Florence - the Intelligent Virtual Assistant for the #EHR [Video] #MHealth

Video of the presentation from the Health2.0 show in October 2013 of the Florence Demonstration
You can see Florence in action at the site
or in the video below

Technology as an Aid vs Hinderance to Doctors

A recent article in Becker Hospital Review:  Technology Should Aid Human Interaction: Q&A with Dr. Nick Terheyden, CMIO of Nuance featured some important points to make

Health IT needs to fade into the background. It needs to become part of the fabric of the office rather than the focal point, and then the interaction will change

• Using the tools to allow the clinician to focus on the patient not the technology
• Human beings deal in narrative and stories, patients want to tell their story and clinicians need the richness of the narrative to help guide medical decision making
• Remove the Physical Barriers to the clinicians patient interaction
• Healthcare is not the focus - the patient is
  

The key to our future and to the successful use of health IT will be turning the focus back on patient and the physician.

Dr Watson Takes on Healthcare

Watson Emerges Champion in Jeopardy

Watson emerged the Jeopardy Champion last night with a resounding $ 77,147, beating both competitors (Ken Jennings:$ 24,000 and Brad Rutter:$21,600) by over 300% ($53,147/$55,547).

Hats off to Ken Jennings and Brad Rutter for taking on the challenge and doing such a great job

This was followed today with the joint Nuance/IBM Press Release: "IBM to Collaborate with Nuance to Apply IBM’s Watson Analytics Technology to Healthcare" (here on IBM and Nuance's site) that recognizes the tremendous value that Watson can bring to healthcare. Watson showed last night the major advances that have been made by IBM's “Deep Question Answering” (QA) research team:

the ability of a computer to understand natural human speech inquiries that pertain to a limitless range of topics, and to make informed judgments about requests

There was plenty of coverage as evidenced by the 15 Million hits in Google for the search (Jeopardy/Watson) with the vast majority highlighting Watson's superiority (Time Magazine: Winner, Dr Watson I presume, ZDNet: Watson's Next Adventure - Healthcare with Nuance, Information Week: IBM Nuance Envision Watson Helping Doctors, NY Times: Computer Wins on Jeopardy, Trivial It's Not andIBM Moving Watson Supercomputer Beyond 'Jeopardy' To Health-Care. Stephen Wolfram of Wolfram Alpha weighed in with this thoughtful post (Jeopardy IBM and Wolfram Alpha) comparing Watson and Wolfram Alpha and this more light hearted post If Google Played Jeopardy: Smartest Search Engine, But It’s No Ken Jennings where he challenged search engines with the simplified task of entering the same Jeopardy Clues

Google showed well

Leading them to suggest that

Either he’s (Ken Jennings) an immediate acquisition candidate in Mountain View and Redmond, or he should just start his own search engine and answer queries as they come in

There were other dissenters including this post Could Google Play Jeopardy Like IBM's Watson. Google has certainly cornered search and access of data but even if the data can be found with a google search and there are certainly some shared concepts in the search technologies I'm not sure that finding the data in a search page is the same as answering the question and really understanding the question. But the summary and analysis of Natural Langauge:

Natural Language Reality Check: The reality is that the technology that Watson demonstrates, while amazing in a game show, is overkill for what most people need. Those behind “natural language” search technologies have long trotted out sentences like the “Who’s in the pajamas” example above to demonstrate how “smart” their search tools are. And yet, most searches people do on search engines are only two or three words long.

Missed the mark and the analogy is flawed on several counts.  I would suggest that our interactions are deliberately simplified when we interact with Google - we work around the lack of understanding of the Google search engine. And while Watson may be overkill for "most people" the potential application in multiple areas dealing with ever increasing volumes of unstructured data it is healthcare that has been struggling with the challenge of overwhelming clinical knowledge and our inability to access and apply this at the time of care delivery. In this discussion on the possible healthcare applications of Wtason/DeepQA


Dr Herbert Chase (Professor of Clinical of Medicine at Columbia University) says:

For at least 30 years it has been impossible for a physician to master all the material to practice medicine at the highest level. Biomedical literature has doubled in size every seven years but patients want those facts at the doctors finger tips when they see him (the doctor)

It is this challenge of data that clinicians face every day as they attempt to deliver the best possible care to each and every patient at the time of the consultation and something their patients expect. Definitely not overkill for patients or doctors and given the Tsunami of medical knowledge and the challenge of sharing clinical data. DeepQA will add a new level of medical intelligence to support to clinicians is with the application of Natural Language Processing taken to a new level of understanding.

