SROTAS IQ: AI-Based Clinical Trial Matching Validation in Breast Cancer

Srotas Health has published a medRxiv preprint validating SROTAS IQ, our AI-based clinical trial matching platform, in breast cancer eligibility screening.

The study evaluates a custom fine-tuned large language model against expert consensus. Ten breast cancer trials were selected across different treatment settings and molecular subtypes, and synthetic patient summaries were generated to represent realistic eligibility scenarios. Two independent oncologists assessed eligibility to establish the reference standard for the evaluation.

Key Results

• 90% or greater accuracy in 5 of 10 trials, including two trials above 97% accuracy.
• 88% overall eligibility classification accuracy across 150 patient-trial evaluations.
• 132 of 150 decisions matched expert consensus, with 18 mismatches.
• Under 0.5 minutes per patient for every SROTAS IQ assessment.
• Good inter-rater reliability between oncologists before consensus discussion: Cohen's kappa = 0.742.

Figure 1. Trial-level accuracy. SROTAS IQ reached the 90% threshold in Trials 1, 2, 4, 7, and 8. The lower-performing trials generally involved more complex eligibility structures, nested criteria, or higher criteria counts.

Figure 2. Overall classification summary. Across 150 patient-trial decisions, SROTAS IQ matched expert consensus in 132 decisions and missed 18, resulting in 88% overall classification accuracy.

Figure 3. SROTAS IQ time-to-verdict. SROTAS IQ completed each per-patient eligibility assessment in 0.40 to 0.50 minutes across the ten trial evaluations.

Publication Details

• Title: SROTAS IQ: An AI-Based Clinical Trial Matching Platform: A Validation Study in Breast Cancer
• Authors: Samuel McInerney, Heeba Gurku, Ramji Balasubramanian, Vikram Parimi, Suman Bhaskaran, Karthik Sekaran
• Repository: medRxiv
• DOI: 10.1101/2025.08.13.25333342
• PDF: Read the full preprint

Why It Matters

Clinical trial matching is often slowed by manual review of complex eligibility criteria and fragmented patient information. This validation study focuses on whether AI can support consistent, scalable screening while preserving the clinical reasoning needed for oncology trial eligibility assessment.

SROTAS IQ was assessed using standard classification metrics and time-to-verdict comparisons, positioning the work around both matching accuracy and operational efficiency for clinical research teams.

Preprint Notice

This article is a preprint hosted by medRxiv. It reports new medical research that has not yet been certified by peer review and should not be used to guide clinical practice.