June 2016

Dive Into Data, Save Lives
By Dava Stewart
For The Record
Vol. 28 No. 6 P. 10

By making EHR data usable, health care organizations can improve patient outcomes.

The basic concept of EHRs is appealing to doctors, patients, researchers, public health officials, policymakers, and anyone else who may have interest in being able to easily access the health information of individuals or populations. Unfortunately, too often the vision of how EHRs can transform care has not crossed paths with reality.

Massive amounts of data must be manually and correctly input into systems, and then stored and extracted before they can be used effectively. Clinicians are frustrated, and government policy complicates matters, but there is good news: EHR data are being used to improve patient outcomes, and there's still plenty of untapped potential for further advances.

Making Data Useful
The frustration surrounding EHR data springs from several sources, including how the data are entered and stored and the difficulty of extracting the information in a usable form. According to Brian Doty, leader of the provider analytics and digital health practice at Deloitte, the question of how organizations can convert EHR data into something usable "gets right at the heart of what all organizations, especially those that have invested hundreds of millions of dollars, are trying to figure out."

Doty explains that EHR fields and notes contain both structured and unstructured data. "The focus up to this point has been on using the structured data," he says, adding that natural language processing and other Big Data technology offer promise when it comes to mining unstructured data.

Several factors influence an organization's ability to successfully access EHR data. For example, how data are input into a system determines in large part whether they will be accessible later. Although structured fields appear simple, if they are being used differently, extracting the data will not be easy. For example, if 80% of clinicians utilize an EHR's blood pressure measurement field, but 20% dictate the measurement into the notes, then accessing that data across the patient population will be problematic. "A lot of organizations have multiple EHRs, and that gets even more complex," says Richard Parker, MD, chief medical officer at Arcadia Healthcare Solutions, adding that having to pull information from two places within each EHR can be daunting and expensive.

Analytics forms a piece of the data puzzle as well. Accumulating data, no matter the storage method, is one thing, but being able to convert them into something useful is another. "That's where we see analytics fitting in," says Joe Warbington, a health care market development director at Qlik. "You can have as much data as you want."

Diving deep into data to make sense of care patterns and identify potential health problems touches on the potential latent inside EHRs. Rick Toren, president of Atigeo, says the idea that each EHR is a singular record is misleading. In the past, patients had individual charts at each physician they visited along the care continuum. Today, multiple providers can contribute data into the same EHR, which makes it difficult to find the most critical and relevant patient-centric insights needed to make care decisions. "The challenge—and what makes it difficult for vendors and everybody else—is you have to have a language translation platform that can take information from all those different sources and parse it," Toren says.

In an effort to defeat this roadblock, software vendors have developed tools that allow clinicians to search EHR data via natural language processing. Just as important, health care organizations are beginning to invest the appropriate resources into developing staff with specialized EHR skills. Bill Fera, MD, a principal in EY's health advisory practice, says an EHR is a tool, and using it effectively requires increasing levels of expertise. "Getting information out for reporting is a skill set, and combining EHR data with other internal data sets, such as from financial systems, as well as external data sets such as patient satisfaction information, is another level of expertise that requires investment," he says.

Reliability Issues
Clinical reports are only as good as the information used to create them, making it imperative that steps are implemented to ensure data are reliable at every stage of the process, from algorithm creation to implementation.

Providing clinically led, appropriate training during implementation helps users understand both why reliable data are critical and how to ensure their accuracy. Fera says data governance is directly related to data quality. "Sometimes people talk about [data quality] as if it is only an issue such as an error in transmission, but a lot of data quality issues are related to data entry and not following the protocols to enter the correct data into the system in the first place," he says.

"We have to face it—the data are only as good as what is entered," Warbington says, noting that once so-called dirty data are identified, clinicians must decide whether to make corrections or omit the erroneous information. Relatively minor errors such as incorrect codes or misspellings are forms of dirty data that clinicians may want to correct, he says.

Interoperability, a popular buzzword throughout the industry, is key to producing reliable data. Aggregating data without corruption is more complex when multiple EHRs are in use across an organization. "The more interoperability there is, the easier data aggregation becomes, but it's going to be a long time before it's easy," Parker says.

Doug Cusick, chief growth officer at Atigeo, says "sometimes [clinicians] are making life-and-death decisions" based on the data they have at hand, adding that there are Big Data platforms available that allow users to immediately discern the quality of the data.

Toren says data must be not only accurate but also comprehensive. "One of the challenges you have is understanding whether or not you are looking at all of the data," he says.

Nancy McMillan, PhD, PMP, a research leader in the advanced analytics and health research department at Battelle, a nonprofit research and development organization, says identifying outliers or unusual values can help ensure that data are reliable. "You can look for unusual things in advance of using the data. You can do range checks or look for sudden spikes in an otherwise normal pattern," she says.

