Dashboards

[Graham]

Does anyone think that a dashboard like this (http://www.ve.com/kavachart/prod_va_demos.html) one ( see the patient management demo ) is useful in other than acute inpatient care? I'm wondering about making something like that for my own use ... the original chart comes from The Lancet 1994.

[mel]

Graham, in the US, the gov and private insurances (that pay our wages) require that our patient notes follow a standard format.

[Graham]

Perhaps you have misinterpreted the purpose of the dashboard? It is a tool that summarises a patients treatment so that trends are discernable at a glance. In this case, it compresses 30 years of data onto one screen. In other words it mines all the patients data and reports it graphically.

My feeling is that it is mainly of use in acute care ... but looking for alternative viewpoints :)

[mel]

Perhaps you have misinterpreted the purpose of the dashboard? It is a tool that summarises a patients treatment so that trends are discernable at a glance. In this case, it compresses 30 years of data onto one screen. In other words it mines all the patients data and reports it graphically.

My feeling is that it is mainly of use in acute care ... but looking for alternative viewpoints :)
Inpatient setting, 30 years of data! :eek: :rolleyes:

[Graham]

Yeah ... we like to keep patients in hospital till we're absolutely sure they are well enough :D

But seriously, I have some patients with 15 years of serial data ( obtained thru hl7 ), and I could make the effort to present this data in this way. But it's going to be not an insigificant effort to do so. The point I'm wondering is if I gain anything ... apart from some pretty charts that I can present at clinical meetings. :cool:

[dawso007]

The point I'm wondering is if I gain anything ... apart from some pretty charts that I can present at clinical meetings. :cool:

I think that graphical representations of data for long periods of time would be worthwhile.

I am currently using a SOA hospital based med records. It basically logs inpatient and outpatient episodes separately. As far as I can tell, it would take a considerable effort to graph years of data across both settings.

On the other hand it does very well graphing all of the data for a particular hospitalization.

GD

[dawso007]

While I am at it I would like to add a couple of other EMR observations.

1. The EMR does a very good job of recording information - but it does not necessarily make it more accessible. On patients with complicated problems you can be facing hundreds of documents online that need to be read. Managing all of this information as free text is becoming a nightmare. What is really needed is some kind of ongoing data mining process that summarizes all of the key information - both graphics and free text.

2. Estimating information flow (for lack of a better term). All of these documents on the EMR are optimized for billing and coding purposes. As far as I can tell - nobody has estimated how much information needs to be discussed by a patient and a physician to solve a problem and how that should be recorded to optimize information flow. Without that basic engineering we are using fairly sophisticated computer systems as word processors.

GD

[Graham]

The dashboard aims to summarise multiple clinical domains in one screen so that you don't have to look at hundreds of screens.

The trick is to determine what variables we need to gather which I guess is why capturing encounter information as free text is ultimately a bad idea.

It is just still not clear to me that it will affect decision making though it may improve one's understanding of where the patient lies.

[BigDoc]

it is difficult to determine what to include where on the Dashboard, if you have too much info, it becomes too busy and useless.
We look at vitals on all patients daily, on ICU, I will want different variables from patient admitted for woundcare, so, how are you going to tease out what goes where?

[Graham]

I guess you would have a set of preferences to decide upon which variables you would want displayed. In the same way you get a plot of serial bloods, you have to choose which variables you want ... and just store this preference for each patient.

So, for a diabetic patient, my dashboard would include weight, Hba1c, glucose, renal function .. but for my rheumatoid patient, I would follow HAQ, DAS28, ESR, CRP etc.

[Graham]

For those mildly interested in this subject, here's the orginal article scanned in by one of the authors

http://www.edwardtufte.com/tufte/lancet_p1

[dawso007]

I just did an N=1 experiment with the EMR yesterday and today on a complicated patient with cardiovascular disease, cerebrovascular disease, endocrine diseases and several surgical conditions. There were 500 pages of text to review. The original H & P took about an extra hour. For comparison, I went back today for an additional 4 hours and read through all of the pages and reconstructed an additional PMH. The 4 hour PMH is clearly superior - there are no holes or question marks in terms of the time course of symptom development or why certain procedures were done.

Does anyone make an intelligent program that can scan the EMR and pull out data rapidly without the need to read every chart and note?

My fantasy would be a data mining program on a jump drive that would rapidly search a particular record and display it. Would result in rapid H & Ps by a simple cut and paste into your note.

If it isn't out there somebody needs to invent it. Don't see why dashboards could not be included.

GD

[Graham]

Basically anytime someone adds a new diagnosis, or a treatment, etc to a chart, they should be adding it to a list included in the chart so that the chart does not have to be mined in this way.

In an ideal world, each diagnosis would be accompanied by some text describing how the diagnosis was made, by whom, and using which criteria. A confidence level should also be assigned to a diagnosis similar to evidence levels for clinical trials.

[dawso007]

Basically anytime someone adds a new diagnosis, or a treatment, etc to a chart, they should be adding it to a list included in the chart so that the chart does not have to be mined in this way.

That's the problem.

When you have docs who are focused on a rather text intensive document because it has to cover CMS bullets for a particular billing code, there seems to be little time left over for more documentation.

The problem list tends to be ignored.

Search engines, operating systems, and bibliography software (EndNote, etc) - encourage searching rather than trying to parse everything into separate folders. As I look at nearly 40 GB of documents on my hard drive, I can see where the searching strategy seems better.

I have thousands of folders and it is fairly easy to misplace a document.

I think a data mining application for the EMR would really fly off of the shelves.

GD

[Graham]

That's the problem.

When you have docs who are focused on a rather text intensive document because it has to cover CMS bullets for a particular billing code, there seems to be little time left over for more documentation.

The problem list tends to be ignored.


I'm not so familiar with the US way of doing things, but don't you have to allocate blood tests to a diagnosis in creating the superbill .. so you must have a problem list somewhere.



I think a data mining application for the EMR would really fly off of the shelves.

GD

Quite likely but the problem I see is that each EMR stores it's data differently, so you would have to create the data miner for each EMR. And if the data is encrypted on the server, you're out of luck.

And since any serious EMR is a client server application, your USB thumb drive application won't have any rights to access the server/database tables. ASP - even harder.

[Graham]

What sort of variables would people like to see graphed as a dashboard view of the patient's data?

I have so far: ESR, CRP, DAS28, WBC, Hb, ALT, AST, HbA1C, glucose, Systolic BP, Diastolic, BMI ...,

[Phillip]

GFR, some tumor markers in some patients/practices (CEA, CA125, etc).

[Graham]

I've solved the problem by allowing users to add their variables to graph .. on my dashboard of 36 minigraphs (http://www.edwardtufte.com/bboard/q-and-a-fetch-msg?msg_id=0001OR&topic_id=1&topic=)or sparklines.