# STAR (stroma-tumor AI risk) assessment: association of AI-derived tumor-stroma proportion with patient survival provides added prognostic value beyond KELIM in epithelial ovarian cancer

## TL;DR

This study shows that AI-based tumor-stroma proportion analysis can predict survival in ovarian cancer, offering a pre-treatment prognostic tool that complements existing methods.

## Contribution

AI-derived tumor-stroma proportion (TSPauto) is shown to be a novel, pre-treatment prognostic biomarker for epithelial ovarian cancer.

## Key findings

- High TSPauto is significantly associated with poor survival in epithelial ovarian cancer (HR 1.99, p = 0.02).
- TSPauto and TSPmanual assessments show high agreement (94%, Cohen’s Kappa 0.89, p<0.001).
- TSPauto provides added prognostic value beyond the KELIM score, which only predicts platinum resistance.

## Abstract

There remains a critical need for prognostic biomarkers of treatment response in epithelial ovarian cancer (EOC). The KELIM score, derived from the rate of CA-125 elimination during the first 100 days of treatment, is a clinically available biomarker of treatment response to platinum-based chemotherapy, its utility is limited by the need for post-treatment data. Tumor–stroma proportion (TSP) has emerged as a prognostic biomarker across several malignancies. Studies from our group have shown that high TSP (≥50% stroma content assessed by pathologist evaluation, TSPmanual) is associated with platinum resistance and poor survival in EOC at diagnosis and before treatment.

We compared the prognostic value of TSP and KELIM by analyzing manual pathologist (TSPmanual) and artificial intelligence–derived assessments (TSPauto) on digitized images from a cohort of EOC specimens.

In this cohort, we showed the prognostic significance of TSPmanual, confirming prior findings. Furthermore, TSPauto and TSPmanual assessments were highly concordant (94% agreement, Cohen’s Kappa 0.89, p<0.001), providing a highly reproducible, automated approach. Unlike KELIM, which was only associated with platinum resistance, high TSPauto was significantly associated with poor survival (HR 1.99, p = 0.02).

These findings support AI-derived TSP as a pre-treatment prognostic biomarker for EOC that complements KELIM.