Too Much of a Good Thing?
By Susan Chapman
For The Record
Vol. 26 No. 3 P. 14
Insiders agree that a focused approach to data analytics provides greater benefits.
As health care swiftly moves from paper-based systems to EMRs, there’s an accompanying shift toward accountable care and population health management. This transition and proliferation of available electronic data are causing health care facilities of all sizes to strongly consider developing data analytics programs. However, industry experts caution it’s not a venture to be entered into lightly nor should it be so encompassing that its value diminishes.
In the past, hospitals weren’t concerned with tracking patient outcomes or the overall health of the populations they served. Hospitals billed for services rendered and received payment. This system offered no incentive to be proactive or understand the health of an individual patient or patient population holistically. Today, hospitals are entering into risk-based contracts, which offer new incentives to manage health and its corresponding costs, a model much like that of payers.
“In the way hospitals got paid in the past, which was fee for service, there wasn’t much need for population analytics,” says John Hansel, vice president of health care provider solutions at MedeAnalytics. “That old model is shifting, though, as everyone is taking on more risk. As hospitals begin to manage populations of patients, they begin to act like payers. To succeed in this environment, health care providers need to adopt the type of sophisticated analytics that payers have been using for decades.”
Both health care providers and payers have spent considerable time and money building enterprise data warehouses, which allow for information to be stored in one centralized data repository that then can make information available for analysis. While the data may seem rich and ripe for the picking, mining efforts can turn messy, Hansel says.
“Health care providers just aggregate data in these big repositories and don’t necessarily know where to direct it all,” he says. “Analytics can sit on top of these warehouses, but there needs to be processes that can turn the information into usable solutions. The problem is that all the new clinical information simply does not lend itself to traditional analysis. For instance, much of the data in an EMR are unstructured text and notes, which are not organized for categorical analysis. This makes the reporting requirements for meaningful use stages 2 and 3 increasingly difficult for health care providers. To do analytics correctly, there must be standardization of data inputs, and no one is enforcing that right now.”
Despite the proliferation of EMRs, the information can’t be easily translated into a form that can be analyzed because the data often are simply electronic versions of paper documents. Therefore, many facilities, much like payers, are using billing and claims information to perform analyses. “There is a lot you can do with billing and claims information, but that is not fulfilling the promise of this incredibly rich EMR data, which aren’t very well structured for analysis,” Hansel says.
Juergen Fritsch, chief scientist for M*Modal, who disagrees with the idea that only structured data are useful, views natural language understanding (NLU) as a way to codify important information that isn’t available to be analyzed in raw form.
“When health care organizations launch an analytics program, they often overlook the importance of unstructured data,” he notes. “While it is important to look at the structured data, what we have seen is that there are a lot of insights to be gained from unstructured information—for example, discharge summaries, progress notes, and other narrative notes. Those pieces of information are sitting in the EMR but are not accessible through standard analytics packages. In the best case, hospitals might miss some critical things. In the worst case, a physician may take the wrong course of action with a patient. One good example is something that Seton Healthcare Family [in central Texas] was able to do. Some congestive heart failure patients kept being readmitted to the hospital after being discharged, yet no one knew why. The answer wasn’t in the structured data; it was in the patients’ social histories and their living conditions. Using NLU to process the notes into a form that could then be analyzed, the hospital was able to identify the contributing factors and is now proactively addressing them.”
Proceed With Caution
Benjamin Loop, vice president of care coordination and analytics at Siemens Healthcare, believes providers must analyze all pertinent data but at the same time be patient as more thorough data analytics programs are developed. “Payers are doing more complex analytics but in a simpler environment than that of the provider world,” he says. “Providers need to understand that they face certain barriers and why those are what they are.”
According to Loop, several factors present challenges to fledgling analytics programs. Among them are inflated expectations. “It’s important to manage expectations around what these programs can do,” he says. “Often, we’re called into board meetings and asked to talk about what is going on in Silicon Valley, but those large tech organizations do not compare to what is going on in a specific health care facility. Therefore, we need to be clear about what executives and board members can expect—and what they need—from analytics programs.”
Dale Sanders, senior vice president for strategy at Health Catalyst, agrees: “The term ‘Big Data’ originated in Silicon Valley from the work of Google, Yahoo, and the other active members of the LAMP [Linux, Apache, MySQL, PHP/Perl/Python] open source community. It refers to Hadoop, MapReduce, and other tools that were developed specifically to address the scalability problems that relational database engines couldn’t address. Unfortunately, a lot of folks in health care immediately assumed that Big Data was the silver bullet to our data analysis needs, but it’s not. For one thing, the volume of data in the health care industry pales in comparision to the data that are collected and analyzed by companies like Google.
“We do not have any problem scaling relational databases to meet our needs, which means we can leverage existing skill sets that are in much greater supply,” he continues. “Also, the nature of the data that we currently have in the ecosystem of health care is not structured to leverage the full capabilities of Hadoop and MapReduce. There is nothing compelling about the value of Big Data technology in health care right now—in five to 10 years, maybe, but not right now—so don’t be distracted by it.”
Another imperative for health care providers is the need to align business goals with a measurement framework. “For instance, organizations need to ask themselves what is our intention, what do we want to do with the information, and how will it make an impact on our organization and its performance? How will we measure success?” Loop says. “Organizations need to link these themes into a comprehensive approach and strategy.”
Loop believes that health care organizations sometimes are trapped by unproductive arguments about what health care data mean. Take, for example, when hospitals measure how many emergency department patients were seen in one month. “Such a snapshot can be taken at the end of the month or once all the charts are coded. Both options can be helpful as long as they are linked to the data’s purpose and why the organization is asking the question,” he says.
