3 site recode C739) and PC (ICD-O-3 site recode C619) were assigned to the exposed group (TC+PC group). The exclusion criteria were as follows: (1) patients with more than 2 malignant tumors; (2) age at diagnosis younger than 18 years; (3) TC or PC identified at death or autopsy; (4) TC or PC not confirmed by histopathology; and (5) follow-up time of 0 days. Additionally, men with only TC were assigned to the unexposed group for analysis of TC prognosis (TC-only group), and men with only PC were assigned to the unexposed group for analysis of PC prognosis (PC-only group). Exclusion criteria for the TC-only and PC-only groups were identical to exclusion criteria for the TC+PC group, except for the first criterion. All data were screened using SEER*Stat software (seer.cancer.gov/seerstat), version 8.4.3, with permission to access the SEER Research Data.

VARIABLE SELECTION AND DEFINITIONS:

All patients were men. To assess socioeconomic characteristics, the following variables were collected: year of diagnosis, age at diagnosis, race, marital status, and median household income at the county level. To evaluate clinicopathological characteristics, the following variables were included: histology, TNM stage, and treatments received (surgery, radiotherapy, chemotherapy). For patients with PC, 2 additional indicators – Gleason score and prostate-specific antigen (PSA) level – were included because of their relationship to disease staging.

Patient survival was selected as the primary outcome of this study. Outcome events included all-cause death, cancer-specific death, and competing events (death due to non-cancer causes). Follow-up time was calculated from the date of diagnosis until loss to follow-up, death, or the last follow-up in 2021, whichever occurred first.

STATISTICAL ANALYSIS:

The normality of continuous variables was assessed using the Kolmogorov-Smirnov test. Continuous variables with a normal distribution were reported as mean±standard deviation, whereas those with a non-normal distribution were reported as median and interquartile range. Categorical variables were reported as frequencies and percentages. Differences between groups for normally distributed continuous variables were analyzed using the t-test, whereas non-normally distributed continuous variables were analyzed using the Wilcoxon rank-sum test. The chi-square test or Fisher’s exact test was used to analyze categorical variables.

Median follow-up time was estimated using the reverse Kaplan-Meier method. Kaplan-Meier curves with the log-rank test were used to evaluate patient survival without adjustment for confounding factors. To address potential confounding effects, Cox proportional hazards regression modeling was used to adjust for confounders and estimate hazard ratios (HRs) for all-cause and cancer-specific mortality; the proportional hazards assumption was verified by inspection of Schoenfeld residual plots. Given that deaths from non-cancer causes are common in the study population and may compete with cancer-specific death, competing risk analyses were conducted using the Fine-Gray test.

Before adjusting for confounding factors, we performed multivariate imputation by chained equations to handle missing data, assuming data were missing at random. The imputation model included survival outcomes and all variables considered for adjustment: age at diagnosis, race, marital status, income level, year of diagnosis, histology, TNM stage, surgery, radiotherapy, chemotherapy, Gleason score (categorized according to International Society of Urological Pathology [ISUP] grade), and PSA level. Ten imputations were performed, and Rubin’s rules were used to combine statistical estimates. Multivariable analyses were conducted in 2 stages to assess the impact of different confounder sets. First, models were adjusted for socioeconomic factors (year of diagnosis, age, race, marital status, and income). Second, a comprehensive model was adjusted for all variables listed above, encompassing both socioeconomic and clinicopathological characteristics.

To assess the robustness of our primary findings and address potential confounding effects related to differential timing of diagnosis, a sensitivity analysis was conducted via propensity score matching (PSM). Patients in the TC+PC group were matched at a 1: 5 ratio with control patients based on diagnostic characteristics: age at diagnosis and year of diagnosis of either thyroid cancer – when analyzing TC survival outcomes (matched with the TC-only group) – or of prostate cancer when analyzing PC survival outcomes (matched with the PC-only group). Nearest neighbor matching without replacement was applied, using a caliper width of 0.2 of the standard deviation of the logit of the propensity score. After matching, all primary survival analyses were repeated in the respective matched cohorts. Data analyses were performed using R software (version 4.3.3), and 2-tailed P-values <0.05 were considered statistically significant.

