Team members from Holmusk recently attended the virtual National Association of Health Data Organizations meeting, where they had an opportunity to share information about how Holmusk enriches its real-world data through the application of a proprietary natural language processing model.

Natural language processing has been successfully used to advance research and care in many other therapeutic areas, such as cardiovascular disease and oncology. To date, behavioral health has not seen the same success because the field faces many challenges, such as complex clinical constructs that require significant domain expertise, lack of standard vocabulary to discuss disease states, and more.

To address this, Holmusk has developed the first transformer architecture (BERT) based natural language processing model that is specifically trained for behavioral health clinical data. The model was trained on over 50,000 clinician notes drawn from NeuroBlu, Holmusk’s industry-leading database for behavioral health that contains real-world clinical data captured from over 1 million patients. With this model, Holmusk has demonstrated the ability to extract key clinical features like anhedonia and suicidality for patients with a diagnosis of major depressive disorder.

Holmusk researchers also developed a model that predicts patients’ scores on the Clinical Global Impression-Severity scale, an important measure that reveals information about disease severity and can be used to predict outcomes.

Holmusk Senior Data Scientist Abhijit Ghosh commented on the value of these models. “Using our natural language processing model, we can quantify clinician notes to reveal information about symptoms and CGI-S scores," he said. "This process extracts research-ready data from information that was previously inaccessible and unusable.”

Holmusk’s Chief Analytics Officer, Joydeep Sarkar, added: “With this model, we can characterize how treatments work on the symptoms of real-world patients, evaluating which symptoms change while others remain. Having this new and rich clinical information on hand will help lead the way for more personalized treatments as we move toward precision psychiatry.”