Pharma companies are increasingly turning to real-world data to answer their commercial business questions, but not all realize that unstructured EMR data is the unsung hero of most queries. Whether a manufacturer is struggling to find a niche patient population, conduct unbiased outcomes research, or generate persuasive proof points, unstructured data can fill in the gaps left by other real-world data sources.
Until recently, this valuable information has been virtually impossible to analyze at scale. Much of the patient data contained in EMR systems—like a patient’s demographic information, vitals and procedural history—adheres to a defined format, which makes analysis feasible. But the qualitative information recorded by a patient’s care team, such as clinical notes, radiology reports and discharge summaries, is stored in free-text fields.
For years, the complexity of turning this data into insights meant manufacturers were unable to see the complete patient journey. But why is this data so pivotal in the first place?
Consider Sarah, a grandmother recently diagnosed with stage 3 breast cancer. While medical claims and lab results reveal glimpses of Sarah’s treatment journey, the richest details about her care—her tumor size, biomarker levels, diagnostic notes, symptoms and physician sentiment—are buried in her electronic medical records.
For the pharma company whose therapy is designed to treat Sarah’s tumor, this data is essential for finding Sarah and others like her: a highly specific subset of post-lumpectomy, stage 3 patients with both ER positive and HER2 negative tumors less than 3 cm in size. Without the ability to parse this unstructured EMR data, the manufacturer will never be able to find Sarah in time to impact her treatment—or improve her outcomes.
Precise patient identification and HCP targeting
With the evolution of AI and natural language processing (NLP), manufacturers can now comb through unstructured clinical data for any combination of terms, clinical scores or test results. A patient’s unstructured EMR data might include a physician’s observations about their family history, potential diagnoses, or nuances of their clinical progression. These tidbits are the missing puzzle pieces that help commercial and HEOR teams understand the full picture of patient’s care.
The oncology manufacturer in our scenario, for example, can now segment its starting cohort of patients based on where they are in their treatment journey. To find eligible patients, this manufacturer needs to know which patients have undergone genetic variant testing, and what those results indicate.
Using integrated lab, claims, and EMR data, the manufacturer can pull in test results and deploy NLP on the unstructured EMR data to see variant results of interest. While traditional datasets may only indicate if a biomarker is positive or negative, EMR data can return nuances like high, moderate or low expression levels.
By analyzing these real-world datasets in tandem, the manufacturer can pinpoint patients with the right metastatic diagnostic codes and exclude those with the wrong codes. Unstructured EMR data further narrows the focus to the patient’s tumor biology, returning patients with both the right variant and mutation status to be eligible for the manufacturer’s therapy.
Instead of a million potential patients, the manufacturer now knows exactly which patients can benefit from their treatment—and who their prescribing providers are. The pharma company’s sales reps can contact Sarah’s primary treating physician within the window of opportunity, ensuring that she is able to benefit from their targeted therapy.
Customized analytics for addressable patient populations
Leveraging unstructured EMR data in combination with other real-world datasets can help pharma companies create a bespoke data asset for a myriad of use cases. After developing a clinical algorithm to surface the right patient information from as many interconnected datasets as necessary, a manufacturer can use this asset on an ongoing basis, moving both forward and backward in the data.
For example, this data can help identify the right patients at exactly the right time: after a positive biopsy, variant testing and tumor grading is completed and before therapy begins. A manufacturer can track patients as they approach the treatment decision, using weekly trigger files to alert their sales reps to reach out to the prescribing physician before that decision has been made.
Pharma companies can also capitalize on the longitudinal nature of this data to run a comprehensive HEOR study that reviews historical patient outcomes. They could create a compelling value story for a new brand by leveraging suboptimal outcomes data for patients treated with competitor products. The data could also reveal trends in physician treatments, referral patterns or unmet needs that might inform future development priorities.
By bridging unstructured EMR data with open and closed medical claims, reference and hospital lab tests and results, and structured EMR data, pharma companies can create a tailor-made data asset that will be instrumental for a wide variety of use cases. Studying the nuances of this longitudinal data can help a manufacturer’s HEOR and commercial teams efficiently map the patient journey, identify barriers and ease access to their therapies.
