Medical records have always been about communication. Communication with yourself across time, communication with other providers and yes, communication with billing and finance. Categorical data is necessary, particularly for the latter, but communicating uncertainty and alternative interpretations has always been an important aspect of medicine, and it is very difficult in the context of a modern EMR.
Informatics is in a hole with administration because most EMRs do not increase provider productivity. There is some evidence the EMRs increasing billing per encounter, but too often this does not offset the increase in time required for providers to complete an EMR note. Billable encounters per provider hour is a critical administrative metric. When informatics proposes enhancements and additions to the EMR, administration is thinking is “You have already whacked encounters per hour by 20% and now you want to add more?”
But, for all the enormous resources devoted to IT in a modern healthcare organization, a surprisingly small fraction of the compute power is actually devoted to the central task of communicating among providers about patient care. Most of the money going to IT is about security, billing, scheduling, inventory management and other necessary but ultimately administrative tasks. And most of the big dollars going to all those consultants is about maximizing reimbursement rather than optimizing care. By comparison, the task of hosting a form based web interface for providers to view history and test results and to enter EMR data is a trivial task requiring only modest compute resources.
We can do better. Much better. Advanced natural language processing should be capable of holding an intelligent conversation with a provider about a patient. Think Siri with a database focused centrally on the patient at hand. Note that this is a greatly restricted space compared to what generic Siri deals with now. Forget stock markets, sports scores or the weather, just hold a conversation about a specific patient’s medical conditions. Smaller scope of interest translates improved accuracy, basic multiple hypothesis testing. This would still require a lot of compute power you say, but if it increased billable encounters per hour by 10%, it would easily be worth $10/hr per provider, and for $10/hr you can rent half a dozen maximal configured high performance compute systems on the Amazon EC2 cloud aggregating to hundreds of CPUs with hundreds of GB of RAM and terabytes of SSD disk PER PROVIDER. That is a lot of compute power. Are you seriously telling me we cannot implement:
Siri: How is Mrs. Doe doing this month?
Doc: She seems to be doing well.
Siri: So the UTI for which you prescribed Macrobid has resolved?
Doc: Yes.
Siri: Her blood pressure has been close to 140 over 90 on the past two visits. Do you want to increase the dose of her Lisinopril or metoprolol?
Doc: How much metoprolol is she on now?
Siri: Extended release 25 mg a day.
Doc: Increase to 50.
Siri: Changing metoprolol extended release to 50 mg daily.
Siri: Her glucometer dump showed a median blood sugar of 138 with no readings above 252 or below 83. Continue her current diabetes regimen?
Doc: Yes.
Siri: Any new problems?
Doc: No.
Siri: Standard outpatient physical exam?
Doc: Yes, no concerns.
Siri: She is due for an A1c and annual diabetic eye exam. Go ahead and schedule those?
Doc: Yes.
Siri: So we are going to bill for a moderate to high intensity visit
addressing follow up on the UTI, hypertension management and diabetes management?
Doc: Sounds good.
Siri: Here is the chart note for your review and signature.
And perhaps at some point, Dr. Siri will be speaking directly to the
patient.
We need a vision of informatics as enhancing communication and productivity. No excuses for forcing people to work for the computer. With the compute resources available in modern environments, the computers can let the people work in the ways people are most comfortable and productive. And yes, we can do better on billable encounters per provider hour.