Comprehensive genomic profiling is routinely used in metastatic prostate cancer to identify alterations that may inform biomarker-directed treatment. Although several individual genes have known prognostic significance, a genomic classification accounting for combinations of tumour alterations has not previously been established for patients with metastatic hormone-sensitive prostate cancer.
A large retrospective study has now developed and externally validated a DNA-based classification that separates patients with metastatic prostate cancer into favorable, intermediate, and unfavorable prognostic groups according to overall survival.
The classification, called Somatic Tumor Risk Assessment for Overall Survival–Prostate, or STRATOS-P, uses information generated through comprehensive genomic profiling already performed as part of clinical care.
The article, titled “Genomic Classification to Predict Survival in Metastatic Prostate Cancer: Development of Somatic Tumor Risk Assessment for Overall Survival-Prostate,” was published in JCO Precision Oncology on July 15, 2026.
Authors: Martin W. Schoen, Jiannong Li, Sihang Zeng, Heena Desai, Ryan Hausler, Candace L. Haroldsen, Lukas Owens, Luca F. Valle, Ruth B. Etzioni, Timothy R. Rebbeck, Brent S. Rose, Michael J. Kelley, R. Bruce Montgomery, Nicholas G. Nickols, Matthew B. Rettig, Kosj Yamoah, Kara N. Maxwell, and Isla P. Garraway.
Study Design and Patient Population
The investigators conducted a retrospective, cross-sectional clinicogenomic study using data from US veterans with metastatic prostate cancer who underwent comprehensive genomic profiling through the Veterans Health Administration National Precision Oncology Program.
A total of 7,201 veterans with metastatic prostate cancer and available comprehensive genomic profiling were identified. The discovery cohort included 2,484 veterans with synchronous metastatic hormone-sensitive prostate cancer and tissue-based genomic profiling. Synchronous metastatic disease was defined as a diagnosis of prostate cancer less than 12 months before the diagnosis of metastasis.
The cohort was randomly divided in a 2:1 ratio into a training set of 1,738 patients and a testing set of 746 patients. The primary outcome was overall survival from the diagnosis of metastatic disease. The median age at diagnosis of synchronous metastatic hormone-sensitive prostate cancer was 72 years. Of the 2,484 patients, 765, or 31%, self-identified as Black. Tumour tissue was obtained from the prostate in 1,951 patients, or 78.5%, and from a metastatic site in 533 patients, or 21.5%.
Comprehensive genomic profiling was performed within six months of metastatic diagnosis in 1,389 patients, representing 55.9% of the cohort. Median overall survival was 59.4 months in the overall discovery cohort and 39.0 months among those whose sequencing was performed within six months of metastatic diagnosis.
Development of the STRATOS-P Classification
The investigators evaluated oncogenic somatic tumour DNA alterations, including short variants, copy-number alterations, and rearrangements.
Genes were considered for inclusion when alterations were present in more than 2% of either prostate or metastatic tissue samples. Associations between individual genes and overall survival were assessed using a multivariable Cox proportional hazards model that included age, prostate-specific antigen level, and Charlson Comorbidity Index.
Sixteen genes were identified as significantly associated with overall survival:
- TP53, PTEN, RB1, CCND1, CDK12, BRCA2, RAD21, MYC, FGFR1, FGF3, FGF4, FGF19, PRKC1, LYN, AR, and SPOP.
These included tumour suppressor genes, genes involved in cell-cycle regulation, DNA repair, cell growth, and androgen signalling.
At least one oncogenic alteration involving the 16 selected genes was identified in 1,741 tumours, or 70.1% of the discovery cohort.
Two or more alterations were present in 694 patients, or 28%, while 289 patients, or 11.6%, had at least three alterations. Approximately 30% had no alteration involving any of the 16 genes.
Genomic Co-Alterations and Risk Categories
The investigators evaluated genomic co-alterations and patterns of mutual exclusivity.
The most common co-alteration involved TP53 and PTEN. Other observed patterns included co-amplifications of chromosome 8q involving RAD21, LYN, and MYC, and chromosome 11q13.3 alterations involving CCND1, FGF19, FGF3, and FGF4.
The genes were then organised into prognostic categories according to their hazard ratios, genomic clustering, colocalisation, and expert clinical review.
Genes with a hazard ratio below 1, or the absence of an alteration, were assigned to the favorable-risk category. Alterations with hazard ratios between 1 and 2, together with relevant co-alterations, were included in the intermediate-risk group. The unfavorable-risk group included alterations with hazard ratios above 2 and the frequent co-alteration of TP53 and PTEN.
Patients with more than one alteration were classified according to the most adverse alteration present.
