In my opinion, documentation assistants will be one of the best and earliest use cases where AI will make a significant difference in the practice of medicine.
Clinical documentation in the EHR is one of the biggest drivers of physician burnout in the USA, causing as much as $90–140 billion in lost physician time per year. Reducing the clinical documentation burden on physicians, which has been exacerbated by the adoption of electronic health records (EHRs), is a priority. A study supported by the American Medical Association (AMA) found that primary care physicians spend almost six hours a day on EHR data entry during a typical 11.4-hour workday. By auto-populating structured data fields (for example, allergies and problem lists) from open-ended physician notes, querying relevant data from prior clinical records and transcribing recorded patient encounters, AI has enormous potential to free physicians from their computers and dramatically reduce documentation burden. An example of this is clinical language understanding applications that analyze physician free-text narratives and extract problems and allergies as structured data.
Technology companies with expertise in automatic speech recognition—including Google and Stanford, Microsoft and UPMC, Nuance (Microsoft) and Epic, and NoteSwift and Athenahealth, plus a handful of startups such as Saykara and Suki—are already tackling this task. Their goal is to develop AI-driven digital scribes that can listen in on patient–physician conversations and automatically generate clinical notes. Only a few months after the release of ChatGPT by OpenAI, Microsoft’s Nuance added the second version of ChatGPT, GPT4, to its medical note-taking tool. It’s integrating GPT-4 into its Dragon Ambient Intelligence platform, which is used by hospitals around the country to ease doctor workloads by using AI to listen to patient-provider conversations and write medical visit notes. Now, Epic is integrating this into its EHR to help physicians with their workflows.
These speech recognition systems can potentially automate the creation of accurate clinical documentation in acute care, ambulatory care, and post-acute settings across both physical and virtual encounters. By freeing physicians up from documenting while they speak with patients, they improve the experience of the encounter, lead to improved patient and financial outcomes, and result in higher quality risk-adjusted quality ratings. The American Academy of Family Physicians has established an Innovation Lab to explore promising technologies that can help in the practice of medicine. They reported using Suki, an AI assistant that uses voice and commands to help physicians to complete their documentation. Their assessment of Suki showed a 60% adoption rate, and those adopters saw a 72% reduction in their median documentation time per note. This resulted in calculated time savings of 3.3 hours per week per clinician. In addition, participants reported improved satisfaction both with their workload and overall with their practice. They reported that physicians described it as a “breakthrough”.
Virtual assistants and natural language processing (NLP) can integrate real-time, AI-powered clinical intelligence within physicians’ workflows and can help to reduce care gaps with contextual diagnostic and treatment guidance at the point of care. For example, NLP-based clinical language understanding generates structured data from an unstructured narrative in radiology reports in real-time to document compliance with clinical standards and improve the closure of follow-up exam recommendations related to incidental findings. As I’ve emphasized before, NLP/NLU (natural language understanding) use cases for AI in clinical notes are works- in-progress, but the fact that they’re not ready for prime time shouldn’t be a show-stopper. These technologies learn over time and putting them out there means that they’ll improve with use and feedback.
