Study population

We queried SEER*Stat (November 2023 update) to extract cases from the SEER 18 registries diagnosed between 2000 and 2021 [8]. Eligible records had the primary site coded as heart using the ICD-O-3 topography code C38.0. A total of 784 cases of PMCTs were initially identified in the SEER database between 2000 and 2021. After excluding 54 cases with missing survival information, the final analytic cohort consisted of 730 patients. Variables were harmonized using a project data dictionary. Demographic covariates included age at diagnosis, categorized as youth (< 45 years), middle (45–64 years), and elderly (≥ 65 years); sex; race/ethnicity; neighborhood median household income (quartiles) based on the national median during 2022 ($75,000); and rural-urban continuum code. These various groupings intentionally reflected factors that are associated with inequities in healthcare, including race, socioeconomic status, and population density. These populations are often less likely to have equal access to healthcare and thus, cardiac imaging and other diagnostic modalities in comparison to their counterparts [9]. Clinical and histologic covariates included tumor size (continuous), ICD-O-3 histology, a prespecified prognosis group, cancer subtype (Soft tissue, Hematologic, Embryonal, Other), and indicators of distant metastases to bone, brain, liver, and lung.

Prognosis groups were designated based on literature-derived estimates of percent survival at 5 years [7, 10, 11]. Prognosis I defines tumors with a prognosis worse than 50% at 5 years, Prognosis II defines tumors with a prognosis between 50–95% at 5 years, and Prognosis III defines tumors with > 95% survival at 5 years. Cancer subtype was determined by literature-derived histological proximity based on author consensus (> 2 authors agreeing). Stratification by author consensus was pursued in an effort to further describe pre-existing biology-driven classification of cardiac malignancy that continues to be useful but limited [12, 13]. This investigation categorized cardiac tumors by literature-derived estimates of 5-year percent survival to ascertain similarities in clinical significance over histology as a distinct tumor niche in primary cardiac tumors to elucidate large, therapeutically informative trends that address the gaps in histology-driven studies.

Treatment covariates included surgery of the primary site (SEER RX Summ—Surg Prim Site [1998+]), radiation type, and chemotherapy (yes/no). These therapies were administered as part of the first course of treatment following the initial cancer diagnosis; subsequent treatments given after disease progression or recurrence were not comprehensively captured within the database. Cases coded as “unknown/death” under the surgery variable represent death certificates that had the treatment information listed as unavailable; thus, it is unknown whether surgical treatment was performed.

OS in months and vital status were obtained from SEER, and death from any cause was considered an event. OS in months began at the time of initial diagnosis and considered death from all causes, not exclusively cancer-related. Because metastasis indicators in SEER are consistently available from 2010 forward, models that included bone/brain/liver/lung metastasis were restricted to diagnoses from 2010–2021, with complete-case analysis for those covariates.

Statistical analysis

We described cohort characteristics with standard descriptive statistics and summarized annual counts (2000–2021) for incidence visualization. To screen hypothesis-generating associations between covariates and survival time, we computed Pearson correlation between continuous variables (e.g., tumor size vs OS months) and Spearman correlation for ordinal or categorical variables vs OS (months); two-sided p-values < 0.05 were considered statistically significant. Unadjusted survival was estimated with the Kaplan-Meier (KM) methods overall and within prespecified strata (All tumors; Prognosis I; Prognosis II; Hematologic; Soft tissue; Embryonal; Other). We reported median OS and KM survival probabilities at 12, 24, and 60 months, and compared curves with log-rank tests. We then fit Cox proportional hazards models separately within each stratum with covariates including age group, sex, race/ethnicity, surgery type, radiation, chemotherapy, household income relative to the median household income in 2022 (above or below $75,000), rural-urban code, tumor size in millimeters (mm), and the four metastasis indicators (when available). Cox proportional hazards regression was used to explore factors associated with OS. Because proportional hazards diagnostics were not formally assessed, the results should be interpreted as exploratory. Given the rarity of PMCTs and small cell counts in some subgroups (particularly embryonal histologies), results from sparse strata were interpreted cautiously. Indeed, results from the embryonal and origins of other types were excluded from the final explorative analysis. Metastasis was also excluded in the subgroup correlational analysis due to low outcome variability. All analyses were conducted in Python within the Jupyter Notebook environment using pandas for data management and lifelines for survival analyses; figures were generated with matplotlib. Two-sided α was set at 0.05.

