Data and analytics increasingly crucial to risk-based payment success

Baystate Health, an integrated health system in Springfield, Mass., has approximately $60 million per year in net patient revenue at stake as a Medicare Next Generation ACO. “That’s a big swing, plus or minus $60 million. If you don’t do well, the loss can be a real problem,” says CIO Joel Vengco, senior vice president and CIO.

The “Next Gen” program is for progressive providers willing to assume sizable two-sided risk to care for a Medicare population, including outpatient, acute and post-acute services. On the upside, ACOs get to keep 80 to 100 percent of any savings achieved over a set annual spending target, plus a 5 percent incentive. On the downside, ACOs must cover 80 to 100 percent of any excess spending.

Baystate is among a minority of organizations taking on this level of risk. According to a 2017 survey, 70 percent of provider organizations involved in value-based payment had less than 20 percent of revenues at risk. In addition, 38 percent were in upside-only risk arrangements.

Baystate Health-CROP.jpg

One reason for the slow uptake is the financial risk involved. Baystate Health’s ACO, Pioneer Valley Accountable Care, has done well in Next Gen, earning more than $4.6 million in shared savings in 2016, while maintaining quality. But seven of the original 18 Next Gen ACOs suffered losses, ranging from $63,570 to $6.15 million. “It’s a tough journey,” Vengco says. “To be able to manage a patient population, you have to really know your patients.”

This involves leveraging sophisticated data analytic tools traditionally used by payers, says Jason Joseph, CIO at Spectrum Health, a 15-hospital system based in Grand Rapids, Mich. “You need a strong analytical capability that helps you understand prospectively, ‘What is the risk of this population?’ Then, the population needs to be effectively managed to ensure appropriate utilization and high-quality care. Without these fundamental capabilities, you are flying blind.”

Providence St. Joseph Health, based in Renton, Wash., has been busy building a comprehensive analytics platform. The health system has seen a 150 percent growth in lives covered under risk-based arrangements over the past three years. When asked to pinpoint the most important data analytic capabilities for managing risk, Deepak Sadagopan, group vice president of population health informatics and government programs, outlines four types—business decision support, care management, performance improvement and operational intelligence.

Business decision support

“Healthcare executives are having to make decisions around the type of payment models their organizations will participate in,” Sadagopan says. “But there’s a lack of adequate infrastructure to support them in making those decisions. All the analysis we’ve done over the past few years has been on spreadsheets.”

To address this shortcoming, Providence St. Joseph Health is building a business decision support structure that enables executives to proactively weigh the potential ROI from participating in risk-based payment models. For instance, many of Providence St. Joseph’s 50 hospitals and affiliated medical groups wanted to participate in Medicare’s new voluntary Bundled Payment for Care Improvement (BPCI) Advanced model. But a key decision needed to be made: Which of the 32 clinical episodes in BPCI Advanced—which range from cardiac valve procedures to kidney failure—were the organizations best prepared to take on?

To provide insight, Sadagopan and his team built an analytical framework using three years of historical claims data to find the average cost of care for BPCI Advanced episodes across Providence St. Joseph’s provider groups. The analysis also compared providers’ historical costs against the benchmark target set by CMS. “We were able to calculate the probability of each provider group being able to meet specific target prices,” Sadagopan says.

The model is currently being expanded to handle what-if scenarios. For instance, how would a provider group perform in BPCI Advanced over time if it could lower post-acute costs for a particular episode by 10 percent? Building this model involved integrating enormous amounts of claims data and tagging critical pieces of information, such as encounter costs and readmissions, before storing it in a data mart for analysis.

Care management

Stratifying a patient population by health risks and other traits is a key part of successful care management. This often begins by identifying patients with high health risks, such as those with multiple chronic diseases, as well as those on the verge of having a serious health event.
“High-risk patients, who make up 4 to 6 percent of any given population, can be as much as 75 percent of healthcare delivery costs,” Vengco says. “The opportunity is in identifying those who aren’t yet in that high-risk pool, sometimes referred to as rising risk, so you can help them before they get chronically sick.”

While a rich source of information, the EHR may not include data on services that patients seek outside a health system or network. “An important part of managing patients is getting real-time access to encounter notifications,” Vengco says. “If you can get alerts when patients are admitted to an inpatient facility or go to the community health center down the street, you have a better chance of engaging them immediately and perhaps triaging them to the right level of care.”

Baystate Health is one of the founders of the Pioneer Valley Information Exchange (PVIX), which includes most healthcare organizations in western Massachusetts. “We use the platform to access the holistic clinical record of the patient, enabling us to view all the care the patient has received,” Vengco says. “We can see all their meds, lab results, allergies and prior clinical notes, regardless of whether it’s a Baystate Health clinic or not.”