Go After Pertinent Data
Fritsch says that in order for health care facilities to build effective data analytics programs, they must gather both the right amount and the right type of information. “I would advise that organizations make sure they are collecting and consolidating the right amount of data based on what questions they need answers to,” he says. “Many hospitals are jumping to analytics and data tools without knowing if they have the right kind and amount of data in electronic form in the first place.”
Fritsch notes that nearly every HIT vendor has an analytics product, which means health care organizations must be sure they’re getting solutions that fit their needs. “Analytics is the big buzzword in health care today,” he says. “Facilities need to do their homework. The bigger, more advanced hospitals tend to do their own homework with people on staff who are experts in this area. For the vast majority of other hospitals, they must look for help from industry experts.”
When it comes to analytics, Fritsch says every hospital has slightly different needs and interests. Each is trying to understand whether it’s using resources appropriately. Consequently, hospitals must make choices and build programs to suit their respective requirements.
“One of the challenges is to get health care to think of starting with the basics, solving problems that we should have been solving many years ago,” he says. “There is a lot of potential for savings and efficiency improvements going after the basics before getting tied up chasing after the more headline-grabbing analytics that are harder to benefit from. Using the data and technology that is available today, organizations can make basic delivery of care more efficient and eliminate waste.”
Ordering, workflow, and adverse events are among the basic care components ripe for data analytics. “Facilities waste a great deal of money by ordering tests and procedures that have no added value and detract from health,” Fritsch says. “In terms of workflow, facilities should be looking closely at simple things—how much and which type of labor it takes to deliver care, for example. It’s more challenging to identify these issues because it’s hard to track the movements of people through the health care delivery process, but it’s necessary to do in order to make care more efficient. Finally, we need to look at such things as adverse patient events: readmissions, safety-related falls, and hospital-acquired infections. These are fundamentally important to improving health care. Once we identify these events and the factors contributing to them, we can use predictive algorithms and prescriptive preventions to intervene when a patient is on that trajectory.”
Building a Foundation
A cautious approach and clear objectives should serve as the framework for any data analytics project, Loop says. Facilities must identify why they’re making such an investment, determine how the information will be used, and create overall goals. Organizations also must build a culture of trust among the technical and operational teams, a time-consuming process. Experts also recommend creating an analytics maturity model to illustrate where the organization stands and where it hopes to go and then periodically benchmark organizational capabilities.
“If health care organizations don’t do these things, there will be a lot of missed expectations, wasted investment, and finger pointing after the project inevitably fails. There is infrastructure that needs to be built that helps you proceed along the path in an intelligent way and not just blindly,” Loop says, adding that facilities must focus on business purposes and build a team to support that level of analytics.
“I always encourage people to treat this as a journey,” he continues. “There are multiple signposts along the way. It’s a project they should approach thoughtfully and deliberately, and they should do it together as a team. That type of approach allows people breathing room to take a smart approach on a very important venture.”
Sanders favors a similar model. “A group of us collaborated over the past several years to develop an analytic adoption model for health care. It’s like a course curriculum for the progressive adoption and utilization of data to make better decisions about quality, cost, and risk management,” he explains. “The model addresses the very specific data governance as well as technical steps required to achieve personalized health care. Everyone in the field is trying to learn and build from past mistakes, which were characterized by a scattered approach to analytics. We borrowed the idea from the HIMSS EMR Adoption Model and applied it to the adoption of enterprise data warehouses and analytics. The model will help organizations build their own analytics road map as well as evaluate vendors’ offerings. Vendors can use it to influence their own product road maps, too.”
No matter the size of a health care delivery system, its data needs and challenges are relatively equal. “What we need to figure out as an industry is how we make analytics for smaller organizations accessible and affordable,” Sanders says. “There is so much that is really busy around health care technology. Lots of organizations are still implementing EMRs, health information exchanges, and the conversion to ICD-10—all of these things combined create a very heavy burden on the technology teams and infrastructure for every organization. Adding the need for an analytics system compounds that already heavy burden. Everything that we do for our clients, whether large systems or small facilities, is aimed at making the adoption of analytics as fast, efficient, and pain free as possible.”
According to Fritsch, tackling analytics isn’t for every hospital. “There are silos all over,” he says. “Small hospitals have their own databases and cannot share their data very well. For analytics to be meaningful, you need to consolidate and aggregate the data into large pools. Payers are able to see the broader stretch of data across many hospitals because they are already doing this.”
Still, Fritsch believes that if a facility wants to improve itself, internal analytics help. “If they really want to improve patient care while managing their cost, then they need to share information and partner with other hospitals in the area,” he says. “It is not cheap to set up this kind of system, and a lot of the smaller hospitals don’t have the budget, capacity, or data required to perform meaningful analytics on their own.”
Fritsch says federal initiatives such as meaningful use, ICD-10, and value-based reimbursement models are driving analytics and catapulting a push toward a new business model. Taken together, these factors are helping hospitals determine what is financially viable. “We need better insights into the patient population,” he says. “People who are doing that have reported positive results in their patient mix and reimbursements. Executives are starting to realize that there is something to be gained from analytics in terms of getting a step ahead of government regulations and other requirements affecting the health care space soon.”
Experts envision analytics taking several shapes and becoming more widespread. Just as the Centers for Disease Control and Prevention and the World Health Organization can mine Internet data to better understand health patterns and epidemics, progressive technology will be able to analyze Twitter updates and Facebook posts to determine where flu outbreaks are occurring in real time.
In the immediate future, however, hospitals must build a basic competency in data analytics to manage the risks of their respective populations. Because there are many models and proven guidelines to follow, providers can construct their programs on an existing science while enhancing them with their own unique data. The most effective programs will solve business problems and answer specific questions that will advance the industry and improve patient care.
— Susan Chapman is a Los Angeles-based writer.