A biopharma company used NorstellaLinQ to identify hard-to-find rare pediatric cancer patients, reaching beyond the limits of claims data to drive real-time demand for its rare oncology therapy. The company:
One biopharma company develops therapies for rare pediatric cancers. Its lead product targets a small patient population that requires confirmation of a biomarker that is largely invisible in standard claims data.
Like many rare disease companies, the client historically relied on claims data for targeting and patient identification. But claims data couldn’t capture biomarker status, reliably surface early-stage patients, or deliver the near-real-time signal needed to time field outreach effectively. By using LLMs, the team could identify and predict a treatment decision. In a disease where a physician may see this type of patient only once every six months, and where that patient may not be top of mind, the timing and precision of outreach is everything. It’s also important for these patients and their families to be made aware of all their treatment options.
The company needed a better way to identify patients it couldn’t otherwise find, get these treatments to patients who needed it, and to get its field team in front of the right physicians at exactly the right moment.
The company partnered with Norstella, using NorstellaLinQ, the linked real-world data foundation for biopharma, to build a rigorous, scalable patient identification algorithm combining curated data, large language models, and unstructured EHR data.
The team started broad, casting a wide net to surface all potential patients from unstructured clinical notes, then systematically refined the cohort: identifying second-line and relapsing patients, then narrowing further to those with confirmed biomarker positivity. Age-based cohorts gave field reps layered signals to prioritize follow-through.
The team ran two parallel approaches: a keyword-mentions-based method as an interim solution, and a more robust extracted data approach developed in tandem. After rigorous iteration with the client’s clinical team and Norstella’s analytics team, including alignment on histology grading, true positive definitions, and cross-functional review, both approaches were retained for their complementary value.
The resulting alerts were integrated directly into the client’s omni-channel engine, fueling both personal and non-personal promotion and feeding its next-best-action platform.
The results exceeded expectations. The team was able to identify a clear subset of patients derived directly from Norstella alerts — patients they would not have found through claims data alone, and who were found far more quickly. More importantly, it made patients aware of a treatment option.
The company exceeded ROI expectations quickly because of the speed and specificity of the data. Where claims data provides high-level directional guidance, NorstellaLinQ delivered biomarker-level precision and near-real-time clinical encounter data. When an alert came in, it became the company’s field team’s highest priority for the week.
Beyond patient identification, the alerts enabled relationship building: Reps gained access to physician offices they would not otherwise have reached, simply by getting there at the right time. Those alerts now feed seamlessly into the company’s omni-channel engine, ensuring that every potential patient encounter, whether through a rep visit or a digital touchpoint, is timed and informed by the highest-fidelity data available.
Norstella’s data lets the company time its outreach across both personal and non-personal promotion, to make sure they’re front and center when a potential patient walks in the door.
Get Started
Frequently asked questions
Real-world data can provide clinical information that may not be available in traditional claims, including biomarker status, clinical notes, laboratory information and treatment history. In this case study, those additional signals helped identify patients who could not be found through claims data alone.
The company used NorstellaLinQ to build a patient identification algorithm combining curated data, large language models and unstructured EHR data. Potential patients were progressively refined into more specific cohorts, including patients with confirmed biomarker positivity.
The case study states that claims data could not capture the biomarker status required for the therapy, reliably identify early-stage patients or provide the near-real-time signals needed to appropriately time field outreach.
In this case, large language models were combined with curated data and unstructured EHR information as part of a scalable patient identification algorithm designed to surface and refine potential patient populations.
The resulting alerts were integrated into the client’s omnichannel engine and next-best-action platform. The case study reports that alerts became a high priority for the field team and helped time outreach to physicians.
Get access to the case study by filling out the form below.
