Building a data foundation: How MD Anderson and Palantir are partnering to accelerate research and improve patient care.

Abstract

e18077 Background: MD Anderson (MDA) strives to maximize learning from the > 44,000 patients seen annually. This has historically been challenging due to the diverse, often unstructured, and siloed nature of patient data. To begin addressing these challenges, MDA constructed an institutional data warehouse and developed Natural Language Processing (NLP) capabilities. This allowed technical users to extract specific cohorts but proved challenging for many non-technical users. In 2018, MDA conducted a proof of concept with Palantir Technologies to integrate disparate data sources into a single cloud-based platform enabling both technical and non-technical users to directly access and analyze data. Methods: Through a highly iterative, collaborative process between clinical/research users, IT experts and Palantir engineers, 150 deidentified datasets were integrated from the existing data warehouse, NLP, molecular reports, public data and manually curated data, representing data from > 1.5 million patients. Data integration was validated and feedback from clinical users gathered. Results: In addition to centralizing data, the platform was found to: (1) Make data accessible to non-technical users by mapping data to real-world concepts, (2) Create reports backed by live data, with clear visibility into inputs, analysis and history, (3) Create templates for common analytical workflows (e.g. KM curve), and (4) Support increased collaboration through transparent, secure access controls. This resulted in dramatic acceleration of cohort identification, data extraction, and outcomes analyses. Conclusions: Integrating siloed data exposed gaps and errors arising from the unstructured nature of medical data and use of NLP. In response, we are piloting a combination of (1) Purposeful structured data capture at the time of care, (2) Use of logic to suggest commonly uncaptured but critical patient events such as progression, and (3) Auditing functionality to allow users to manually fill in gaps and correct data in a transparent manner. Expanding the data foundation unlocks the potential for more advanced research, fueling scientific discovery, while enhancing the quality and safety of patient care.