Simple Three-Point Score Predicts Survival in Prostate Cancer Patients Receiving Radioligand Therapy

A three-point scoring system that requires nothing more than a blood count, a brief assessment of a patient’s physical condition, and a look at where the cancer has spread has now passed one of the toughest tests in clinical medicine: independent external validation. In a study published in the European Journal of Nuclear Medicine and Molecular Imaging, researchers led by Thomas Büttner of University Hospital Bonn and collaborators at University Hospital Frankfurt demonstrated that the Bellmunt Risk Score reliably predicts how long men with metastatic castration-resistant prostate cancer will survive after receiving lutetium-177 PSMA-617 radioligand therapy. The finding matters because this therapy, while transformative for many patients, produces strikingly uneven responses, and clinicians have long lacked a simple bedside tool to identify who stands to benefit most.

Metastatic castration-resistant prostate cancer, abbreviated mCRPC, is the lethal stage of prostate disease in which tumors continue to grow despite androgen-deprivation treatment. The arrival of [177Lu]Lu-PSMA-617, a radiopharmaceutical that delivers beta radiation directly to prostate cancer cells expressing the prostate-specific membrane antigen, changed the treatment landscape after the pivotal VISION phase III trial established it as a standard of care. Yet the survival benefit varies enormously from one patient to the next. Some men live for years after treatment begins, while others progress within months. Weighing the potential benefit against the burden of repeated nuclear medicine cycles demands an honest estimate of prognosis, and that is precisely what a validated scoring system can provide.

The Bellmunt Risk Score was originally developed for a completely different cancer: metastatic urothelial carcinoma. In its original setting, it stratified patients with advanced transitional cell carcinoma who had failed platinum-based chemotherapy using three binary variables that are universally available in any oncology clinic. Each patient receives one point for an Eastern Cooperative Oncology Group performance status greater than zero, meaning any degree of functional impairment; one point for anemia, defined as a hemoglobin concentration below 10 grams per deciliter; and one point for the presence of liver metastases. The resulting score ranges from zero to three, with higher scores indicating worse expected survival. Its elegance lies in its simplicity: no specialized software, no genomic sequencing, no additional laboratory panels.

In a recent primary analysis, the same research group showed that the score translates surprisingly well to prostate cancer patients undergoing PSMA-targeted radioligand therapy. But a demonstration in the development cohort is only the first step. Clinical prediction models are notorious for performing well in the population where they were built and then faltering elsewhere, a phenomenon statisticians call overfitting. External validation in a genuinely independent patient population, treated under different institutional pathways and evaluated by investigators uninvolved in the original data collection, is the mandatory test of generalizability. The new study was designed to meet exactly that standard.

The validation cohort comprised 243 men with confirmed PSMA-positive metastatic castration-resistant prostate cancer who received at least one cycle of [177Lu]Lu-PSMA-617 at University Hospital Frankfurt, a tertiary high-volume center with no patient overlap with the original development cohort. The score was calculated at baseline, within twenty-eight days before the first therapy cycle. The patient population was heavily pretreated: 9.9 percent received radioligand therapy as the first treatment line for metastatic disease, 22.6 percent as the second, 39.1 percent as the third, and 28.4 percent in even later lines. Over a median follow-up of 17.7 months, 111 deaths occurred, corresponding to 45.7 percent of the cohort, providing substantial statistical power for survival analysis.

The results revealed a strikingly clean stepwise pattern. Patients with a score of zero, meaning none of the three risk factors, had an estimated median overall survival of 24.0 months. Each additional point shaved months off that figure: 15.8 months for a score of one, 11.2 months for a score of two, and 8.7 months for the small group of seven patients with all three risk factors. The Kaplan-Meier comparison across the four groups was highly significant, with a log-rank p value below 0.0001. In univariable analysis, compared with the low-risk group, the hazard ratio for death was 1.63 for score one, 3.38 for score two, and 5.30 for score three, and these estimates aligned closely with the hazard ratios observed in the original development cohort.

