分子肿瘤登记库——真实世界数据（RWD）的学习系统

Background: Genomic data is a key component for drug discovery and precision medicine, an innovative approach for tailoring treatment and prevention that is useful for clinicians and researchers. Patient care now often involves comprehensive genomic profiling through certified clinical lab vendors. The clinical interpretation of molecular alterations is at the heart of providing the value of precision medicine and yet this Real-World Data is not always seamlessly integrated along with Real World Evidence (RWE) in electronic health records (EHR). We present the challenges and lessons learned in developing an end-to-end secure platform, along with the opportunities that a molecular registry opens for RWE in clinical care and research. Design: The molecular tumor registry IMPACT (Individual Molecular Registry of Patients for Accelerated Clinical and Translational Medicine) is built within a secure data enclave, “Soteria”, a HIPAA compliant high-performance compute (HPC) cluster. The software tools include an R package of JSON parsers to extract NGS and other results and store them within a DuckDB. An interactive RShiny web application is deployed on a secure VM within the secure enclave. This provides a real-time, de-identified genomics data feed of test metadata including tumor site, test types, genes, variants, and several associated elements for every case. Optimized data pipelines have been constructed for bulk NGS files, for comprehensive analysis. AI tools have been deployed for connecting the National Clinical Trial registry and associated FDA-approved targeted drugs. Results: The data is accelerating in size as more tests are being ordered by clinicians and new tests are developed by the vendors. Common challenges are a) time devoted to advance the process due to legal hurdles and inflexible contracts, b) lack of common nomenclature and data formats to compare and interpret results across vendors, c) constant change of technology and tests and d) Integration with the EHR. The registry opens up opportunities - 1) to normalize and harmonize results across vendors, 2) enable genomically informed clinical trials and patient recruitment, 3) facilitate a molecular tumor board, 4) enable cohort discovery and integrative analysis with RWE data from EHR, 5) support biomedical trainee education 6) provide opportunity to deploy LLM (Large Language Models)-powered tools for data investigation and deep learning models for research. Conclusions: The learning system, IMPACT, enables bringing molecular sequencing and clinical data together in a secure enclave to learn from RWD from patients utilizing power discovery approaches. Real-world insights from visualizing the pan-cancer results from both a high and granular level is an invaluable resource for accelerating research and treatment strategies across cancers, the epitome of translational research.