Clinical and radiomics parameter prognostication in metastatic uveal melanoma patients treated with hepatic arterial infusion chemotherapy.

INTRODUCTION

Metastatic uveal melanoma (MUM) has a poor prognosis, but hepatic arterial infusion chemotherapy (HAIC) may improve outcomes in patients with hepatic metastases. To identify reliable prognostic factors for patient stratification and treatment allocation, we analyzed the clinical and imaging data from a large single-center cohort using machine learning (ML) models.

METHODS

Pre- and post-first treatment clinical data of 235 patients with MUM treated with HAIC between 2009 and 2019 were retrospectively analyzed using Cox regression to identify prognostic factors for overall survival (OS) and time to change treatment strategy (TTCS). Furthermore, ML models were trained on clinical and computed tomography (CT) data for endpoint prediction.

RESULTS

Pre-treatment multi-variate analysis identified elevated lactate dehydrogenase (LDH) (OS: 6.5 vs. 16.4 months, hazard ratio [HR]) = 1.87, P = 0.006) and gamma-glutamyl transpeptidase (GGT) (OS: 7.6 vs. 16.4 months, HR = 1.67, P = 0.012) as prognostic factors for inferior OS. Decreased albumin (TTCS: 1.3 vs. 6.1 months, HR = 6.26, P < 0.001) and elevated LDH (TTCS: 2.9 vs. 7.6 months, HR = 1.72, P = 0.011) and alanine aminotransferase (ALT) (TTCS: 3.7 vs. 6.4 months, HR = 1.65, P = 0.004) predicted shorter TTCS. Scoring enhanced the power of the prognosticators for OS and TTCS. Post-first treatment multi-variate analysis emphasized the importance of inflammation management and liver protection. ML models incorporating radiomics features from baseline CT imaging were not superior to models based on pre-treatment clinical data alone.

CONCLUSION

We identified independent but synergistic prognostic factors for outcome stratification to guide treatment decisions and optimize patient management. ML-based radiomics features did not significantly enhance prognostic performance.