Within the discovery cohort:
- 947 patients, or 38%, were classified as favorable risk
- 1,180 patients, or 48%, were classified as intermediate risk
- 357 patients, or 14%, were classified as unfavorable risk
The TP53 and PTEN co-alteration was present in 234 patients, representing 9.5% of the cohort.
Overall Survival According to STRATOS-P
STRATOS-P separated patients into groups with different overall survival outcomes.
Median overall survival was:
- 87 months in the favorable-risk group
- 52 months in the intermediate-risk group
- 31 months in the unfavorable-risk group
Among patients whose genomic profiling was performed within six months of metastatic diagnosis, median overall survival was not reached in the favorable-risk group. Median overall survival was 36 months in the intermediate-risk group and 24 months in the unfavorable-risk group.
In the overall multivariable model, intermediate-risk classification was associated with increased mortality compared with favorable-risk classification, with an adjusted hazard ratio of 1.54. Unfavorable-risk classification was associated with an adjusted hazard ratio for mortality of 2.37 compared with the favorable-risk group.
The average time-dependent area under the curve was 0.83. In the training cohort, the adjusted hazard ratios for mortality were 1.55 for the intermediate-risk group and 2.30 for the unfavorable-risk group. In the held-out testing cohort, the corresponding adjusted hazard ratios were 1.59 and 2.73, respectively.
External Validation in the MSK-IMPACT Cohort
STRATOS-P was externally evaluated in 1,039 nonveteran patients with synchronous metastatic hormone-sensitive prostate cancer included in the Memorial Sloan Kettering–Integrated Mutation Profiling of Actionable Cancer Targets dataset. Median overall survival in this cohort was 89.8 months.
Of the 1,039 patients:
- 457, or 44.0%, were classified as favorable risk
- 453, or 43.6%, were classified as intermediate risk
- 129, or 12.4%, were classified as unfavorable risk
Compared with the favorable-risk group, the adjusted hazard ratio for mortality was 2.45 in the intermediate-risk group and 4.37 in the unfavorable-risk group. The average time-dependent area under the curve between 12 and 60 months was 0.79.
Findings Across Disease Settings and Sample Types
The investigators also evaluated the genomic classification across other metastatic presentations, hormonal states, and sequencing sample types. STRATOS-P remained associated with mortality in patients with synchronous or metachronous metastatic disease, hormone-sensitive or castration-resistant prostate cancer, and tissue-based or liquid-biopsy profiling.
Among 384 veterans with synchronous metastatic hormone-sensitive prostate cancer who underwent liquid-biopsy profiling, the adjusted hazard ratio for mortality was 2.23 in the intermediate-risk group and 5.13 in the unfavorable-risk group compared with the favorable-risk group.
The time-dependent area under the curve at 12 months was 0.71 in this cohort. The authors noted that the performance of STRATOS-P in liquid-biopsy samples was important because tumour tissue may be unavailable or difficult to obtain in some patients.
Study Limitations
The investigators acknowledged several limitations. Because the study was retrospective, the findings may have been affected by unmeasured confounding. Veterans who underwent comprehensive genomic profiling may also have differed from other patients with metastatic hormone-sensitive prostate cancer, creating potential selection bias.
Many patients underwent genomic profiling several months after their metastatic diagnosis because comprehensive genomic profiling had historically been used primarily to inform treatment in metastatic castration-resistant prostate cancer.
The investigators addressed potential immortal-time and left-truncation biases through an analysis restricted to patients whose sequencing was performed within six months of metastatic diagnosis. This analysis produced similar results.
Patients with metastatic disease were identified using a natural language-processing algorithm applied to Veterans Health Administration records, which may have included some patients with regional metastases.
The clinical model did not include features such as Gleason score and visceral metastases. A sensitivity analysis incorporating tumour volume improved model performance.
The authors additionally noted that tumour tissue may not always be available and that comprehensive genomic profiling may fail in some patients, limiting the applicability of genomic prognostication.
Conclusion
The STRATOS-P classification uses somatic tumour DNA alterations identified through comprehensive genomic profiling to categorise patients with metastatic prostate cancer into favorable, intermediate, and unfavorable prognostic groups.
The classification was developed and validated in patients with synchronous metastatic hormone-sensitive prostate cancer and remained associated with overall survival in an external nonveteran cohort and across additional metastatic disease settings and sample types.
Because STRATOS-P uses genomic information obtained through contemporary clinical practice, it does not require additional tissue procurement or sequencing. The authors concluded that the classification could contribute to shared decision making and support prospective studies evaluating genomic risk together with clinical variables to optimise treatment strategies.
The full article is available in JCO Precision Oncology.
Find more videos on OncoDaily YouTube TV.