Over two years ago, I wrote a little blurb on the subject, mostly in reaction to a NEJM article http://onhealthtech.blogspot.com/2010/03/ehr-checklists-bayesian-diagnosis.html

I am not certain that things have changed much since then, and I have a mild level of discomfort with the NEJM authors’ thesis. I am also aware of Dr. Weed’s work in this field (which has been commercialized, but not with much success…)
Do you see any current breakthroughs in this science, or are we still where we were back then?

Bayes RULES!

http://www.bgladd.com/Total_Information_Awareness/

“While the relative “accuracy” (sensitivity & specificity) levels of many clinical methods that estimate disease probabilities (or any type of experimental assay with anterior empirical underpinnings using Bayesian statistical methods … are tolerably well-defined (and uniformly well below 99.9%)… no test is infallible, there are inescapable trade-offs in terms of relative false-positive/false negative levels associated with any assessment.”

By probabilistic I mean an approach that will be based on bayes theorem and statistical principles which are needed in order to put some structure on vague patient complaints, borderline lab values and also for assessing the need of certain diagnosis tests or the importance of their results in the overall context of diagnosis.

The problem here, as you have mentioned is the chicken and egg…we certainly do not have enough statistical data to build a medical diagnosis process based solely on probabilities… (We will have this if we will gather structured data for couple of years and mine this data)

In my opinion, even without accurate numbers – the principles can still be used with rough estimates/simple models and help reach better decisions.

Margalit – this is something that really needs to be seen in a live setting, and I think there may be a site fairly close to you. While a canned and/or remote demonstration can be arranged, that simply can’t demonstrate the real effects at the point of care. Let’s message privately.

I know I’m pushing my luck, Dr. Oates, but is there a way I could take a look at new design. I am really curious now…. 🙂

This is a very valuable observation. The issue with structured data is the chicken and the egg dilemma. We could probably do something useful with lots of discrete data, but we can’t do much more than hypothesize until we have enough data, and people are not willing to go to the trouble of collecting it, until we show what we can do with it….

I am intrigued by the term “probabilistic”, Dr. Gal. What do you have in mind there, as far as data collection goes?

I agree with the points raised about UI/UX of EHRs and the ideas of simplicity, paper metaphors. Most of the EHRs that I saw in the last couple of years were problematic at that and that causes errors and slows the clinical process.
Dr.Oates outlined great design principles, some of which are even quite easy to implement. However, I believe that this is not a strict UI/UX problem – there is an underlying problem in the way medicine is currently being practiced that can not be solved with a good UI/NLP and etc. (Computers, unfortunately, are less artistic in their nature than physicians)

Using even an optimized EHR to capture the entire story of the patient or a semi-processed version of it in a “paper like” manner,
is not sustainable.Hints to that are in these two quotes:
“A simple visit…. I couldn’t keep up.”
“whenever you have two doctors in a room, there are at least three different preferences.”

A unified,logical and probablistic way for interacting with patients while capturing their data in a meaningful way is needed in order for computers(EHRs) to help us and with the abundance of clinical information, shortened doctor-patient visits, standards of quality and etc. this help is really needed.

Amen to this comment. But, as long as the needs of the few outweigh and just crush the needs of the many, good luck seeing progress.