EMR adoption is skyrocketing, in no small part due to government incentives. The office of the national coordinator lauds this hockey-stick curve as a success. Advocates promise electronic records will improve patient care, reduce mistakes, and save healthcare costs. At the same time, doctors love to complain about implementation cost and poor usability. How can we reconcile these differing opinions? The truth is they are describing very different technologies. EMRs, the way they are implemented now, will not accomplish these goals. In fact, early adopters can become stuck at a rudimentary level of functionality, and the extensive feature lists described by meaningful use criteria fail to address the most basic needs for patient care.
I have been at medical institutions at different levels of technological development. Each has a different attitude toward the EMR; for some its loathing, others longing. Some devote resources to try to improve it, but others give up. I realized the parallels with Maslow’s Hierarchy of Needs, people are motivated to attain something only after their very basic needs have been fulfilled. So are EMRs good or bad? Well, it depends on where you are on the hierarchy.
The figure above describes the steps to building a technology infrastructure that will lead to improved patient care. Yes, incentives help us achieve some very basic needs, but the problem is that decisions and investments we make now will determine the ceiling as well.
Step 0. Paper Charts
Explanation: We’ve heard the horror stories of interpreting handwritten notes and the errors that occur with handwritten medication orders.
Step 1. Make it digital
Explanation: The conversion from paper to digital charts involves electronic charting, order entry and documentation. The major advantage is accessibility.
An analogy: Research used to be done at the library with volumes of journals. Now we can search Pubmed and Wikipedia.
Are we there yet? Yes. All EMRs, by definition, offer this, and the majority of Meaningful Use Measures define digitization of different types of data. But you don’t need a commercial EMR to accomplish this—the most basic setup would be a series of Word documents on a shared hard drive (which many departments still use)
Step 2. Make it aggregate data (dashboard)
Explanation: Before this level, the data exist electronically, but are still organized in a way that mimics the paper chart. Here, data are displayed in dashboards, organized by usage (e.g. admitting, covering, discharging, etc) or thought process (e.g. hemodynamics, infectious, etc) rather than how they were collected.
Are we there yet? Not quite. Commercial EMRs have “custom views” that are cumbersome and limited. Microsoft Amalga, or homegrown solutions like the SMART platform at Harvard and MedView at Penn sift through EMR databases to retrieve and aggregate data (bypassing the commercial EMRs) to enable custom views.
Step 3: Make it communicate
Explanation: Doctors commonly request that their systems should “talk to each other” for better usability (e.g. single sign-on, pre-populating forms)(). Perhaps more importantly, poor exchange of data hinders new apps for quality improvement (Provider error prevention: online total parenteral nutrition calculator. Lehmann et al, Proc AMIA Symp. 2002:435-9).
Analogy: e-commerce websites can extract payment from Paypal, so users do not need to enter credit card information repeatedly.
Are we there yet? No. Meaningful use describes e-prescribing and secure messaging, but communication between applications is lacking (and therefore the capacity to build apps is too). The standards for data exchange exist (e.g. RESTful APIs), but EMRs are more interested in keeping data in a silo.
Step 4: Make it synthesize data
Explanation Existing EMRs flag lab values in a rudimentary way, using absolute thresholds. This leads to alert fatigue and misses the big picture. Instead, alerts can be more sensitive and specific by the following methods:
– using trends: a rising White Blood Cell count is more informative than the actual level. A low platelet count is less significant if it has been at this level for a long time.
– personalizing: hemoglobin ranges differ by gender, and creatinine differs by age
– using multiple variables: disseminated intravascular coagulation is often detected by hemoglobin, platelets, fibrinogen, and several other lab values
Analogy: Your bank tells you when there is suspicious activity.
Are we there yet? No. Meaningful use specifies simple decision rules, medication interactions, and patient reminders (for preventive health and followups). No EMR, however, utilizes trends in data, and some have argued they have no reason to .
Step 5: Make it predict
Explanation: If I knew my patient was going to get intubated or get started on antibiotics tomorrow, I might give him diuretics or start antibiotics today. Often in retrospect we see signs that should have alerted us to impending deterioration.
Are we there yet? No commercial EMR offers prediction, and they way they are structured actually inhibit development of these tools, using math models or machine learning because data are sequestered. There are implementations at my institution that alert for sepsis, acute lung injury, and acute kidney injury when certain lab and vital sign criteria are met, but there’s no platform to make this process scalable. Some companies are attempting to apply machine learning
Step 6: Self-actualization
Explanation: This is the point at which my EMR has become a collection of tools that enable my job as a doctor. Here we realize the promises of improved safety, outcomes, and costs.
Are we there yet? Perhaps one day