Wednesday, July 23, 2014

Adding Some Structure to Last Week’s #HITsm Tweetchat

Last week’s #HITsm tweetchat topic – Unstructured Data – was of particular interest to me. Kudos to Bernadette Keefe (@nxstop1) for wrangling some good information, ideas and opinions out of the participants. The tweets were flying through my tweet stream fast and furious.  In an attempt to pull a little more value from that stream, I went ahead and added a little order and structure. Here’s what I came up with.

Note: Thanks to all who contributed. I’ve not noted who tweeted each item in this post as I didn’t want to misquote or omit anyone – which I surely would have done. You can learn who said what and to whom via the transcript of last week’s #HITsm tweetchat here.


Topic 1: How will the analysis of Unstructured Data aid in healthcare delivery: for MDs/Providers/Patients/Payers?

Are we talking systemic use or manual? If data can be systemically used, a flood gate of use cases open up.

Think the promise is truly in analysis of the unstructured data~eventually.

Gain valuable info about patient experience with drugs and interactions

Well good portion is 'the patient's story' how their health makes them feel. may never be structured

Yes, lets talk re value in unstructured data. It contains all patient narratives, the stories told to countless MDs ~inval…

Agree but worth asking the question, problem is too structured means less innovative thinking

IMO, biggest #unstructured data miss is in the 4th "P" a patient's peers. Blog: http://t.co/BEZaYlSTnL

No because there is high value in "the narrative" for telling and communicating the patient story

Need all: structured, unstructured, human intelligence together

For Providers – reduce over-utilization, eliminate redundant diagnostic tests see JACR http://t.co/cCXnWl8E6A

Value in unstructured data--patient narratives, the stories told to countless MDs ~invaluable

Interpreting clinical text ultimately requires more than just sound, structure & meaning. It involves goal, plan & task recognition

#1 question I always ask when someone brings up BI and BIG DATA is what problem are we trying to solve

It's possible NOW to analyze unstructured data - query it & add value to existing structured datasets

What percentage is unstructured today? Seems like a big hole in the "complete record". Possible risks for systemic analysis.

Yes, the benchpoint number given is 80% ~ Massive!

Should we make HCP use structured data only, or develop tools to analyse unstructured data?

Unstructured data is ALL TEXT ~ so all MD/ other HCP provider notes, all radiolog reports, surgical reports, etc

Indexing a semantic searching is a good start and facilitate finding relevant data, but still may require manual analysis.

SOAP notes and plans the gold mine of unstructured, free-text data. RT @nxtstop1 T1 Unstructured data is ALL TEXT

My MAC just puked some 0s and 1s out the side seeing "SOAP"

Please educate me - what about data from wearables? Is that unstructured? I believe @Brian_Eastwood touched on this recently.

One benefit of "unstructured data" is that using can can add value w/o changing physician workflow.

If no impact, no worth RT @JMCelio RT @georgemargeli yes -- how is data driving clinical decision-making, improving quality of life?

But have to keep in mind safety issues, ambiguity can be dangerous

Can analyze common terminology used...first within institutions & expanding outward-->standard terminology-->structured data

I suppose we only know a little if funnel is limited to ‘structured data’ & way beyond nuance?

Explicit or implied delimiters, tags and/or “context” can add structure to unstructured data.

Any analysis & exchange between these constituents (structured, semi-structured or unstructured) can only help.

Topic 2: How can you, your firms, #HealthIT in general, help in the retrieval & analysis of Unstructured Data?

Really like this from @CLOUDHealth "Structure is not defined by databases but by relationships with people."

Yes -- how is data driving clinical decision-making, improving quality of life? #NoDataForDatasSake

No #data is #unstructureddata Structure is not defined by databases but by relationships with people.

We get so excited over having more data > don't spend enough time *thinking* about what it's really for

Adding time layer to storytelling aspect of Patient's history. It's very complex and multi dimensional story to tell.

Realized I'd misspoken. Data is structured, identifying quantifiable things, but unreadable to clinical systems.

T%, or ‘business as usual no longer an option…question is ‘what is new biz model’? Is there ‘one’?

That's the common problem, right? There's structured data here, but won't make sense there

Asking questions of unstructured narrative isn't an exact science but it's a better than spending valuable clinical sight time

Unstructured just a tech term b/c its unstructured TO TECH.New thought:text data is Gr8>& tech will step up 


I believe the solution we're looking for is I-N-T-E-R-O-P-E-R-A-B-I-L-I-T-Y.

I wrote bit of a rant on #wearabletech and an inability to get tangible insight from it: http://t.co/nD61pkh9Cg Does that count?

Some standardisation around terminology would make it easier, just because its free text doesn't mean we need to be too free :)

SNOMED gives us some reference clinical terminology

Room for Web crawlers in EHRs? :) Think unstructured data us fine if smartly presented in collective to MD viewing files.

Start turning unstructured data to semi-structured data. Turn semi-structured data to structured data.

Consider adding delimiters, tags & considering “context” for everything you touch and/or have control over.

Don’t just run on sentences. Add white space. Use CR/LF’s… Spell correctly. Make the text “processable”

Actually, IMO it's really just grunt work and getting people to agree on "code translations"

Whoa! Not one single ‘de-limiting’ reference to HIPAA today. A miracle? Or simply context/competence appropriate?

Topic 3: If you have worked on a project with Unstructured Data-please describe pros/cons of the experience. 

One benefit of "unstructured data" is that using can can add value w/o changing physician workflow.

Unstructured healthcare data is not just free text, it includes a variety of multimedia content: video, audio, etc.