Descriptive results

Correlational analysis

Cox proportional hazards model

Table 7 depicts statistically significant predictors of hazard from explorative Cox proportional hazards models stratified by cohort.

In the All tumors cohort, younger age was protective [middle vs elderly hazard ratio (HR) 0.69 (0.50–0.95); p = 0.019; youth vs elderly HR 0.52 (0.35–0.78); p = 0.003] while brain and lung metastases conferred higher hazard (HR 2.04 and 1.89; p = 0.047 and p = 0.011, respectively).

In Prognosis I, larger tumor size was associated with a small per-mm increase in hazard (HR 1.01; p = 0.031). Younger age was protective (middle HR 0.56; p = 0.005; youth HR 0.49; p = 0.004). Limited surgical interventions were associated with lower hazard, including local excision (HR 0.63; p = 0.050) and partial resection (HR 0.44; p = 0.021).

In Prognosis II, younger age again, reduced hazard (middle HR 0.59; p = 0.003; youth HR 0.52; p = 0.004). Excisional biopsy was protective (HR 0.50; p = 0.049), whereas “unknown/death” surgical status was associated with increased hazard (HR 1.54; p = 0.004). Brain and lung metastases remained adverse (HR 2.43 and 1.93; p = 0.017 and p = 0.010).

In Hematological cancers, the age gradient was pronounced (middle HR 0.30; p = 0.001; youth HR 0.13; p < 0.0001). Non-Hispanic Asian/Pacific Islander (NHAPI) race and Hispanic ethnicity were associated with lower hazard (HR 0.41; p = 0.026 and HR 0.38; p = 0.020). Excisional biopsy was protective (HR 0.44; p = 0.047), while unknown/death surgical status increased hazard (HR 2.18; p = 0.006).

In Soft tissue tumors, size showed a modest per-mm hazard increase (HR 1.02; p = 0.017). Younger age was protective (middle HR 0.46; p < 0.0001; youth HR 0.30; p < 0.0001). Several procedures were associated with reduced hazard, including enucleation (HR 0.39; p = 0.003), excisional biopsy (HR 0.36; p = 0.001), local excision (HR 0.35; p < 0.0001), and partial resection (HR 0.44; p < 0.0001).

In Other tumors, NHAPI race and Hispanic ethnicity were associated with lower hazard (HR 0.12; p = 0.004 and HR 0.17; p = 0.009). Partial resection was protective (HR 0.37; p = 0.007), whereas upper-quartile income was associated with higher hazard (HR 1.73; p = 0.022).

Kaplan-Meier survival curves across all tumor types

As seen in Table 8, the statistical significance in survival probability was noted by prognosis grouping and cancer subtype most definitively across time points (p < 0.001). Prognosis II maintains consistently improved survival compared to Prognosis I at 30, 60, and 120 months time points. Chemotherapy status also showed strong separation of survival curves at 30, 60, and 120 months (p < 0.001). Brain metastasis was consistently associated with poorer survival at all evaluated intervals (30 months: p = 0.017; 60 months: p = 0.017; 120 months: p = 0.017), while bone metastasis was associated with poorer short- and intermediate-term survival (30 months: p = 0.017; 60 months: p = 0.028) with attenuation by 120 months (p = 0.084). Lung metastasis was associated with poorer short-term survival at 30 months (p = 0.046), but was not sustained over time. Liver metastasis did not significantly separate survival curves at any interval examined. Sex, age, race, tumor size category, income, and rural-urban status were not associated with durable long-term differences in unadjusted survival.

Categorical analysis

Incidence trends

Figures 1 and 2 demonstrate a gradual increase in PMCT incidence over time, representing overall incidence and poor-prognosis incidence, respectively. While this could reflect a true biological increase, it is more likely due to improved detection through advanced imaging and greater diagnostic awareness. This parallels trends in other rare tumors, where incidence rates have risen with increased imaging utilization [2]. The continual uptick in reported cases emphasizes the importance of earlier implementation of cardiac imaging, such as echocardiography or cardiac MRI, in patients presenting with unexplained cardiopulmonary symptoms. When examining Figure 3, rates increased and plateaued in the 2010s compared with the 2000s, followed by a gradual decline in the later years of the dataset. This trend contrasts with the patterns observed in poor-prognosis and overall PMCT incidence. These findings may suggest a shift in the distribution of PMCT prognostic categories over the study period, with potentially fewer improved-prognosis tumors being diagnosed in recent years. This may be related to surveillance factors, such as increased reporting, clinical recognition of more aggressive tumors, or improvements in diagnostic tests that enhance identification of higher-risk tumors. Overall, the trends shown in the figures likely reflect evolving diagnostic and reporting practices that may influence observed incidence patterns rather than reflecting true changes in tumor biology.

