Skip to main content

The Analytics Challenge: Timing, Talent, Value, and Ethics

 |  By Jonathan Bees  
   August 30, 2017

Healthcare analytics is evolving to a greater focus on analyzing data using prescriptive analytics and providing proactive solutions.

This article first appeared in the September 2017 issue of HealthLeaders magazine.

Healthcare analytics is evolving from analyzing what has happened (descriptive) to anticipating what will happen given past data (predictive) and, in its most powerful iteration to date, expecting what will happen plus providing proactive solutions based on those predictions (prescriptive). So what comes next?

The next step in this evolutionary process is the application of artificial intelligence to healthcare data to help providers make the most effective decisions possible, both financially and clinically.

While AI is still in its early stages and has yet to be fully embraced by healthcare leaders, many believe that it offers tremendous potential for analyzing the vast amount of data generated by
the industry. 

At this point in time, the application of AI in healthcare is fairly minimal. According to respondents in the 2017 HealthLeaders Media Analytics in Healthcare Survey, for example, only 14% of respondents say that their organizations use a software platform that provides an artificial intelligence capability.

However, 35% of respondents say they don't currently have this capability but plan to within the next three years, indicating that there is potential for growth.

That said, AI is not for everyone, and 35% of respondents say that their organization does not plan to have this capability. A fairly large share—17%—don't yet know how their organization will proceed.

Steve Hess, the lead advisor for this Intelligence Report, is chief information officer at UCHealth, an integrated health system serving the Colorado area.

"At this point in time, the application of AI in healthcare is fairly minimal."

The system includes seven hospitals, 1,620 hospital beds, and more than 17,000 employees. He points out that the difficulty today is not access to data or having enough tools to analyze it—far from it; the real challenge is not having the time to extract analytical value from the data.

"We don't have a shortage of data. We don't have a shortage of dashboards. We have plenty of tools. What we don't have is plenty of time to analyze all of that data. Prescriptive analytics are what the healthcare industry needs to move the needle."

Clearly, the solution is not more tools, but more intelligent ones.

Financial and clinical analytics now

There are currently a wide variety of uses for healthcare analytics, an indication of its effectiveness and flexibility.

For example, respondents in our survey say that cost containment efforts (78%) and maximizing reimbursements (64%) are the leading uses for financial analytics now, with the next four responses falling in a tight cluster between 57% and 59%.

Note that compared with last year's survey, responses have increased for cost containment efforts (up 12 percentage points) and population risk assessment (up 9 percentage points), evidence of provider interest in continuing to bend the cost curve and applying analytics tools to manage the increased risk associated with value-based care.

The top responses for finance-related data drawn on now are both forms of claims data—Medicare/Medicaid patient claims data (75%) and commercial payer patient claims data (69%)—the same responses as in last year's survey.

This is a reflection of provider interest in analyzing the relationship between patient care and revenue. It is also worth mentioning that care partners' cost data (26%) and care partners' provider productivity data (23%) receive low responses, likely because of the
difficulty involved in gaining access to this type of data.

These data types will be critical as providers seek to understand costs and improve care across the continuum in the future.

The leading response by a large margin for clinical analytics use now is improving clinical quality (79%). Identifying gaps in care (65%) and lowering cost of care (60%) are second-tier uses. The top two data types for patient-related data drawn on now for analytics activity are clinical data from EHR (80%) and patient demographics (71%), part of an increasing focus by providers on analyzing patient care more deeply.

Two relatively new sources of patient-related data—social determinants of health (26%) and patient genetic data (9%)—are also expected to see growing analytics use in the coming years, in both population health and personalized medicine applications.

"The real holy grail is being able to look at all that data along with genetic data and really tailor treatment plans to the individual, not to the disease."

 

"I would call that precision or personalized analytics," says Hess. "Population health can mean so many different things. It could mean that you're just looking at utilization data, you're looking at claims data, you're looking at social determinants data, you're looking at remote monitoring data, and you're trying to intervene with a patient prior to them having to get treated in a more expensive care setting.

"But the real holy grail is being able to look at all that data along with genetic data and really tailor treatment plans to the individual, not to the disease. And I think the larger organizations, including us, are starting to invest in biobanks, DNA sequencing, and the pharmacogenomics data, and really trying to bring that data back into the EHR to impact the therapies and the treatments that an individual gets by virtue of them as an individual, not as a disease. So I think we're at the very early stages of that right now in the industry."

Hess says that UCHealth has been collecting DNA data through a partnership with the University of Colorado. The university has a biobank and is sequencing data for UCHealth patients through a special consent process. He estimates that there are approximately 23,000 DNA samples collected at this point, and pharmacogenomic work has started to match it up to the drugs. 

"We haven't quite yet closed the loop between the DNA data that's being sequenced and collected back to the EHR, but we're close to that. Over the next 12–24 months we fully expect to close the loop between the DNA sequencing and the EHR."

Data-related challenges

Respondents say that their organizations expect to face a number of data-related challenges in performing analytics over the next three years. The top three challenges are integrating internal clinical and financial data (47%), establishing/improving EHR interoperability (43%), and integrating external clinical and financial data (43%). In each case, the data comes from multiple sources and is not easily collated, which is likely why it is a challenge for providers.

Although the response for ethical concerns regarding the use of patient genetic data (8%) receives a low response, it is also true that the provider usage rate of patient genetic data for analytics is small.

For example, only 9% of respondents say that they draw on patient genetic data when conducting patient-related data analytics. As such use becomes more widespread, concern about ethical issues may also grow. 

