Flatiron Health at ISPOR 2024

Prevalence and incidence of ROS1 and NTRK rearrangements in real-world patients with advanced non-small cell lung cancer in the United States

Author affiliations: Bristol-Myers Squibb, University of North Carolina, Chapel Hill, University of Colorado Cancer Center, Flatiron Health

Recent advances in targeted therapies benefit patients with advanced non-small cell lung cancer, particularly those with ROS1 (ROS1+) and NTRK (NTRK+) rearrangements. However, limited biomarker testing data hinders understanding of the burden of disease and its economic implications. This study addresses this gap by examining ROS1 and NTRK biomarker prevalence in the US NSCLC population through comprehensive genomic profiling and patient characteristics among biomarker positive cohorts, informing personalized treatment and healthcare policy for improved outcomes.

Racial and ethnic inequities in endometrial cancer survival from 2017 to 2022: Assessing the influence of social determinants of health

Author affiliations: Flatiron Health

Amidst the rising incidence of endometrial cancer, concerns are mounting about exacerbated inequities during the COVID-19 pandemic due to care delays. This study examines recent endometrial cancer survival, focusing on racial inequities and the role of social determinants of health (SDOH). It uncovers persistent survival gaps, particularly among Black patients compared to their White counterparts, both before and after the pandemic, and highlights the significant influence of clinical and SDOH factors. Addressing the drivers of later and more aggressive disease in Black patients is essential for improving endometrial cancer outcomes and advancing health equity.

Oncology practice setting classification in the US: Considerations for real world evidence

Author affiliations: Flatiron Health

In oncology real-world evidence (RWE) studies, variations in treatment patterns and patient characteristics often complicate the analysis due to differences in practice. Furthermore, the lack of a standardized classification methodology for oncology practices in the US poses challenges in addressing these complexities effectively.

This study evaluates existing classification schemes used in RWE studies and summarizes practice characteristics of the Flatiron Health Research Database. Similar to academic sites, some community oncology practices have high patient volume and conduct clinical trials. Researchers should not assume community oncology practices are homogeneous.

Performance assessment of a disease-agnostic treatment sequencing heuristic for deriving line of therapy in a real-world, rare, multi-tumor cohort across tumor mutational burden status

Author affiliations: Flatiron Health

Deriving clinically-appropriate treatment sequences across different cancers poses challenges in retrospective studies. This study introduces a new heuristic approach to tackle this issue. By identifying treatment sequences across a spectrum of cancer types, it enables a deeper understanding of treatment patterns in multi-tumor analyses, including biomarker-defined cohorts.

Evaluating real-world response-based endpoint in the clinical trial settings among aNSCLC, mCRC and mBC patients

Author affiliations: Flatiron Health

Understanding real-world response rates (rwRR) is crucial for comprehending clinical outcomes outside of oncology clinical trials. Using machine learning, this study examines rwRR in lung, colon, and breast cancer patient cohorts aligned with clinical trials. By providing significant insights into treatment response rates beyond clinical trials, the research aims to deepen our understanding of the relationship between endpoints in trials and real-world data.

Assessing the potential value of the Flatiron Health commercial mortality data combined with publicly available death data for non-oncology research

Author affiliations: Flatiron Health

Combining mortality data from EHRs, commercial databases, and the US Social Security Death Index (SSDI) offers reliable measures in oncology research. This study explores using Flatiron Health's commercial mortality database and SSDI data alone as an alternative for research without EHR access. The findings have significant implications for mortality research beyond oncology, providing a valuable alternative for mortality assessment in settings without EHR data. With viable levels of sensitivity and specificity, this approach provides a near-real time option for mortality outcomes in diverse research contexts.

Using large language models to extract PD-L1 testing details from electronic health records

Author affiliations: Flatiron Health

This study assesses the effectiveness of large language models (LLMs) to extract key PD-L1 details (test date, percent staining, etc) from a diverse medical record database, spanning numerous different cancer types. The research demonstrated that LLMs were able to perform this complicated task and were more effective in extracting these clinical details after being fine-tuned using human labeled examples. The work showcases the potential of these models to expedite access to crucial data from the EHR to inform treatment decisions and facilitate research. It also highlights the importance of high quality labeled data to improve and understand performance of these LLMs.