Put the shovel down

Four steps are necessary to extract oneself from a hole:

1. Recognize you are in a hole

2. Decide you want out of the hole and

3. Put the shovel down.

4. Plan and execute an exit strategy

The hole (nay, canyon) we here argue that we must recognize traps us, is the one dug by the electronic medical record (EMR). Without doubt, the EMR was introduced with great promise¹. However, rather than improve the critical care we deliver, the EMR has hard coded 100+ year old workflows that are no longer optimal. Yes, typed characters are now legible, but from the perspective of one of us old enough to have begun their career using paper and pen, the content of most notes is categorically worthless². Yes, data show up from the labs and the bedside monitors and maybe some stand-alone bedside devices, but the flow sheets are inconsistently nested, data entry is variable, and situational awareness is difficult at best³. Patients are not benefiting and perhaps are being harmed. Step 1 achieved: we are in a hole.

Step 2 is to decide to get out. No one can argue that current critical care (regardless of the substantial strides made in the last few decades) has reached peak clinical efficiency and therapeutic perfection. “Dystopia” may be a bit strong, but given the remarkable progress brought by automation over just the last decade (think, GPS navigation, autonomous vehicles, and handheld computers that also work as mobile telephones⁴) by comparison the use of the EMR to instantiate the ancient medical workflows the word dystopia seems quite rational. “Burnout” is believed, in some significant part, due to the EMR⁵. We are asking you to deliver the same message.

In this issue of PCCM, Sauthier et al. compared humans and computers performing the same computation using data from the EMR⁶. First the obvious: after a display of biostatistics prowess, they found (a) computers are more accurate and (b) sometimes the EMR has wildly bogus data. More surprising is why we are interested in automating a score developed 20 years ago still using the same Delphi method chosen variables and thresholds⁷. Are computer scientists still talking about Y2K? Are Google’s search results generated from 20-year-old data using 20-year-old algorithms? Yes, considerable thought went into creating PELOD. But every year to treat thousands of critically ill children, we employ new diagnostic and therapeutic techniques, collect exponentially more data, and sometimes even understand what we’re doing better (lung protective ventilation was first described in the past 20 years). The impressive work that went into engineering PELOD should be reapplied every year, or better yet, continuously. Outside of fields like EMR software and the government, with distorting incentives and lack of competition, innovation is the reality. If we are not innovating, we are (or should be) innovated away (long live Friendster!⁸). Spurred by the HITECH act, EMR vendors focused on expanding market share, meeting arbitrary and temporary “meaningful use” requirements, and helping hospital and physician billing. A leading EMR vendor did not even allow clinicians to search for text in a patient’s chart until 2015. The irony is that real EMR innovators such as Vanderbilt and Partners were squeezed out by HITECH because it was too expensive to meet meaningful use requirements. Please put the shovel down and move to step 4.

Rajkomar, Dean, and Kohane envision medical care where “every diagnosis, management decision, and therapy [are] personalized on the basis of all known information about the patient, in real time, incorporating lessons from a collective experience⁹.” This vision is possible, but not if we remain locked in combat with the EMR vendors on their terms. The path out requires freeing the data from the EMR to enable innovation on our terms for both discovery and delivery of care. On the discovery side, hospitals are racing to create data warehouse platforms for precision medicine. Finally recognizing the avalanche of innovation sparked 25 years ago by the release of the open MIMIC dataset¹⁰, researchers have realized the value of data to gain physiological insights, improve patient care, and create new products. Sadly, not all these efforts are as open as MIMIC. To learn from the collective experience, we need to share data. This requires trust in deidentification, technical innovation (for example federated data warehouses where algorithms are shared instead of data), and even rethinking the ethics of clinical data use in a learning healthcare system^11,12. Data will drive future innovation in pediatric critical care, so pediatric intensivists need to lead on these issues by pushing for greater data access within institutions and supporting collaborative efforts like the PICU data collaborative¹³.

On the delivery side, Mandl and colleagues lead the effort to open and standardize access to live data within EMR with SMART and FHIR¹⁴. By creating an architecture where interchangeable “apps” access patient data through a standardized APIs, innovators will be able to deliver new functionality broadly, and providers will be able to pick and choose the best components to match their own workflows. Progress has been slow as it took a nudge from Congress to get EMR vendors to even condescend to participate. There are technical challenges too dealing with EMR’s internal mess. Every hospital’s EMR implementation is customized practically from the ground up, ostensibly to best match the hospital’s workflows. The consequence is that every EMR implementation (even from the same vendor) has different internal names for labs and flowsheets (and different misspellings too). Moreover, EMR’s have incomplete concepts like ventilator settings, forcing each hospital to create their own schemas for documentation. These are solvable problems (for instance there are only a finite number of ventilator manufacturers) but will take continued effort.

So, let’s put the shovels down and turn off the backhoes. We’re deep in a canyon, but there is an extraordinary world of possibility above us. Together we can move beyond the limitations of the EMR to put data to work improving patient outcomes.