Because more heavily pretreated patients naturally have worse survival, the researchers adjusted their multivariable analysis for the number of prior treatment lines. Even after this adjustment, the score remained an independent predictor of survival. The hazard ratio was 3.04 for score two and 4.76 for score three, both statistically significant, while score one showed a non-significant trend at 1.49. Given that only seven patients fell into the highest-risk category, the team performed a sensitivity analysis using Firth’s penalized likelihood Cox regression, a method designed for small sample sizes, which confirmed the stability of the estimate with a hazard ratio of 5.10.

Predictive accuracy was assessed with two complementary metrics. Harrell’s concordance index for the multivariable model combining the score and treatment line reached 0.659, comparable to the performance achieved in the development cohort. Time-dependent receiver operating characteristic analysis, which evaluates discrimination at specific time points, yielded area-under-the-curve values of 69.3 percent for six-month survival, 71.3 percent for twelve-month survival, and 67.5 percent for twenty-four-month survival. Calibration for twelve-month survival, examined with two hundred bootstrap resamples, was excellent, with a mean absolute error of just 0.027 and a maximum absolute error of 0.059, meaning the predicted and observed survival probabilities were nearly identical.

The authors are candid about the study’s limitations. The retrospective design and the absence of a central review of imaging data introduce potential bias, since liver metastases were identified through local institutional interpretation of PSMA PET/CT scans. The multivariable model deliberately adjusted only for treatment lines, omitting established prognostic markers such as baseline PSA, lactate dehydrogenase, alkaline phosphatase, and quantitative metastatic burden, which leaves open the possibility of residual confounding. The researchers argue that this restraint was intentional: adding complex variables would risk overfitting and would compromise the score’s identity as a strictly clinical, standalone, bedside-ready tool. The very small size of the score-three subgroup also demands cautious interpretation of its survival estimates.

What emerges is a pragmatic message for oncology. More sophisticated nomograms that incorporate quantitative PSMA-PET volume metrics or extended biochemical panels may theoretically offer superior accuracy, but they require specialized software or additional laboratory costs and are often impractical in routine care. The Bellmunt Risk Score, by contrast, can be computed in seconds from information already gathered at every oncology visit. As indications for [177Lu]Lu-PSMA-617 expand into earlier treatment lines following trials such as ENZA-p, which tested the radioligand in combination with enzalutamide, the need for rapid risk stratification will only grow. The researchers emphasize that the score is intended to complement, not replace, comprehensive clinical and imaging assessment, serving primarily to guide patient expectations and support shared decision-making. With external validation now complete, a tool born in urothelial carcinoma has earned a place in the prostate cancer clinic, offering clinicians and patients alike a clear-eyed, evidence-based starting point for one of the most consequential conversations in modern cancer care.

Subject of Research: External validation of the Bellmunt Risk Score for predicting overall survival in metastatic castration-resistant prostate cancer patients undergoing lutetium-177 PSMA-617 radioligand therapy

Article Title: External validation of the bellmunt risk score for survival prediction in mCRPC patients undergoing [177Lu]Lu-PSMA-617 therapy

Article References: Büttner, T., Wenzel, M., Mandel, P., Schmitt, P., Hoffmann, C., Marinova, M., Essler, M., Chun, F. K. H., Ritter, M., Groener, D., & Krausewitz, P. (2026). External validation of the bellmunt risk score for survival prediction in mCRPC patients undergoing [177Lu]Lu-PSMA-617 therapy. European Journal of Nuclear Medicine and Molecular Imaging. https://doi.org/10.1007/s00259-026-08169-7

Image Credits: AI Generated

DOI: 10.1007/s00259-026-08169-7

Keywords: metastatic castration-resistant prostate cancer, Bellmunt Risk Score, Lu-PSMA-617, radioligand therapy, survival prediction, prognostic markers, external validation, PSMA PET, overall survival, Kaplan-Meier analysis, Cox regression, nuclear medicine