Plugging Holes
From patient outreach and monitoring medications to helping physicians improve performance, there are many ways in which data can identify and fill gaps in care. At nearly every level of the health care system, there are uses for the data contained in EHRs.

For example, a Battelle team led by McMillan developed a program that uses EHR data to predict acute kidney injury (AKI), a risk that occurs frequently in critical care settings. For example, patients who undergo a procedure involving dye have an increased risk of AKI. "Science and technology can be used to support the health care environment, but if there's a problem you can't do anything about, it's not helpful," McMillan says, noting that because AKI is a known risk in the critical care setting, it made sense to focus on it.

Algorithms can be used to monitor complex patient information. "The AKI algorithm Battelle developed is an example of one such algorithm that has been shown to have high sensitivity and specificity for identifying when a hospitalized patient is at risk of an AKI event within the next 24 hours," McMillan says. Sensitivity refers to correctly predicting when something will occur, while specificity has to do with when to send alerts (this helps prevent alert fatigue).

Retrospective analysis of EHR data can also fill gaps in care. "[Organizations are] looking at records to see procedures or medications for a patient, then analyzing the full encounter or episode of care, evaluating the outcome, and drawing correlations of both positive and negative outcomes," Doty says. Clinicians treating patients with chronic illness can "see what treatments were provided and what the outcomes were, then compare those with instances when the outcome was demonstratively different," he says.

Patient outreach initiatives can mine EHR data to improve outcomes. "The EHR data can augment and strengthen the claims data by helping offices understand what patients are late for what procedures and who is scheduled to come in the next week. Then you compare it with the claims data," Parker says. "It's a powerful way to make sure patients are getting what they need." For example, by analyzing EHR data, staff members can reach out to patients who have not received necessary procedures.

A less common but nevertheless beneficial use of EHR data is as a mechanism to examine variations in care among providers. Parker says in an office with 20 physicians, there will be a few who are receiving high patient satisfaction scores, staying on schedule, and generally doing a good job. On the other hand, there will be a few at the opposite end of the spectrum, with the majority somewhere in between. EHR data can provide physicians with insights into how efficiently they run their practice compared with their colleagues. Because physicians are data driven, Parker says having this sort of information available may make them more willing to correct poor habits.

At the research and development level, EHR data have become invaluable. For example, according to Warbington, a leading private research university is using EHR data in a clinical study of patients with COPD. He says the research team "has a set of analytics that wrap around that, and they are showing the full 360 degrees of the patient and how they stack up against their cohort."

Measuring Success
Once an organization has implemented an EHR, taken all measures to ensure the stored data are accurate and reliable, and selected and invested in an analytics strategy, there must be a method to determine whether it's generating the desired results. Measuring patient outcomes may seem to be a straightforward affair, but several key elements must be in place.

For example, what is the organization's definition of success? Benchmarks can help. The source of the benchmarks may vary. An organization may choose a set of internal criteria or incorporate those developed by the Centers for Medicare & Medicaid Services. Either way, taking the first set of measurements to determine the starting line is the only way to determine real progress. "Basically, you've got to be able to measure where you are starting from," Parker says. "Then you remeasure at three, six, and 12 months and see how much you improved."

When describing how success is measured in the AKI program, McMillan says, "Real measure of in-use success will depend on comparing rates of negative outcomes, that is, AKI before and after the decision support system is put in place." The Battelle team measured the number of AKI incidents, implemented the algorithm, and regularly compares results to the original measure.

Once the program produces its findings, then the real work begins, McMillan says. "Real success depends on the care team acting on these alerts and its ability to mitigate risk within the 24-hour window this tool provides," she says.

For an analytics-based program to be successful, all of the stakeholders must fill their roles effectively. Parker says the analytics team's relationship with IT is paramount. "We are going to be very accountable in showing you the progress, but it's going to be a joint venture. It's a clinical/IT marriage. The clinical part of the medical operation has to be functioning at a high level, as does the IT operation," he says.

Analytics teams that work hand-in-hand with clinicians as programs are developed and implemented increase their chances of long-term success. McMillan says that learning how the caregiving community preferred to have decision-making support tools integrated into their environment was a huge factor in the AKI project. "Without the feedback, you are going to develop tools they don't need," she says.

"One of the biggest hurdles is physician adoption," Fera says, adding that garnering physician buy-in helps everyone in the entire organization understand what is changing and why. While there are several factors that determine an analytics program's success, Cusick says "the most important thing is to integrate back to the workflow. You don't want to be disruptive, because the clinicians won't accept the change."

Warbington believes analytics teams must keep in mind that different organizations have different needs. "They need a platform that can grow with them and that is customizable and personalizable," he says.

While some in the health care community have been dissatisfied with the performance of EHRs, the technology's ability to influence patient outcomes may be a game-changer. As analytics efforts make the data manageable, this is becoming a commonplace occurrence at more organizations.

— Dava Stewart is a freelance writer based in Tennessee.