Hess says that the use of patient genetic data has many ethical implications for patient care, and that it is an issue currently being studied at UCHealth.

"Using patient genetic data adds a bunch of ethical implications. As an example, what if we sequence your DNA data and you are a patient of ours but we haven't seen you in two months, and we notice that based upon your DNA sequencing, the medication that we prescribed for you two months ago probably could have been a better choice? Or there's another nonpharmaceutical therapy that might be better for you? What are our obligations for reaching back out to you? 

"It might be that we find something based upon DNA sequencing that wasn't available to us a year ago when we treated you, which is now available to us. And what if we find something in your DNA that is game-changing—what do we do about that? And do people really want to know some of those things? Do they really want to know what their DNA is showing in terms of being predisposed to cancer, diabetes, or some other disease?"

Tactical challenges

According to the survey, the top three tactical challenges respondent organizations expect to face in performing analytics are the need to deliver timely analysis (46%), overcoming insufficient skills in analytics (42%), and insufficient funding in light of other priorities (37%).

Note that two of the top three tactical challenges are either indirectly or directly related to financial resources—the solution to overcoming insufficient skills in analytics is investment in training or adding new analytics staff, and insufficient funding in light of other priorities is
clear-cut.

Delivering timely analysis is perhaps a universal problem—life as we know it requires a real-time response to information needs, and healthcare is no different.

It is worth mentioning that the response for insufficient staff (35%) is fourth on the list of analytics challenges, which is also an indicator of insufficient financial resources.

Lacking financial resources is something that plagues most healthcare IT departments when it comes to analytics, particularly the advanced forms of prescriptive analytics—the needs almost always exceed the budget—because there is simply too much data and the complexity of analyzing it effectively can be overwhelming.

According to Hess, outsourcing the more advanced parts of the analytics workload while retaining core EHR data responsibilities and mainstream analytics functions can be an effective strategy.

"About 18 months ago I made a decision that we can never find nor retain the data science experts that we're going to need in the future. So even though we're partners with the University of Colorado and we have really intelligent individuals in the Denver area, it is extremely difficult to attract and retain that level of expertise. That's why our partnerships with companies that are in Silicon Valley and competing with Google and Microsoft and all these different global organizations, that's a better approach for us. 

"We have a talented, talented team that knows how to deconstruct the Epic data model really well, but we could never get to that next level that we're going to need to go to. I think you can count on one or two hands the number of healthcare organizations in this country that can actually do what we're talking about in terms of advanced analytics."

As an example, Hess mentions two outsourced initiatives that involve scheduling efficiency, one for the University of Colorado Cancer Center's chemotherapy infusion centers (starting May 2015) and the other for UCHealth's operating rooms (starting May 2016).

"We don't have a shortage of data. We don't have a shortage of dashboards. We have plenty of tools. What we don't have is plenty of time to analyze all of that data."

The initiatives were outsourced to California-based LeanTaaS, which uses advanced predictive analytics and machine learning algorithms to provide efficient scheduling and usage of the resources.

The results for the infusion centers were positive. According to Hess, once the initiative was formally launched, patient wait times were reduced by 60% during peak periods and averaged 33% lower throughout the day. There was also a 28% reduction in infusion center staff overtime.

Hess describes the process this way: "[We] feed our data to LeanTaaS in the cloud, deidentified. Then they would come back with solutions on how to change our scheduling templates [in Epic] to get more patients through. So existing scheduling data is sent to the cloud, run through machine-learning algorithms, and then it comes back out with, hey, change these four-hour infusions to six-hour infusions, these eight-hour infusions to four-hour infusions, and so on, like a big game of Tetris to get more patients through.

"And we're now on iteration four of their algorithms, and it gets better every time. It's not a one time in, change it, and then set it and forget it. It's constantly learning and changing." 

According to Hess, the operating room scheduling initiative, which is still being rolled out across UCHealth System's ORs, has also been successful.

"And we're now on iteration four of their algorithms, and it gets better every time. It's not a one time in, change it, and then set it and forget it. It's constantly learning and changing."

So far in 2017, the program has been implemented at 86 ORs across the organization. At the time the program was launched in May 2016, UCHealth's OR block utilization—the share of the allocated OR time that is actually utilized—was about 65% on any given day.

As an example of the progress that has been made, current block utilization within the 25 ORs at University of Colorado Hospital's Anschutz Inpatient Pavilion (AIP) is at 76.5%, up 4.7 percentage points over the most recent six-month period. The current 2017 utilization target is 80%.

Note that Hess says that each percentage point of increased utilization at an OR adds approximately $100,000 to the unit's bottom line.

Analytics development

Survey respondents say that the three most promising areas for analytics development are clinical best practices (56%), real-time delivery of actionable information (54%), and population health data (47%).

It is worth mentioning that the need to deliver timely analysis (46%) is the top tactical challenge that respondents say they expect to face in performing analytics over the next three years. The good news for providers is respondents expect progress in an area that is important to them.

Achieving real-time delivery of actionable information, given the quantity and complexity of healthcare data, is likely only going to be accomplished through the use of prescriptive analytics. Hess describes the state of the industry this way:

"At some point, the IT departments and COOs and CEOs across the country, they're going to realize that they're spending hundreds and thousands or even millions of dollars on getting data out. And it's not helping them figure out what levers to pull any better. They have to go after prescriptive analytics capabilities that actually come back with, ‘Based upon what we're seeing, here are the three things that you need to be doing to be more effective.' " 

Jonathan Bees is a research analyst for HealthLeaders.


Get the latest on healthcare leadership in your inbox.