Results

BASELINE INFORMATION OF THE STUDY POPULATION:

Comparisons of baseline data related to TC between the TC+PC group and the TC-only group are shown in Table 1; comparisons related to PC between the TC+PC group and the PC-only group are shown in Table 2. According to the order of cancer diagnosis, the TC+PC group was divided into the TC-1st group, in which TC was diagnosed prior to PC, and the PC-1st group, in which PC was diagnosed prior to TC. Baseline data for the TC-1st and PC-1st groups are also summarized in Tables 1 and 2.

In terms of socioeconomic characteristics, the total population (TC and PC) was predominantly married and had a moderate household income ($40 000–$120 000). Age at TC diagnosis was significantly older in the TC+PC group than in the TC-only group (median [interquartile range], 65 [58–72] vs 51 [40–62], P<0.001); year of diagnosis was earlier in the TC+PC group (P 0.001). Age at PC diagnosis was slightly younger in the TC+PC group than in the PC-only group (65 [58–72] vs 66 [60–72], P<0.001); year of diagnosis was later in the TC+PC group (P<0.001).

In terms of clinicopathological characteristics, for TC, the vast majority of patients in both groups received surgical treatment (96.5%), and histology was predominantly differentiated thyroid cancer (papillary thyroid cancer+follicular thyroid cancer, 91.3%). The TC+PC group had a more advanced American Joint Committee on Cancer (AJCC) stage than the TC-only group (stage III/IV, 28.0% vs 21.4%, P<0.001). Given the age-dependent staging of differentiated thyroid cancer [18,19], this stage difference is consistent with the pronounced age disparity between groups. The proportions of patients receiving radiotherapy were 44.3% and 49.0% in the TC+PC and TC-only groups, respectively. There was no significant difference between groups in the proportions receiving surgery or chemotherapy. For PC, more patients in the TC+PC group underwent surgery compared with the PC-only group (44.2% vs 38.3%, P<0.001). Histology was predominantly adenocarcinoma (98.9%). Compared with patients in the PC-only group, those with TC+PC had lower Gleason scores, lower PSA levels, and earlier AJCC stages. The proportions of patients receiving radiotherapy or chemotherapy were similar between the TC+PC and PC-only groups.

The reverse Kaplan-Meier method estimated median follow-up times of 10.17 years in the PC-only group, 8.00 years in the TC-only group, and 9.75 years in the TC+PC group; the last follow-up was recorded in 2021. Median overall survival was 17.4 years in the PC-only group but was not reached in the TC-only and TC+PC groups.

RELATIONSHIP BETWEEN PC HISTORY AND SURVIVAL IN PATIENTS WITH TC:

The comparison of all-cause mortality among patients with TC according to PC history is shown in Table 3. Before adjustment for confounding factors, the Kaplan-Meier curve showed that overall survival was worse in the TC+PC group than in the TC-only group (P<0.001, Figure 1); the risk of all-cause mortality was higher in the TC+PC group than in the TC-only group (HR=1.32; 95% confidence interval [CI], 1.20–1.46; P<0.001). After adjustment for socioeconomic characteristics (year of diagnosis, age at diagnosis, race, marital status, and income level) via Cox regression modeling, the TC+PC group showed a lower risk of all-cause mortality (HR=0.67; 95% CI, 0.60–0.74; P<0.001) than the TC-only group. When clinicopathological characteristics (histopathology, AJCC T stage, AJCC N stage, AJCC M stage, surgery, radiotherapy, chemotherapy) were added as adjustment factors, the favorable prognosis persisted in the TC+PC group (HR=0.78; 95% CI, 0.71–0.87; P<0.001). This group was then divided into 2 subgroups (TC-1st and PC-1st) according to the chronological order of cancer diagnoses, and subgroup analyses were performed. The results showed that the TC-1st group had a lower risk of all-cause mortality than the TC-only group regardless of adjustment for confounding factors, with HRs ranging from 0.45 to 0.71 (all P<0.001). In contrast, the PC-1st group had a higher risk of all-cause mortality than the TC-only group before adjustment (P<0.001); no statistically significant difference was observed after adjustment for confounders (Model 1, P=0.102; Model 2, P=0.544).