One challenge of unstructured data is that using it allows physicians to avoid change

Other hard part is asking the right Qs @HealthFusionKMc: Agree! And if do the data prep right, analysis even easier

Attributing data to member and provider (provider in all its variations: individual, practice, network, etc) is hard.

Biggest prsnl complaint: get unstructured data of my Rx. So tired of getting the wrong thing at the Pharm

A concern kinda BUT~ MDs will always (and should always) use free text to render appropriate nuance to pts stories

Con=Unstructured text includes ambiguity. Need to balance specificity with sensitivity. May result in FP/FN. Set user expectations.

I collect a lot of data which has never been used...yet. I'm confident it will be one day.

Will be even huger with HIEs RT @ShimCode: identifying and properly handling CHANGES to data previously processed is huge.

I'd almost rather see investment in smarter systems to interpret unstructured data then more standards

Wide variation of how “standards” are used & re-purposing of fields without consistent use are 2.

You know hand writing is a type of encryption, especially coming from docs :)

This may generate some controversy but analyzing data is easy. It’s getting the data ready to analyze that’s hard.

Yep. Cleaning data is supposedly 80% of the workload.

Agree! And if do the data prep right, analysis even easier @ShimCode analyzing data easy

Exactly @nickdawson: when I've done big financial models, most of the work is in prepping and normalizing the data

I used to do NLP, which got me into standards like #CDA, here are some of my observations: http://t.co/sQzKspzLZx

Workflow highly important 2 speech recognition & text disambiguation. Used 2 more accurately estimate tokens & meaning probabi…

Topic 4: Given the base data is text,can you think of creative presentations of the analyzed data which might especially engage MDs/Providers/Patients/Payers? 

Can you think of creative presentations of the analyzed data/text that may engage caregivers/Patients/Payers?

IMAGINE all of us having our INDIVIDUALIZED , PERSONALIZED WORD CLOUD AT THE BEGINNING OF #EHR -and it could chg (improve hope)

"Creative" is key. As a patient, I don't get pumped about taking charge of my health by looking at pie charts, bar graphs

Yes, lets get around to the creative presentation. Particularly fond of this as we see w/ Pinterest Interest, FB pix, VISUALS ROCK

Longitudinal progression of a health concern over time, from out of shape to overweight to diabetic...

Good txt summarizer can [make] pages of unstructured text afew sentences

MDs, MAs, Nurses.. Always looking for data on patient. Spending more time being a detective than doing actual work.

It's all about the use case - not the tech, not the data - but the value in unstructured clinical narrattive to improve healthcare

Tie in the social #data for prevention will be good

Word clouds hit the popular pairs but the real relationships are often hidden

Nice, but might b better to see (creatively) predictions of how my lifestyle choices will affect me 5, 10, 15 years on.

NLP is cool tech but it should be about solving problems - not tech for tech sake

Crawl it, index it, search it, rank it, and present it. http://t.co/izYhdV8BRZ

Here is a great tool to help convey your analyzed data "Visualizing Health" http://t.co/y4fDCnAFIS

Teaching clinical analytics key to future HIT pros - http://t.co/cxNzzfQXbd

"Communicating Data for Impact" http://t.co/vaE0DOccen

Absolutely!!! Crawl it, index it, and apply context.

There are some real HIPAA issues with web crawlers in EHR.

Topic 5: Any other thoughts on the current use of EHR data? 

Creating registries for reseach and quality. Have worked on this for CKD successfully

Need to teach HCP value of EHR data and focus on clinical outcome, not just data for data sake

Developing tools to analyze both structured and unstructured data will be incredibly valuable

Really need tools to make data collection passive. I'm not slowed down @ bedside,you get your data

Painful doing manual abstractions for research in #EHR feels like little progress from paper#decretefields

Your point well taken & key 2emphasize. he goal is actionable data which will better Rx/prevent disease

I think about data libraries and a place to house all the structured data once it's clean. But may violate HIPAA 


"The Business Case for Using Unstructured Text Analytics for Critical Decision Making"

Comments on Statistics about Unstructured Data

Good question. Do you know how much health data is unstructured today?

We're talking about how much HIT data is currently unstructured & how to address that.

Said to be 80% , but that's a quess RT @jonmertz @Greg_Meyer93 Good question. Do you know how much health data is unstructured today?

I’ve seen estimates that up to 90& of healthcare ‘data points’ are unstructured. Is that true?

I'd also say 90% of collected data is never used again

Of 1.2 billion clinical docs produced-US each year, approx 60% valuable data unstructured documents

And what would any #healthit discussion be without some chatter about 'Standards?'

A ‘what if? Might single payer enable a ‘standards’ agenda? Or pluralism worshipped at any cost?

The industry really need standards defined. without that we will have this discussion forever

We have standards.... just may need to be more firm than fuzzy.

That there are no standards is a common complaint. It usually indicates lack of knowledge OR lack of implementation

Agree! No shortage of standards. Some are even worthwhile.. Some result from confusing activity w/results.

Standards are one thing: but anything can be structured with any standard. Get to structure first, then apply standards.

Which we won't have until we start thinking message sets as opposed to documents (so 1950s like carbon paper). (Forgot)


Even without standards master data management can help


Could addressing unstructured data be elementary for Watson?

Any #watson projects happening in this area? Sorry if someone already mentioned it.


Couldn't that be a #watson job? dealing with legacy code?


The installed base of legacy code prolly won't go away under Single Payer. Still need to process data? 

Closing Tweet

Can't forget ultimate end user... The patient. "How will this data help them be healthy in future?"



No comments:

Post a Comment

Note: Only a member of this blog may post a comment.