**Background:** Uveal melanoma (UM) is the most common primary intraocular malignancy in adults, with a 15-year metastatic rate of 32–45% despite successful local treatment. Current prognostic methods (e.g., tumor biopsy for gene expression or BAP-1 status) are invasive, costly, and not universally available. Liquid biopsies from peripheral blood offer a less invasive alternative, but most existing tests only detect metastases when they are already radiologically visible. This study aimed to develop a prognostic test (serUM-Px) based on a single blood sample obtained at diagnosis that could predict long-term metastatic risk.
**Methods:** 83 patients with posterior UM (choroidal or ciliary body) diagnosed between 1996 and 2000 were included. Exclusion criteria were detectable metastases at baseline, inability to consent, or recurrent melanoma. Peripheral serum samples were collected at diagnosis and stored at -80°C. In step one, pooled serum from 10 patients who later developed metastases and 12 who did not was screened for 84 cancer-related proteins using a Proteome Profiler array. Five candidates (leptin, osteopontin, progranulin, tenascin C, DLL1) were selected based on differential expression and non-overlapping functions. In step two, these five proteins were quantified by ELISA in a training cohort of 18 patients. Receiver operating characteristic (ROC) analysis identified leptin and osteopontin as meeting the minimum requirement (lower bound of AUC 95% CI >0.5). Optimal cutoffs were: leptin <5 ng/mL for males and <15 ng/mL for females (low levels indicating risk), and osteopontin >5 ng/mL (high levels indicating risk). The serUM-Px test classified patients into low (both markers favorable), intermediate (one unfavorable), or high (both unfavorable) metastatic risk. In step three, the test was validated in an independent cohort of 56 patients (9 of 65 originally randomized were excluded for poor sample quality). Follow-up was complete for all patients, with median follow-up for survivors of 22.7 years (IQR 22.3–23.0). Metastasis-free and overall survival were analyzed by Kaplan-Meier and log-rank tests, and cumulative incidence of UM-related mortality was assessed with competing risks analysis (Gray's test). Multivariate Cox regression was used to evaluate independent predictors.
**Key Results:** In the validation cohort, patients in the low (n=20), intermediate (n=24), and high (n=12) risk categories had 10-year metastasis-free survival of 95%, 68%, and 49%, respectively. Metastasis-free survival decreased significantly with increasing risk category (log-rank P for trend = 0.007). Overall survival also worsened (P = 0.018). In competing risks analysis, cumulative incidence of UM-related mortality was significantly higher with increasing risk category (Gray's test P = 0.029), while death from other causes did not differ (P = 0.74). In multivariate Cox regression adjusting for tumor diameter and patient sex, serUM-Px remained an independent predictor of metastasis (hazard ratio 3.2 per increasing risk category, 95% CI 1.5–6.9, P = 0.003). Among 12 patients who underwent enucleation, all four with low nuclear BAP-1 expression (a known poor prognostic marker) were in the high serUM-Px category, with moderate inter-method agreement (κ = 0.53, P = 0.038). Primary tumor osteopontin expression did not correlate with serum levels or prognosis, suggesting the serum osteopontin may originate from micrometastatic niches rather than the primary tumor. Serum leptin levels correlated with BMI (R² = 0.18, P = 0.041), but BMI itself was not a significant predictor of metastasis.
**Clinical Implications:** The serUM-Px test provides a minimally invasive, inexpensive prognostic tool that can be performed at the time of UM diagnosis using a single peripheral blood sample. It stratifies patients into three risk categories with markedly different long-term outcomes (10-year metastasis-free survival ranging from 95% to 49%), which could guide personalized follow-up intervals and selection for adjuvant clinical trials. Unlike most liquid biopsies that only detect metastases when radiologically visible, serUM-Px predicts risk years in advance. The test's independence from tumor size and patient sex suggests it adds complementary prognostic information. Limitations include the relatively small sample size (56 patients in validation), long-term storage of samples (over 20 years at -80°C), and lack of correlation with BAP-1 expression in a small subset. Prospective validation in larger, contemporary cohorts is needed before clinical implementation.