Back in June I talked about this technology and the potential for application in healthcare (NLP in Healthcare). In October Nuance announced our strategic partnership with IBM and I covered the news in this post: Clinical Documentation Challenges and then again building towards the Jeopardy challenge NLP in Healthcare Part 2 and most recently this week as part of the potential solution to the challenge of shareable clinical data and clinical data models (The PCAST Opportunity, HL7 CDA, UEL and SAGE)

The announcement today builds on a deep research and technology partnership that already exists in multiple areas between the companies. As one person pointed out in a note to me yesterday

"I cannot understand that Watson cannot 'hear' an opponents wrong answer. They text in the question and then Watson parses the english, algorithms swirl, etc. However, when an opponent speaks it cannot hear. So when a bad answer is spoken it is bound to repeat it.. "

That is part of the innovation and solutions that Nuance will contribute to the partnership which will also include the Clinical Language Understanding (CLU) Technology that will be used create new solutions that provide hospitals, physicians and caregivers access to critical and timely information expanding from recognizing what was said and parsing data to now understanding the intent and providing guidance. This will push us one step closer to intelligent medical analysis real time with the clinician to assist hospitals to utilize facts and knowledge as they migrate toward evidence-based and accountable care models:

Recognizing the tremendous value that Watson can bring to healthcare, Nuance and IBM have teamed to co-develop solutions that will transform vast amounts of clinical data into actionable information across the continuum of care. This endeavor is intended to unlock important medical knowledge and facts buried within huge volumes of data repositories, providing healthcare provider organizations, payers and individual physicians a new level of medical intelligence

It has been yet another exciting day in the world of healthcare technology

If you want to see the Jeopardy games you can watch them below
Part 1:

Part 2:

and Part 3:



And it is not just healthcare:

NLP in Healthcare Part 2

Following up from my post back in June (NLP in Health care) the Jeopardy challenge is on - coming to your TV February 14 - 16. There was lots of coverage of the announcement
This from the Washington Post - "'Jeopardy!' to pit humans against IBM machine" and the IBM release
Ken Jennings and Brad Rutter two of the leading Jeopardy winners will go up against Watson. Jennings had the game show's longest winning streak at 74 games in a row and Rutter has won the most money standing at $3.3 million.

What is Watson



But what's important is the possible future applications:

IBM is hoping the technology it exhibits will have some practical uses eventually, for instance helping doctors diagnose illnesses or solving customer problems at technical support centers.

Applying this in healthcare is part of the partnership announced back in October and featured in the posting  "Clinical Documentation Challenges". It will be exciting to see the performance of Watson in Jeopardy on live TV but its the application of this in healthcare that will present some revolutionary opportunities...watch this space

Clinical Documentation Challenges

We are on a path to roll out a large swathe of Electronic Health Records (EHR) but a recent report published on the AISHealth.com Audits of Electronic Health Records Cloning Reveal Documentation Problems That Put Compliance at Risk will give many folks reasons to pause and consider their strategy in rolling out electronic medical records and reconsider how they capture information in these systems.

Interestingly the article suggests that EHR's "can reduce the time it takes physicians to document patient encounters" but there is a fair amount of research suggesting that EHR can increase the burden and time taken to document. A study in the Healthcare Ledger in March 2009 showed an increase of ~ 4x when documenting using the EHR and there is increasing concern that the current burden of clerical work being required of medical residents is limiting the educational opportunities and failing to teach our future doctors the process of reflection and distillation of a patient history that is an essential part of the diagnostic process (Doll and Arora 2010). Add to that that many EHR systems limit the potential for capturing the complete clinical story of the patient as outlined in a recent study “Communication of Clinically Relevant Information in Electronic Health Records: A Comparison between Structured Data and Unrestricted Physician Language” (Resnik, Niv et al 2010) in a systematic comparison between free natural language dictations and information codified by structured categories in an EHR demonstrated a failure to capture clinically significant information. Even in the most conservative estimate the study demonstrated 25% clinical omissions that rated 4-5 on a 1–5 scale of seriousness (1 being minimal severity and 5 meaning severe).

