During his tenure as a medical student at UC San Francisco, Dereck Paul harbored growing concerns regarding the stagnation of innovation within the realm of medical software when compared to other sectors, such as finance and aerospace. This apprehension led him to a fundamental conviction that the optimal way to serve patients is by equipping physicians with software that embodies the cutting-edge technological advancements. Consequently, he envisioned establishing a company that would prioritize the exigencies of patients and healthcare providers over the interests of hospital administrators and insurance entities.
In pursuit of this vision, Paul collaborated with his colleague Graham Ramsey, an engineer at Modern Fertility, a pioneering women’s health technology company, to co-found Glass Health in the year 2021. Glass Health introduces a digital notebook designed for physicians to facilitate the seamless storage, organization, and sharing of their diagnostic and treatment methodologies throughout the trajectory of their medical careers. Ramsey aptly characterizes it as a “personal knowledge management system” tailored to enhance the practice of medicine.
Paul elucidated on the genesis of this endeavor, stating, “Amidst the pandemic, Ramsey and I bore witness to the overwhelming burdens besetting our healthcare system and the escalating crisis of healthcare provider burnout. My own firsthand experience of provider burnout during my medical student rotations and subsequent residency at Brigham and Women’s Hospital underscored the imperative for us to create a company committed to harnessing technology comprehensively to ameliorate the practice of medicine.”
Glass Health garnered early traction across social media platforms, particularly on X (formerly Twitter), garnering significant interest from physicians, nurses, and medical trainees. This resonance translated into the company’s maiden funding tranche, securing $1.5 million in a pre-seed funding round led by Breyer Capital in 2022. Subsequently, Glass Health earned acceptance into Y Combinator’s Winter 2023 cohort. However, earlier this year, Paul and Ramsey made the strategic decision to pivot the company towards generative AI, aligning with the burgeoning trend in this domain.
At present, Glass Health offers an AI-powered tool underpinned by a substantial language model (LLM), akin to OpenAI’s ChatGPT. This tool is engineered to generate diagnoses and proffer “evidence-based” treatment options for consideration by healthcare professionals. Physicians can input patient descriptions such as “a 71-year-old male with a history of myocardial infarction presenting with subacute progressive dyspnea on exertion” or “a 65-year-old female with a history of diabetes and hyperlipidemia experiencing acute-onset chest pain and diaphoresis.” Glass Health’s AI system subsequently delivers likely prognoses and clinical plans.
Furthermore, Glass Health has the capacity to draft case assessment paragraphs for clinician review, inclusive of elucidations regarding pertinent diagnostic studies. These explanations are amenable to editing and subsequent utilization for clinical documentation or dissemination within the broader Glass Health community.
Notwithstanding the theoretical promise of Glass Health’s tool, it is imperative to acknowledge that even state-of-the-art LLMs have encountered challenges when providing health-related advice. Several AI healthcare startups, including Babylon Health and Cass, have faced scrutiny for assertions regarding the superior capabilities of their AI systems and the potential dissemination of harmful advice.
There exists an overarching concern pertaining to bias within LLMs, including those deployed in healthcare, such as Glass Health. These models, due to their training on existing healthcare records, may inherit latent biases inherent in the data, potentially resulting in an inadvertent favoring of certain demographic groups. However, Paul underscores Glass Health’s commitment to mitigating bias through ongoing engagement with a cadre of academic physicians tasked with reviewing and refining AI guidelines.
It is important to underscore that Glass Health’s AI system does not intend to proffer definitive or prescriptive solutions but rather aims to furnish potential diagnoses for consideration by healthcare professionals. This nuanced approach serves to circumvent heightened legal scrutiny and potential regulatory oversight by the FDA.
Paul emphasizes that Glass Health fosters collaboration with academic physicians to construct and subject guidelines to rigorous peer review, thereby affording healthcare practitioners meticulous control over the AI’s outputs. Moreover, Glass Health collects user data to continually enhance the underlying LLMs, with users retaining the prerogative to request data deletion at any juncture.
Despite these challenges and concerns, Glass Health has effectively attracted in excess of 59,000 users and offers a subscription service directly tailored to clinicians. On the horizon, Glass Health intends to pilot an enterprise offering characterized by electronic health record integration and compliance with the Health Insurance Portability and Accountability Act (HIPAA). Currently, 15 health systems and corporate entities await access to this offering.
With total funding amounting to $6.5 million, Glass Health envisages the allocation of resources toward physician oversight, guideline development, AI refinement, and comprehensive research and development endeavors. Paul asserts that Glass Health is fortified with a four-year runway to sustain its operations.
