AI Clinical Summaries Market Map: Easy to create, hard to get right

This is part of Elion’s weekly market map series where we break down critical vendor categories and the key players in them. For more, become a member and sign up for our email here.

Until recently, chart prep meant reading dozens of pages of patient records, intake forms, discharge summaries, and labs to understand a patient’s medical history and generate a summary of the most important information.

AI Clinical Summarization products show promise of dramatically decreasing the time to generate summaries, while delivering meaningfully higher quality output. The implications are wide-ranging, including

• Increasing the speed and accuracy of pre-encounter chart prep

• Improving risk adjustment and coding and surfacing care gaps

• Streamlining referral intake and handoffs

• Curating real world evidence and improving clinical trial matching

Put simply, solutions in this space pull records from EHR systems, HIEs, lab and imaging results, and patient intake forms, then use LLMs to create meaningful clinical summaries. But having an LLM produce an accurate summary is a heck of a lot harder than just plugging it into ChatGPT. Some of the things we’d recommend digging into as you evaluate solutions are:

• Use case: While many of these products have a broad spectrum for functionality, some stand out for chart prep (Abstractive, HealthKey, Hona Health), others on discharge and handoffs (Pieces), while others are focused on risk adjustment (Navina, Credo, Fourier) or intake and eligibility (Synthpop), and some on life sciences or insurance-oriented workflows (DigitalOwl).

• Data sources: Do you need HIEs, and which ones? Can it pull from your EHR? Handle uploaded documents, labs, and imaging results? Ability to OCR handwriting?

• Clinical workflow: Is the clinical summary integrated into your provider’s workflow, i.e. notes uploaded into the EHR or sent to their email as a briefing?

• Speed vs. safety: Is the clinical summary entirely AI-generated, or is there a human-in-the-loop? Given the potential for hallucination or leaving out critical information, you might be willing to trade latency for potentially higher accuracy.

We’re starting to see clinical summarization products building AI scribe solutions, and AI scribe solutions building clinical summarization tools. We view this as the early days of the race to become the AI operating system. Which companies will emerge as frontrunners? Will there be consolidation as companies encroach on each other’s territory? We’re eager to see how things shake out.