How Does AI Work For Medical Note Taking and Risk Scoring? (Augmedix, HDAI)
If you’re still trying to wrap your head around the use of AI in healthcare, this episode will give you an idea about the use of generative AI to create clinical notes during an interaction between a doctor and a patient.
Key topics in the discussions with the CEO of Augmedix Manny Krakaris and CEO of HDAI Nassib Chamoun:
Generative AI in Healthcare Documentation
Development of Risk Prediction Scores
Data Utilization and Model Training
Industry Challenges and Differentiation
Augmedix is a healthcare technology company that delivers ambient medical documentation and data solutions. Their clinician-controlled mobile app uses generative AI to instantaneously create a fully automated draft medical note after each patient visit.
How is their data model built?
Augmedix AI model works by training on medical notes and structured data generated within the company. This data is then fed back into the machine learning process to refine the models for better documentation. Augmedix has generated over 6 million medical notes and continues to produce over 70,000 weekly. This extensive dataset, accumulated over 11 years, provides a significant advantage in training their models compared to new entrants in the field, said Manny Krakaris - CEO of Augmedix.
Avoiding Hallucination Rate Inaccuracies
One of the key challenges in the industry is the public relations (PR) battle amidst rising competition. Many companies can claim to produce automated medical notes using open-source LLMs.
Augmedix addresses the accuracy of the outputs by using their vast data library to train models. “In the process, we generate structured data from the medical note creation process, which goes beyond a flat file. This structured data, rich in metadata, is used to further train their models,” said Manny Krakaris.
Risk Scoring
AI has tremendous potential for easing the work of clinicians and offering them insights that can aid decision-making. Most vendors designing AI tools emphasize that AI is only there to help clinicians either get a better overview of the patient or predict possible clinical outcomes. The final clinical decision still lies in the hands of clinicians.
Health Data Analytics Institute has been developing clinical risk prediction scores since the early 2000s, focusing on creating statistical tools that synthesize patient data for operational use in healthcare. These models are not designed for diagnostic purposes but rather to consolidate data for clinicians to make informed decisions, said CEO Nassib Chamoun.
Scope of AI Prediction
Currently, the company collaborates with Houston Methodist and Cleveland Clinic, with about a million patients in the Medicare ACO program utilizing their predictive models. These models, which cover a broad spectrum of risk assessments, including mortality, adverse events, and healthcare utilization, are designed to be easily integrated into healthcare systems with little need for customization.
They also predict adverse events such as strokes, myocardial infarctions, and other conditions like acute kidney injury and heart failure. Beyond clinical outcomes, the models forecast patient care needs, including visit frequency and potential costs, which is vital for managing value-based care, said CEO Nassib Chamoun.
The models are still in the early stages of implementation, and clinicians are suggesting the development of new use cases. But some are already in place. For example, at Houston Methodist, high-risk patients are identified for expedited postoperative follow-up to prevent or mitigate complications.
Tune in to the full episode.