Predicting immunotherapy outcomes in advanced melanoma: Clinical, radiological, and pathological approaches

Abstract

Immune checkpoint inhibition (ICI) has significantly improved survival for advanced melanoma patients, yet nearly half do not respond to treatment. Identifying predictive biomarkers remains a challenge, as no single marker reliably predicts outcomes. This thesis explores clinical, radiological, and pathological predictors of ICI response to improve patient selection. Part I - Clinical predictors The time to first distant recurrence (TFDR) may indicate immune surveillance effectiveness. Analysis of 3,462 advanced melanoma patients treated with either ICI or BRAF(/MEK) inhibitors showed that longer TFDR correlated with improved progression-free survival (PFS) and overall survival (OS). Patients with a TFDR >5 years had a median OS of 37.3 months compared to 25.0 months for those with TFDR <2 years (p=0.014). A prediction model using clinical data and primary melanoma characteristics was developed in a nationwide cohort of 3,525 patients. The model, incorporating 10 clinical variables, demonstrated good discrimination between response groups (AUROC 0.657). Internal-external validation across different geographic regions confirmed its generalizability, supporting its potential for clinical implementation. Part II - Radiological predictors Baseline imaging features may provide insights into tumor response heterogeneity. A systematic review of 119 studies (15,580 patients) identified several imaging predictors of ICI outcomes, including tumor burden, subcutaneous adipose tissue, muscle mass, and liver metastases. However, no single imaging biomarker reliably predicted treatment response. Next, radiomics, the extraction of imaging features using machine learning, were investigated. Analysis of 620 patients showed that a radiomics-based model (AUROC 0.607) did not outperform a clinical model (AUROC 0.646), nor did their combination improve predictive performance. Overlap between radiomics-derived information and clinical factors, such as lactate dehydrogenase (LDH) levels and number of affected organs, likely explains the lack of added value. This study highlights the necessity of evaluating radiomics against clinical models before implementation. Part III - Tumor-Infiltrating Lymphocytes (TILs) as predictors TILs have historically been associated with better melanoma prognosis. In this thesis, TILs were assessed using three different scoring systems. While primary melanoma TILs did not correlate with response, the presence of TILs in pre-treatment metastatic lesions was associated with improved response, PFS, and OS. Patients with the highest TILs scores had a median OS of 49.4 months versus 19.5 months for those with absent TILs (p=0.003). These findings suggest metastatic TIL assessment could be a valuable biomarker for ICI response. Furthermore, the research in this thesis also explored whether TILs were associated with severe immune-related adverse events (irAEs). No significant correlation was found between TIL presence and the likelihood or timing of irAEs. The findings in this thesis support a multimodal approach to prediction, integrating clinical, pathological, and imaging data to refine patient selection and improve ICI treatment outcomes.