Why Healthcare Data Fails to be Information

And How to Build Actionable Insights from Healthcare Data

Written by Megan Bultema, Ph.D., Chief Data Scientist and Ben Doremus, B.S. in Electrical and Computer Engineering, Data Scientist


It is rare to find a modern healthcare system which does not purport to make “Evidence based decisions.” Even the most primitive surgical settings are flush with charts and graphs, and staff are inundated with multitudes of reports, suggesting the need for improvement. As is evident in practice, this overwhelming-yet-superficial presentation of data drives neither meaningful nor lasting change.

Underlying this inconsistency is the stark difference between using Data to narrowly guide decisions, and the real goal: compiling Information which gives insight into which decisions are best. It is easy to assemble investigations and disseminate spreadsheets, but just because clinicians have Data does not mean it can be interpreted as Information. In short, Data becomes Information when it has enough context to be actionable. Until the information becomes personally meaningful to a clinician, it is unlikely to drive any enduring change in their practice.

To fully utilize that information, it needs to be easily obtainable and critiqued by all relevant stakeholders. This points to an even higher goal; Information Transparency. The Lucian Leape Institute defines transparency as “the free flow of information that is open to the scrutiny of others” (Shining a Light). Transparent information has the power to permanently transform healthcare, at scale and with speed. However, there are some daunting challenges when striving to develop Information and Transparency in a surgical setting.

Compiling the Data

The initial struggles are often technical. Before we can display meaningful information, we have to accumulate all of the relevant data in one place. Despite the advantages that EMRs bring, there are still significant hurdles to overcome.

In large part, these issues are caused by the wide variety of data which needs to be brought together. These data are siloed in different parts of the organization, and OR staff themselves cannot compile everything they need to make decisions that are optimal in multiple dimensions. Critical data points can be spread between the central EMR used for clinical documentation, claims data, Supply Chain, Quality Analytics, and staffing software, with no obvious means for connecting it all. It is impossible to mine data for existing “best practices” without one resource that compiles and meaningfully interprets data from all of the relevant sources.

It is a heavy lift to compile it all in one place and, even when you are able to do so, there is still more work to be done. Combining the data sources together is non-trivial. Some information is pertinent to individual surgery cases, others to the hospital encounter, and still other information is centered on the patient. Finding the right level of aggregation for each element is essential to having data reflect reality, and to telling a meaningful story from which we can begin to identify actionable information.

By utilizing a wide data set that includes all the relevant metrics for the patient’s story, it is now possible to begin inspecting the correlations between various data elements and attributing causality to a surgeon’s choices and their patients’ outcomes.

Presenting Data as Information

With so much critical information, the right visualizations are critical to preventing information fatigue. Depending on where you are in perioperative services, different surgeon groups will have very different Key Performance Indicators that are relevant to their work. It is essential to be able to easily highlight the specific metrics which they find valuable, without distractions. As a guide, Edward Tufte has phenomenal resources for Information Design and the multitude of ways data can be presented.

Being able to compare surgeon’s results to their peers is a powerful way to instigate change, but it falls flat unless strong correlations can be shown between choices that they’ve made and their outcomes. Being able to quickly show the large impact of minute choices is difficult, but becomes much easier with the right visual tools that include the right data elements.

Separating the Signal from the Noise

Providing value in healthcare is a key focus area for many providers. To provide higher quality outcomes at reasonable prices, many hospitals are attempting to reduce unwarranted clinical variation. To truly target unwarranted variation, it becomes necessary to select groups of encounters which are clinically similar, to expose what variation is warranted and what is unwarranted.

The common practice of using procedures or DRG codes to group data is wholly inadequate for a meaningful inspection of surgeon practice. To effectively compare the different outcomes between different patients who underwent the same surgery, we must control for the variety inherent within the patient population, differences in severity, and any other complicating factors. At Empiric Health, we use the term “cohort” to represent these clinically comparable groupings..

While there is relevant information in diagnosis codes and Trauma levels, experience has proven that those fields are inadequate for accurately identifying these clinically comparable cohorts. In some cases, all of the discrete data fields that exist in modern EMR systems still do not express the level of detail needed to separate the data into similar cohorts. Without more sources of data that help us segment the encounters, it is impossible to distinguish the signal from the noise.

Fortunately, there is one last resource which contains a wealth of information. Dictated notes contain the critical nuances of patient presentation and surgical approach which are often not present anywhere else in the EMR. Unfortunately, extracting the information captured in these free-text notations is exceptionally difficult, and often requires deep clinical expertise paired with cutting edge computational methods in Natural Language Processing.

Information that Drives Change

At Empiric Health, our most essential, actionable information comes from analyzing clinically comparable patients, grouped as Cohorts. To define the Cohorts, we group clinically relevant procedures and applying thousands of rules that meticulously winnow out clinically different patients. What remains are exclusively similar encounters uncluttered by complicating comorbidities or other relevant complexities. Armed with these consistent Cohorts and all of the relevant metrics for performance, we finally have a window into which we can make meaningful comparisons between surgeons and weigh the different choices they’ve made on equal footing.

Through this massive transformation of Data we have created actionable Information. Once all of the caveats and reasonable inconsistencies are accounted for, we can see the patterns that lie under all of the noise. This sort of transparency represents the antithesis of drowning in irrelevant spreadsheets or incomplete dashboards. These sorts of tools naturally drive innovation, increase patient safety, and ensure optimal efficacy in the OR.

When armed with this concise and clinically meaningful view of surgical cases, clinicians can use this new level of Information Transparency to move their practice from the frustration of unusable data overload into actionable insights that drive better outcomes and better value for their patients.

Let Us Know

Where have you seen Data fail or succeed in becoming Information in your work? What are you doing in your role to help promote Information Transparency? To continue the conversation, email us at info@empirichealth.com or visit empirichealth.com.