Unstructured Data: Q&A with Elliott Green

Elliott Green
CEO and co-founder
Dandelion Health

As the pharma industry continues to move further into the digital world, the collection and analysis of data is becoming more complex. Elliott Green, CEO and co-founder of Dandelion Health spoke with Pharmaceutical Executive about how the industry can handle all of this unstructured data.

Pharmaceutical Executive: How important is it for the Pharma industry to access new types of unstructured data?
Elliot Green: Accessing new types of unstructured data is becoming increasingly important for the pharmaceutical industry. Traditional data sources - such as data from clinical trials, claims data or ICD-10 codes found in electronic health records (EHRs) - provide valuable insights, but they often fall short of capturing the full spectrum of patient experiences and outcomes. This is because these data points were created for billing and administrative purposes but often don’t capture the full complexity of patient care or disease.

But so much lies in the richer, unstructured data which make up over 80% of real-world healthcare data. These data assets provide a truly novel understanding using raw biological data (imaging waveforms), alongside nuanced human observation (clinical notes).

By integrating these data sources, pharma companies can gain a much deeper understanding of disease progression, treatment responses, and patient subpopulations. This deeper insight then enables more precise and personalized therapeutic approaches. This is particularly important as we move toward a more patient-centric model of care where the goal is not just to treat diseases but to optimize individual patient outcomes.

PE: Can you give a few examples of these unstructured data sets?
Green: Examples of unstructured data sets include:

• ECGs or EEGs (continuous waveforms), that provide real-time insights into physiological states. For instance, the analysis of ECG waveforms can reveal subtle cardiac issues that might not be evident through traditional structured data, like lab results or demographic information typically found in EHRs. ECGs are also a highly scalable, low-cost method for identifying patients at risk of certain conditions, such as the potential for major adverse cardiac events like stroke or myocardial infarction.
• Imaging Data: Medical images, including X-rays, MRIs, and CT scans, are inherently unstructured. Advanced image analysis algorithms can extract clinically relevant features that contribute to a more accurate or earlier diagnosis and personalized treatment planning.
• Clinical Notes: Physicians’ notes, which often contain observations and insights that are not captured in structured fields of EHRs, offer a wealth of information about patient history, symptoms, and treatment responses. For example diagnostic information not captured by ICD-10 codes, or reasons for discontinuing a certain medication (e.g., side effects or access issues).

PE: How important is the use of AI to the growth of precision medicine?
Green: AI is indispensable to the growth of precision medicine as it allows us to extract the key value from all these data. It changes the nature of the practice of medicine by facilitating the prediction of disease progression in a way that hasn’t been possible up to now.

The complexity and volume of unstructured data require sophisticated algorithms to process and interpret them effectively - AI can identify patterns, correlations, and anomalies in these data sets that would be impossible for humans to discern. This capability allows for a more accurate picture of disease progression and the impact of drugs on patient outcomes over time. It also allows for more accurate predictions of how different patient subgroups might respond differently to alternative treatments, leading to more targeted and effective therapies.

Equally important is the validation of these AI models. Validated AI ensures that the insights generated are accurate, reliable, and equitable. Without rigorous validation, AI-driven analyses could be biased against certain patient populations or lead to incorrect conclusions, potentially compromising patient safety. To this end, one of the first initiatives Dandelion launched into the market was a free public service for algorithm audits, supported by the Gordon and Betty Moore Foundation and The SCAN Foundation.

PE: How will this (multimodal data and validated AI) specifically impact the GLP-1 market?
Green: The integration of multimodal data and validated AI will have a transformative impact on the GLP-1 market, mainly because it allows us to understand what is happening at this moment, rather than waiting to conduct time and resource-intensive clinical trials.

GLP-1s, initially developed for glycemic control in diabetes, are now being explored for a range of indications, including weight management, cardiovascular health, renal impairment and even neurological disorders. By harnessing unstructured data - like imaging and waveforms - alongside traditional structured data, we can gain a more comprehensive understanding of how GLP-1s are affecting diverse populations. For example, for those with a GLP-1 in the market, our RWD allows them to understand the wide range of impacts their medication is having. For those that are developing GLP-1s, or are affected by the impact of GLP-1s (almost all companies), the value is understanding both where there is impact and where there isn’t so they can quantify either the scale of the opportunity or the challenge.

Earlier this year, Dandelion launched one of the most comprehensive real-world datasets of GLP-1 patients and we have been working with several companies and researchers who are already utilizing this data. The role that AI algorithms could play in structuring and analyzing the unstructured multimodal data is very exciting in terms of better characterizing disease progression and understanding the impact of these drugs on secondary outcomes and patient populations that have not been studied before. Some of the findings we are already seeing (that we will be sharing in the coming months) would traditionally take years, and billions of dollars, to uncover using clinical trials. We have been able to do it in weeks and months for a small fraction of the cost.

Finally, there is huge potential for precision medicine here, too. AI can help identify biomarkers from imaging or waveform data that predict a patient's response to GLP-1s, enabling more personalized treatment plans. This AI-powered approach can expand the potential of GLP-1 therapies, optimizing outcomes in areas such as obesity management, cardiovascular risk reduction, and even neurological conditions, ultimately broadening the market and enhancing patient care.