With much hoopla towards a push for AI (artificial intelligence ) in EMRs now palpable across the United States, I hope EMRs don’t try to run before learning to walk. Since the launch of the first EMR in 1972 and federal mandates for adoption and meaningful use in place since 2014, EMRs have had ample time to evolve. In the past 13 years of my clinical life in the US as a Hospitalist, I have had the chance to work with various EMRs and I find they are still mostly passive record-keepers who need to morph into efficient secretaries & active clinical partners.
From the Hospitalist point of view, these are 8 basic and practical ‘AI’s I hope EMRs adopt :
- Adaptive Information: No matter what clinical issues patients have, their information display looks the same! An EMR designed to dynamically display relevant information depending on active clinical issues and urgencies would save clicks and help align our thought processes.
- Automated Inquiry: Automated herding of contextual information for a clinical problem can save hundreds of clicks a day. For example, a low Sodium level display could have a clickable alert that pops up historical Sodium levels, last Urine Sodium & Osmolality, Last TSH, I’s & O’s, Weight trend and enlist causative meds like HCTZ, SSRIs, etc. We providers could easily custom-populate desired data fields on such pop-ups themselves, we just need the EMR to provide that framework or integrated app.
- Advised Intervention: A responsive, auto-sensing widget could display orders relevant to what I am browsing – much like the Google Adsense technology that has been around since 2003. Thus, a low Potassium level could trigger a suggested potassium supplement order and a lab recheck that just needs my blessing now to execute.
- Augmented Information: Say what you will about social media, but Twitter & TikTok videos are a masterclass in effective communication by keeping it short, engaging & visually easy to remember. Pictures do speak louder and faster than words – almost 50% of our cortex is dedicated to visual processing. Illustrated representation of clinical issues and visual trackers for hospital courses that give a birds-eye view can accelerate the assimilation of clinical information, aid problem-solving, and create safer & more effective hand-offs.
- Abbreviated Information: We providers are traditionally verbose when it comes to dictation and official documentation. “Note-bloat” is a painful issue. I often tinker with prompt engineering of chat-based AI programs like ChatGPT, Gemini, Claude, etc in creating distilled versions of dictated voluminous clinical notes that communicate just the essence without compromising vital information using the least possible text. Such functionalities can be leveraged to ‘cut the fat’ in bloated notes and create concise sign-outs for other providers.
- Assisted Interpretation: We now have a plethora of evidence-based and clinically validated medical calculators to assess severity, prognosis & mortality risk for a myriad of conditions – why not automatically calculate & display relevant scores?
- Audio Integration: What better way to reduce clicks than not needing to click at all? Both browsing and ordering on EMRs should ideally be voice-enabled like the Alexa family of devices. I have created macros on Dragon-dictation software to create keystroke sequences for voice-activated command execution. EMR should have either standardized APIs that allow dictation devices to voice-activate every clickable feature or just like Epic’s “Hey Epic” feature, have an inbuilt voice assistant that can be valuable for mobile app experience. And just like ‘Routines” on Alexa, I hope these voice features will allow automated cascading of various tasks.
- Apocalyptic Intuition: We providers know what the possible complications of a clinical presentation or interventions are, we simply do not have the time to watch for evolving changes and have situational awareness round the clock for every patient. So instead we deploy the inbox method of processing complications retrospectively. Recently FDA approved the Sepsis-DART app has automated round-the-clock monitoring for sepsis, something similar for various possible contingencies for both critical and non-critical patients will be very valuable.
The primary focus of EMRs has been billing and federal mandates on quality, safety, and communication. Improving provider efficiency and workload has not been considered monetizable. With burnout entering the mainstream conversation, the value equation has changed. EMRs like Allscripts and Epic have migrated to core-OS and third-party integration models, much like the Apple App-store which allows flexibility for innovation. So there is hope.
With Artificial Intelligence, though we haven’t fully addressed issues of accountability, hallucinations & HIPAA, by always keeping the providers in the driver’s seat, we can co-exist. With time and technology, our EMRs will improve. At present most of them are like a vast library – but a librarian is missing.