However the access to real time information and improved legibility deliver significant benefits and advances into our healthcare system. But as Nina Youngstrom points out:

CMS and Medicare contractors are wary of classic EHR physician documentation shortcuts — cloning (cut and paste), macros and templates — and audits are bearing out their concerns

As the audit demonstrated "Each note should contain individualized data that supports the medical necessity of the visit or procedure.” and problems stem from EHR documentation shortcuts:

• Cloning (cutting and pasting): Physicians copy information from previous patient encounters (e.g., demographic, history of present illness, exam, medical decision making) and paste it in the current encounter.
• Templates: Physicians fill out templates for patient encounters that cover a lot of ground with a few key strokes. The review of systems is pre-filled with the term “negative” for each organ system. For positive answers, physicians must change “negative” to reflect the positive response given by the patient.
• Macros: Macros are a type of EHR shortcut that allows the entry of generous customized data quickly. Though initially CMS resisted the use of macros, the agency gave its approval for their use by teaching physicians (see Medicare Transmittal 811). With macros, teaching physicians, for instance, type in “.liv” to convey “liver exam,” which triggers a drop-down menu of choices for the next step.

So avoiding these pitfalls and capturing the essence of the clinical consultation in the "Medical decision making" which is the cognitive process and is hard to document with templates and macros is key to both good quality documentation as well as avoiding potential CMS audits and challenges in the future.

There is no one size fits all to these challenges and in different clinical circumstances different solutions will be beneficial but providing tools to document the complete narrative and extract key data elements will help drive clinically actionable data into the EHR while maintaining the decision making process that includes taking and documenting a full history without burdening the physician with mundane data entry tasks. Clinical Language Understanding offers to bridge this divide capturing the voice with proven voice recognition Technology (Dragon Medical) and the exciting addition of Clinical Language Understanding: The video demonstrates the CLU in action offering an alternative path for clinicians weary from screen based data entry:



Or you can see the video here, and can see Paul Ricci on CNBC "Street Signs” hosted by Erin Burnett "Nuance Partners with IBM":




and read about the collaboration with IBM here. What would you rather be doing? Manual data entry or dictating. This announcement and the work underway offers a solution that bridges the divide between the need for clinically actionable data and physicians desire to capture the complete clinical story for the patient.

The Science of Medicine

Medicine is complex and providing the best possible care is a challenge, and it is getting more complex on a daily basis. I'm willing to bet that at least a few readers will still hold onto the belief that playing music, in particular Mozart, can improve brain function. IN fact searching Google Scholar reveals about 8,400 articles related to Mozart and brain function. There are even pages promoting the this concept "How to improve Brain Function". But as this NPR podcast (The Mozart Myth and More) points out it all started with an one page article published in 1993 Francis Rauscher: Listening to Mozart enhances spatial-temporal reasoning. What followed took the researcher unawares with television interviews, lots of media attention and even some hate mail and calls (because she was misquoted that rock music was not good for brain function)! In fact this limited experiment (36 students total) were able to improve spatial temporal test - nothing else, not general intelligence.

It's easy to misinterpret results and data and human nature probably pushes us towards believing in miracles and cures but what is important is we review the data carefully and base our decisions on science. In the blog "Bad Science" Ben Goldacre spends a lot of time and effort debunking myths and researching the the data to uncover the facts and interpret them correctly. He has exposed the Nutrition and Pills Industry and Fish Oil and a related piece on science and health reporting errors and this expose of Obvious Quacks. The challenge is we are deluged with information and sorting myth from reality is difficult for everyone from patients to clinicians.

For clinicians this problem is even more acute as they deal with debilitating work loads, time pressures and information overload that oftentimes are overwhelming even for standard cases. I've watched several colleagues recently treating patients an dealing with the information available in the electronic medical record (EMR). In all but a few instances the volume of data is overwhelming and processing this in the limited time available is a challenge. Add to this the need to verify existing data and update with new information and then capture the latest data relating to the patient which will become too much to burden the existing system of clinical care. Meaningful Use and the final rule making is pushing us towards EMR but for these systems to begin to address this information overload rather than add to it we have to find ways to capture clinical data without adding further work to the clinicians who is time challenged