Display full size

Figure 1. Incidence of all primary malignant cardiac tumors between 2000 and 2021.

Display full size

Figure 2. Incidence of poor-prognosis primary malignant cardiac tumors between 2000 and 2021.

Display full size

Figure 3. Incidence of improved prognosis primary malignant cardiac tumors between 2000 and 2021.

Implications and future directions

These findings reinforce that prognostic grouping carries clinical significance:

• Prognosis I (sarcoma-dominant): Aggressive multimodality therapy is warranted, with early surgical referral emphasized. Partial resections provide tangible benefit, and chemotherapy should be integrated into multimodal treatment planning. Early imaging to detect resectable disease is also essential.

• Prognosis II (hematologic-dominant): Systemic therapy, not surgery, should be prioritized. The poor outcomes associated with chemotherapy highlight the need to adapt hematologic oncology advances like rituximab-based regimes or CAR-T into cardiac manifestations of hematologic malignancies.

• Across all PMCTs: Younger age and absence of metastasis are the strongest consistent predictors of survival. This emphasizes the importance of earlier diagnosis, comprehensive staging, and targeted therapeutics for optimal survival outcomes.

Limitations

This study design has various limitations. First, the rarity of PMCTs, as well as the use of SEER data collection, introduces inherent selection and reporting biases that may have affected the generalizability of the findings. Given that this is the largest database that includes this tumor type, it is the best current understanding of PMCTs, but it is worth noting that as more data becomes available, the results could shift. Another key limitation of using the SEER database is the potential for confounding variables inherent in its retrospective design. Treatment modalities such as radiation, surgical resection, or chemotherapy cannot be interpreted as independently or collectively effective due to unmeasured confounders and lack of randomization. Also, the heterogeneity in the collection of data into the SEER database, such as institution-to-institution diagnostic criteria, follow-up, or treatment approaches, may have indirectly skewed the results. Furthermore, survival outcomes may have been further influenced by unmeasured variables such as referral patterns, supportive care differences, among others, all of which may have played a role in the variables measured in the paper. Next, the sample size for some of the variables, such as people who received radiotherapy, was very small, which may have played a role in the statistically significant findings noted. Another limitation revolves around the fact that since metastasis data only became consistently available in the SEER database starting in 2010 until 2021, analyses based on metastatic data would reflect only a subset of the overall study population, which spanned from 2000 to 2021. Thus, analyses done between metastatic burden and the overall study population may not be directly comparable due to various shifts in medicine, including diagnostic practices, staging accuracy, and treatment patterns during these 10 extra years. These potential shifts in medicine and PMCT care may influence the interpretation of any results and skew potential relationships investigated. The improved survival associated with surgery and chemotherapy in sarcomas may reflect selection bias, as those eligible for surgery and/or chemotherapy could have had a less severe manifestation of PMCTs to be eligible for said treatment options. Finally, because proportional hazards diagnostics were not formally assessed, interpretation of the Cox regression findings should be considered exploratory. Still, our results provide hypothesis-generating factors that contribute to primary cardiac tumor therapy and survival.

Conclusions

PMCTs remain an exceptionally rare cancer that is functionally challenging to manage. This study, using the SEER database, serves as a large analysis of this tumor type, offering insight into trends in incidence, demographic patterns, histologic distributions, and survival predictors. Prognosis I tumors, which were predominantly soft tissue sarcomas, were largely influenced by size and resectability, whereas Prognosis II tumors, which were often hematologic, were more strongly affected by metastases. Younger ages consistently conferred survival advantages for both groups, and select surgical interventions and chemotherapy helped improve outcomes in Prognosis I. Both groups were severely affected by metastases, especially to the lung and brain. Furthermore, PMCTs are very heterogeneous, where tumor biology not only affects prognosis, but also the ability to intervene surgically and the presence of metastatic spread.

While limited by the retrospective nature of the SEER database and incomplete areas of data collection, the results indicate an important principle of early diagnosis, screening, and tailored treatment strategies. Moving forward, prospective multicenter studies and genomic profiling may refine the classification of these tumor types, identify strong therapeutic targets, and ultimately improve outcomes for this rare but devastating group of tumors.