But clinical data is only one type of data needed to understand patient risk. Research generally shows that social determinants of health, such as food insecurity and lack of transportation, account for about 50 percent of health outcomes. Recognizing this, Baystate has started asking Medicare ACO patients about any social, economic or environmental challenges they face.
“If one of our Medicaid patients is unable to see a doctor because of transportation issues, we need to know that so we can try to find a way to get them to the clinic, or perhaps they are a candidate for our telehealth services,” Vengco says.

Community health workers are interviewing patients face-to-face, using a care needs assessment app. “That data is then stored in PVIX and can be reviewed by clinicians in the ACO and the region when those patients end up in their care,” Vengco says.

To further engage patients, Baystate is partnering with PatientBond to incorporate psychographic segmentation into its patient registry. PatientBond uses a variety of consumer data, as well as a 12-question survey, to pinpoint patients’ communication preferences (for example, whether they prefer contact by text, email or phone) as well as what tends to motivate them in managing and improving their health.

“Population health is really about managing behavior change,” Vengco says. “It’s not unlike what retailers like Amazon and Target have been doing. They understand what motivates people, and they use that knowledge to engage them and modify their behaviors. We need to take lessons from other industries on how to leverage various data types beyond healthcare to truly know our patients and ultimately better manage patient behaviors and, mostly importantly, their health.”

Performance improvement

Spectrum Health has its own health plan, Priority Health, which has proven beneficial in moving toward risk-based payment. “As a health plan, we’re in the risk business already,” Joseph says.

The health system has built a sophisticated analytics framework that incorporates statistical packages, including SAS and Python, and the visualization tool from Tableau. It has also developed risk-scoring and care-management processes that leverage predictive analytics.

In addition to aiding population management, these analytical capabilities help data scientists uncover drivers of variation in patient care. “We use the data to go after improvements that will lower the cost or improve the quality of care,” Joseph says.

One lesson learned is the importance of focusing analytical efforts on specific high-priority issues. “Data analytics is not ‘a build it and they will come’ strategy,” Joseph says. “In other words, don’t build a data warehouse with all the data anyone could ever want hoping that somebody will come and use it one day.”

Spectrum Health is using an iterative approach. “We start with, ‘Here’s the problem we’re trying to solve.’ Then we build analytics around that problem and pull together data to solve the problem,” he says.

For example, after Spectrum Health identified hospital throughput as an improvement priority, staff gathered and analyzed data to identify the bottlenecks and opportunities to improve performance across the system. That led to a redesign of how patients were triaged, moved and put in exam rooms. A lot of useful data for the analysis was already being collected via the process log in the health system’s EHR, including the time when patients are checked in, see a physician, are moved through the organization and are discharged.

“We follow the principle, ‘Let’s start with analytics that we can develop based on data we already have before we talk about analytics that will require additional data,’ ” Joseph says. “Sometimes data fields are there already, but we haven’t built the analytical tools yet to leverage that data. As we move up the analytics maturity scale into predictive analytics and leveraging machine learning, we will need to ensure our data is well organized and well managed.”

Operational intelligence

Sadagopan believes the greatest analytics opportunity lies in integrating intelligence into the clinician and care manager workflow. This includes providing physicians with visibility into any care gaps patients have, including those related to quality metrics, such as patients needing immunizations, as well as lists of high-risk patients who would benefit from care management.

“If we make commitments in specific payer contracts, but do not conduct those activities, we will not comply with the terms of the contract and may leave money on the table,” Sadagopan says.

One challenge in implementing payer contracts is the disparity in how administrative data, including member rosters and claims data, is formatted and shared. “We see formats and structures all over the place,” Sadagopan says. “Some payers send us data as flat delimited files. Others send us additional bits of information in an Excel file. In some cases, we get rich clinical data with risk codes. In other cases, we don’t get any rich data like that.”

This Wild West situation can make even simple tasks difficult. “We even struggle to identify patients attributed to Providence St. Joseph Health,” Sadagopan says.

Resolving this situation will require collaboration across providers and payers to standardize administrative data exchange, similar to how clinical information exchange has become more uniform. Providence St. Joseph has started to partner with payers to achieve this.

Another challenge is the sheer scale of data being collected. As an example, Sadagopan points to the large number of quality metrics Providence St. Joseph has to report to Medicare, Medicaid, and commercial payers. “A foundational challenge is how do we scale our information infrastructure to track, compute, and report on all these quality measures across our whole system?”

Providence St. Joseph is investing in a cloud-based big data infrastructure that will be able to deliver needed information to leaders and staff on a real-time basis. Currently, data is gathered from multiple settings and stored in an enterprise data warehouse. “As we move into a 2.0 world, we want to be able to organize those massive information sets into data marts by specific business-focused need, which will dramatically improve the efficiency of processing information,” Sadagopan says.

The health system sees its data management infrastructure as a strategic asset it needs to shore up to ensure success under value-based population health. “We’re looking at every aspect of how we do business—in terms of acquiring patients, managing care, and managing risk—and we need information on our performance in all those areas. That makes scaling our data infrastructure one of our core priorities for the next three years.”

For reprint and licensing requests for this article, click here.