As part of that initiative the Healthstory Project has created a vision of a comprehensive electronic clinical record that captures the all important data while retaining the complete patient clinical story. Part of the projects has been to develop a range of data standards for sharing that information and to date 5 draft for use standards have been issued and they are developing an additional 4 more. Getting the information into these formats will be a challenge and there are several efforts underway to facilitate this process. One of the members of the Healthstory Project Nuance (full disclosure they are my employer) is looking for pilot sites to test a prototype of their Clinical Language Understanding technology that is aimed at easing this burden and providing a bridge between the narrative documents generated currently by physicians and the structured data that is essential to fill these EMR's that will help deal with the information Tsunami in medicine and help guide patients and clinicians in delivering the best possible care. There are other developments underway and I have no doubt over the coming months we will see a range of solutions aimed at plugging the doctor more directly into the clinical knowledge base to help them (to help the patient) make clinical decisions with all the information, processed and assessed each and every time we reach a clinical decision point. Some of this will be about user interfaces and we might even end up with a Neo like interface

Its not as far fetched as you might think). some of it is about work flow and processing of information but the building block for all these improvements is based on capturing and processing information from the clinical interaction.

What's your experience been - have you got systems in place and have you developed work-flows to facilitate the clinical patient interaction? Has your doctor been able to capture information and process while you visits?

Keeping Medicines Narrative

Thought critical free-text physician notes are under threat in the current slew of Electronic Medical Records and are at risk of being washed away in the rush towards the digitization of medical records. In Jeff Barry's article in Health Management Technology; "Value of Unstructured patient narrative" he cites examples of

Throwing the patient out with the paper

In fact there has been increasing coverage of the challenge associated with delivering high quality medicine when you drive out the clinical narrative. As one colleague complained to me;

"the narrative is a big part of the value I bring, removing it dumbs down the information"

The challenge is described well in the piece "Mining Clinical Data: Road to Discovery"

For administrators and researchers, who need to extract data to develop reports, they would prefer all clinicians to enter information to the exact same manner, which means time is saved for the administrator, but not for the clinician...it's a trade off of who spends more time and who saves more time

For the busy clinician who currently derives little from the capturing of structured codified data there is little benefit for the extra and often frustratingly challenge extra time required to capture data in this structured form. It's very difficult to get clinicians to enter coded data as opposed to entering patient conditions.

There is a big push demanding structured data that is sweeping over clinicians who are increasingly highlighting the issue and pushing back refusing to become data entry clerks. While some may see this as simple resistance to change it is not just the physician who is loosing information. Patients will find their records relegated to a series of check boxes and lists and while this may provide information it does not cover the full story. The Clinical Narrative must be integrated into the EHR.

In addition to the patient and the clinical team as Jeffrey Barry notes the public health researchers also stand to gain from richer electronic patient narrative. As I frequently cite - Henry VIII medical record remains a shining example of the value of the narrative providing far more detail than could ever have been captured using a rigid form based data entry tool. There are many details available and exposes including this brief review refuting the common view that he died of syphilis



And this posting reviewing possible causes of death comparing to his symptoms and this article in the Journal of History of Medicine: "Henry VIII and Medical Study".

The unstructured free text of the physicians progress notes provide color to the unstructured data's black and white

In a recent case I observed the clinical note captured in free form using speech recognition that allowed the physician to record the information directly into the patents digital record providing immediate access to critical information to everyone on the care team. There was no delay in generating the note and even if it was possible to put this information into a structured form it would have taken far more time than the 1-2 minutes this took to create using speech recognition and since it was generated immediately and included in the EMR it was available to the full clinical team.

But all is not lost and as both articles highlight there are emerging tools that will bridge the divide between the free from narrative and the need to generate structured codified data both for analysis but also driving improvements in quality of care. In the examples provided a post processing by Natural Language Processing (NLP) technology that in a study from 2008 at the Regenstrief Institute and Indiana University School of Medicine where

researchers were able to produce sensitivity, specificity and positive predictive values exceeding 99 percent (detecting Methicillin Resistant Staphylococcus aurerus - MRSA)

That's impressive and already showing value from automating the extracting of information from free form text. There are limitations and the accuracy as variable as the terminology used by clinicians for the same condition but as IBM is showing with Watson the application of NLP in healthcare is rapidly approaching significant exciting new frontiers

Human interaction also is required because of the complexity of language and providing tools and a blended approach to capturing this information will be key to making this work. In this example Rob Stewart demonstrates a variety of ways of capturing structured information in a radiology setting:



There are many ways of approaching the problem and offering multiple choices that make physicians more efficient as they interact with patients will be the key to success and adoption.

Do you have tools or techniques that work well in your setting. What balance have you achieved between the digital need for structured data and the human need for narrative. Let me know - share your experiences or views.