The first substudy using the multi-omics and multi-modal data generated prospectively within SPRINTR-Real
We are building and validating an AI tool that reads a single routine H&E biopsy slide and predicts two treatment-defining biomarkers: PTEN status and homologous recombination repair (HRR) status.
Today, these biomarkers are identified by sequencing, a process that is slow and costly. By training an AI model on biopsies from 1,000 men (600 in Sweden, 400 in Denmark) with matched H&E, PTEN immunohistochemistry, and genomic profiling (HRR), we will build an algorithm that can serve as a first-line screen: only the flagged tumors undergo genomic testing. The goal is to reduce the number of biopsies requiring genetic analysis by at least 50%, while maintaining ≥99.5% certainty that no clinically relevant PTEN loss or HRR deficiency is missed.
The validation can be performed within the continously growing SPRINTR-Real population, with the goal to use the algorithm to actively select patients for biomarker-driven in SPRINTR-TRIAL - turning routine pathology into a precision medicine tool.