We then used Cox proportional hazards regression (Table 3) and competing risk modeling (Table 4) to compare TC-specific mortality risk among patients with TC according to PC history. Cox regression models showed that TC-specific mortality risk was significantly lower in the TC+PC group than in the TC-only group (P=0.003 for the univariable model; P<0.001 for the multivariable models); the HR after adjustment for all confounding factors was 0.58 (95% CI, 0.47–0.72). In further subgroup analyses, the TC-1st group had significantly lower risks of TC-specific mortality regardless of adjustment for confounders, with HRs ranging from 0.15 to 0.25 (all P<0.001). The PC-1st group had a lower TC-specific mortality risk than the TC-only group after adjustment for socioeconomic characteristics (P<0.001). Competing risk modeling showed results similar to those of the Cox regression models (Table 4). Taken together, these findings indicate that a history of PC is linked to improved survival in patients with TC, as reflected by lower risks of all-cause and TC-specific mortality. This association was particularly pronounced in the TC-1st group. These findings were corroborated by PSM sensitivity analyses (Tables 3, 4).

RELATIONSHIP BETWEEN TC HISTORY AND SURVIVAL IN PATIENTS WITH PC:

The comparison of all-cause mortality risk for patients with PC according to TC history is shown in Table 5. Compared with the PC-only group, the Kaplan-Meier curve showed that the TC+PC group had more favorable overall survival (P=0.003, Figure 2). Consistent with this finding, Cox regression models revealed that all-cause mortality risk was lower in the TC+PC group than in the PC-only group before and after adjustment for confounding factors (all P<0.001). The HR after adjustment for all confounders was 0.83 (95% CI, 0.75–0.91). Subgroup analysis showed that the PC-1st group had a lower risk of all-cause mortality than the PC-only group across all 3 models (all P<0.001); the TC-1st group showed no significant difference after adjustment for confounders, despite a lower risk before adjustment (HR=0.73; 95% CI, 0.62–0.86; P<0.001).

We next used Cox regression and competing risk models to analyze PC-specific mortality in patients with PC according to TC history (Tables 5, 6). All 3 Cox regression models showed lower PC-specific mortality risk in the TC+PC group than in the PC-only group (all P<0.001, Table 5), and the HR after adjustment for all confounding factors was 0.62 (95% CI, 0.50–0.78). In subgroup analyses, the PC-1st group demonstrated low risks similar to the overall cohort (all P<0.001). The TC-1st group had a lower PC-specific mortality risk than the PC-only group before adjustment and after adjustment for socioeconomic characteristics (P=0.001 and 0.020, respectively), but the statistical significance did not persist after adjustment for both socioeconomic and clinicopathological characteristics. As shown in Table 6, the results from competing risk models were consistent with findings from the Cox regression models. Complementary PSM analyses (Tables 5, 6) confirmed the robustness of the primary findings, given that the significant association persisted after matching in the overall cohort and in subgroup analyses. In summary, a history of TC was associated with improved survival outcomes in patients with PC; the association was particularly prominent in the PC-1st subgroup.

Discussion

As the population of long-term cancer survivors grows, the prognosis for patients with multiple primary malignancies is an increasingly important aspect of clinical management. Our analysis of SEER data revealed that patients with both TC and PC experience superior survival relative to those with either cancer alone. This association persisted after comprehensive adjustment for socioeconomic and clinicopathological confounders; it was corroborated by PSM sensitivity analyses.

