AI CDS Safety & Limitations
Our Glass Health AI Clinical Decision Support (CDS) platform is designed to enhance the clinical decision-making processes of clinicians. Our AI CDS generates drafts of differential diagnoses, assessments & plans, and answers to clinical reference questions. The available core features of the platform are not designed for and should not be used for acquiring, processing, or analyzing medical images, signals from in vitro diagnostic devices, or patterns or signals from signal acquisition systems.
All of our AI CDS responses provide the basis for their specific recommendations for independent review by the clinician user. Our AI responses are recommendations or recommendations in the form of drafts that should be reviewed in detail by the clinician user. The available core features of our AI CDS platform should only be used as an adjunct in clinician reasoning and never replace or supersede a clinician's judgment. Clinical testing and optimization of the current version of our platform revealed limitations of the AI CDS, particularly in its handling of clinical scenarios in the realms of obstetrics & gynecology, pediatrics, endocrinology, and palliative care. Our validation of this version of our platform revealed limitations in AI CDS handling of endocrinological issues, complex statistical calculation, complex and nuanced decision-making in uncertainty, complex clinical/ethical legal issues, complex patient-centered decision-making, complex psychiatric presentations, planning and evaluating patient safety and quality improvement initiatives, and the identification of underlying social context as a contributing factor in diagnostic reasoning and determining next steps in management. Our Glass Health Team is continuing to develop the platform with these current limitations in mind to achieve our mission of empowering clinicians with the best-in-class AI CDS and improving patient outcomes worldwide.
Large language models have inherent limitations, including the ability to perpetuate biases from large language model pre-training and fine-tuning and biases introduced by user inputs. Although Glass Health has made significant attempts to remove the ability for the model to perpetuate harmful biases, at the service of our dedication to ensure safety, equity, and alignment in the deployment of AI CDS worldwide, at this time, we recommend the omission of any elements of a clinical scenario related to the handling of race, ethnicity, sexual orientation, gender, socio-economic status, disabilities, age, geographical location, and language and cultural background when using our AI CDS. The elimination of the perpetuation of biases when using a large language model and ensuring that AI serves to help accelerate the achievement of health equity are core elements of the Glass Health mission.
The Glass Health AI CDS platform is expressly not intended for use by patients. Users should not use this model for personal health advice or as a replacement for professional medical consultation. This platform offers recommendations to clinicians that aid in their clinical decision-making, and AI responses often require a clinician's expertise to interpret responses, which often consider a wide range of potential diagnoses, diagnostic options, and therapeutic options as part of an exercise in probabilistic clinical reasoning.
Lastly, HIPAA-protected identifiers should not be entered into this version of the Glass AI CDS Platform, and all patient summaries should be de-identified completely. To access a HIPAA-Compliant version of Glass via secure web browser, or to access a HIPAA-compliant version of Glass through your electronic health record (EHR), please have an authorized representative of your health system or enterprise contact Glass Health at enterprise@glass.health