[Paper Review] Large Language Models-Enabled Digital Twins for Precision Medicine in Rare Gynecological Tumors

Jacqueline Lammert, Nicole Pfarr+17 more|arXiv (Cornell University)|Aug 31, 2024

Radiomics and Machine Learning in Medical ImagingMedicine3 citations

TL;DR

This study proposes a large language model (LLM)-enabled digital twin system for precision medicine in rare gynecological tumors (RGTs), integrating clinical, biomarker, and literature-derived data from 655 publications and 21 institutional cases to generate personalized treatment plans for metastatic uterine carcinosarcoma. The system identifies therapy options missed by conventional single-source analysis, demonstrating potential to improve outcomes through biology-based, patient-specific treatment matching.

ABSTRACT

Rare gynecological tumors (RGTs) present major clinical challenges due to their low incidence and heterogeneity. The lack of clear guidelines leads to suboptimal management and poor prognosis. Molecular tumor boards accelerate access to effective therapies by tailoring treatment based on biomarkers, beyond cancer type. Unstructured data that requires manual curation hinders efficient use of biomarker profiling for therapy matching. This study explores the use of large language models (LLMs) to construct digital twins for precision medicine in RGTs. Our proof-of-concept digital twin system integrates clinical and biomarker data from institutional and published cases (n=21) and literature-derived data (n=655 publications with n=404,265 patients) to create tailored treatment plans for metastatic uterine carcinosarcoma, identifying options potentially missed by traditional, single-source analysis. LLM-enabled digital twins efficiently model individual patient trajectories. Shifting to a biology-based rather than organ-based tumor definition enables personalized care that could advance RGT management and thus enhance patient outcomes.

Motivation & Objective

• To address the clinical challenge of rare gynecological tumors (RGTs) with low incidence and poor prognosis due to lack of standardized guidelines.
• To overcome limitations of manual curation of unstructured biomarker data in therapy matching.
• To develop a digital twin system leveraging large language models (LLMs) for personalized treatment planning in RGTs.
• To shift from organ-based to biology-based tumor classification for improved precision oncology.
• To evaluate the system's ability to identify effective, potentially missed therapeutic options using real-world and literature data.

Proposed method

• The digital twin system integrates clinical and biomarker data from 21 institutional cases of metastatic uterine carcinosarcoma (UCS).
• It ingests and synthesizes information from 655 published studies involving 404,265 patients using LLMs to extract and contextualize biomarker and treatment data.
• LLMs process unstructured text from publications to identify actionable biomarkers and corresponding therapies, such as HER2, ESR1, FR-alpha, and PRAME.
• The system generates individualized treatment plans by correlating patient-specific biomarker profiles with evidence from clinical and preclinical studies.
• It employs retrieval-augmented generation (RAG) techniques to ground LLM outputs in verified literature and clinical data.
• The system evaluates treatment options based on clinical benefit rates, response durations, and survival outcomes reported in trials and case reports.

Experimental results

Research questions

• RQ1 Can an LLM-powered digital twin system effectively synthesize heterogeneous clinical and literature-based data to generate personalized treatment plans for rare gynecological tumors?
• RQ2 What therapeutic options for metastatic uterine carcinosarcoma are identified by the system that may be missed by conventional, single-source analysis?
• RQ3 To what extent can biology-based tumor classification, enabled by LLMs, improve precision oncology in RGTs compared to organ-based classification?
• RQ4 How does the integration of real-world case data with large-scale literature data enhance the reliability and clinical relevance of digital twin-generated treatment recommendations?
• RQ5 Can the system identify durable responses to combination therapies, such as pembrolizumab + lenvatinib + letrozole, in ESR1-amplified UCS patients?

Key findings

• The digital twin identified a 36-month durable partial response in an ESR1-amplified, pMMR metastatic UCS patient treated with pembrolizumab, lenvatinib, and letrozole—a regimen potentially missed by standard approaches.
• For ER-positive UCS, the system highlighted a 43% clinical benefit rate with anastrozole in a phase II study, with median clinical benefit duration of 5.6 months.
• The system identified potential benefit from mirvetuximab soravtansine plus pembrolizumab in FR-alpha-positive UCS, supported by a phase II trial with 37.5% objective response rate in endometrial cancer patients.
• Preclinical evidence showed that HRD-positive UCS cell lines were significantly more sensitive to olaparib than HR-proficient lines, suggesting PARPi utility in HRD-positive cases.
• Sacituzumab govitecan demonstrated a 35% objective response rate and 5.7-month median progression-free survival in Trop2-positive recurrent endometrial cancer, including UCS patients.
• The system identified that elevated serum CA-125 at progression correlated with partial response, and postoperative elevation predicted poor survival, supporting its use as a dynamic biomarker.