Interestingly, compared with patients who had TC or PC alone, the decreased mortality risks associated with TC+PC co-occurrence were more pronounced when TC or PC occurred earlier than PC or TC, respectively. For example, the adjusted HR for TC-specific mortality was 0.58 in the TC+PC group and decreased to 0.25 in the TC-1st group. Similarly, the adjusted HR for PC-specific mortality was 0.62 in the TC+PC group and decreased to 0.51 in the PC-1st group. Although the PC-1st group had a lower mortality risk than the TC-only group and the TC-1st group had a lower mortality risk than the PC-only group after adjustment for socioeconomic characteristics, these effects did not remain statistically significant following additional adjustment for clinicopathological factors. In a more refined analysis using PSM, the association between the TC-1st subgroup and reduced mortality relative to the PC-only group remained statistically significant after full adjustment, whereas the association for the PC-1st group compared with the TC-only group was not sustained.

The observed survival advantage must be interpreted in the context of substantial baseline differences between the TC+PC and the TC-only or PC-only groups, especially regarding age, year of diagnosis, and certain clinicopathological factors. These imbalances create potential confounding effects that warrant careful consideration. In particular, patients already under surveillance for their first cancer likely undergo more frequent screening, which could lead to earlier diagnosis and more intensive management of the second cancer. Furthermore, the requirement of surviving the first cancer to be diagnosed with the second inherently introduces immortal time bias. Although our sequential multivariable adjustments and PSM analyses were designed to mitigate these issues, residual confounding effects from unmeasured differences in surveillance intensity, comorbidities, or their complex interplay cannot be fully excluded. Therefore, these factors represent plausible explanations that should be considered when interpreting our findings.

In addition to methodological factors, the observed association raises the question of whether a biological interaction contributes to this relationship. A potential mechanism warranting exploration is crosstalk between thyroid hormone and androgen signaling pathways. The effects of the hypothalamic-pituitary-thyroid axis on TC and the hypothalamic-pituitary-gonadal axis on PC are well documented [4,6,20–22]. Notably, recent studies suggest additional crosstalk between these 2 hormonal systems, influencing cancer development and prognosis [23–27]. For example, the thyroid hormone-binding protein μ-crystallin can inhibit both thyroid hormone and androgen signaling; its low expression is linked to early recurrence and poor prognosis in patients with PC [25,27]. Additionally, testosterone is an important androgen that upregulates troponin T type 1 (Tnnt1) expression, activating the p38/Janus kinase (JNK) pathway and promoting the growth, migration, and invasion of papillary thyroid cancer [26]. Recent studies have also shown that androgen receptor activation promotes aging and apoptosis in TC cells by downregulating programmed death ligand 1 expression or by regulating several key signaling pathways [23,24]. Although these mechanistic insights are suggestive and provide a plausible biological framework, they remain speculative in the context of our clinical findings and require direct experimental validation.

The strengths of our study include its large, population-based sample from the SEER database, the use of competing risk models to address non-cancer deaths, and the application of both multivariable adjustment and PSM to address confounding effects. However, several limitations must be acknowledged. First, because this retrospective study used registry data, the findings are susceptible to residual confounding effects from unmeasured factors, such as detailed treatment regimens, comorbidities, and lifestyle factors, as well as potential data inaccuracies. Pronounced baseline imbalances in age and stage heighten this concern despite statistical adjustment. Furthermore, the association was not consistent across all subgroups. This heterogeneity suggests that unmeasured factors or residual confounding vary across patient subsets, indicating that the observed effect may not be universal. Second, the SEER database does not capture disease recurrence, preventing analysis of recurrence-free survival. Third, although PSM mitigated some biases related to diagnostic timing, it cannot fully eliminate immortal time bias or surveillance bias, which may partly explain the observed associations. Fourth, no statistical correction for multiple testing was applied; therefore, interpretation of P values near 0.05 requires caution. Finally, the speculative nature of the proposed hormonal crosstalk mechanism highlights the need for confirmatory experimental studies.

Conclusions

This retrospective cohort study revealed a mutually favorable survival association between TC and PC, indicating that a history of either cancer warrants consideration during prognostic evaluation. These findings highlight the need for further research to validate this association and explore its underlying biological mechanisms.