PROVIDENCE – In 2015, the Centers for Medicare and Medicaid Services awarded Rhode Island a $20 million State Innovation Model grant, with the expectation that the funds would be used to transform the way health care is delivered and paid for.

The major architects and champions of the SIM grant, Elizabeth Roberts, the former secretary at the R.I. Executive Office of Health and Human Services, and her former deputy director, Jennifer Wood, are no longer serving in government.

In turn, much of the current attention by state officials in health and human services has been directed at rescuing the more than $400 million Unified Health Infrastructure Project, or UHIP, following the botched roll out of the Deloitte software system in September of 2016, in the attempt to create a one-size-fits-all technology portal system for benefits.

At the same time, with the election of President Donald Trump, many of the priorities and investments made under the Obama administration, particularly around the work of the Centers for Medicare and Medicaid Innovation, appear to be changing, under the leadership of Dr. Tom Price, the new director of the U.S. Department of Health and Human Services. Exactly what many of those changes are involved with has been difficult to discern, given the current lack of Congressional oversight.

Still, the work around the SIM initiative in Rhode Island is ongoing, focused in large part on developing data metrics around measuring population health outcomes.

High-risk patient identification
Recently, on July 11, the SIM High Risk Patient Identification Workgroup met for the first time, taking over responsibility for what had been a similar group under the direction of the R.I. Office of the Health Insurance Commissioner.

The members of the working group have a high level of expertise if not “wonkism” when it comes to health care policy initiatives. Attendees at the July 11 meeting included: Libby Bunzli policy analyst; Dr. Peter Hollman, the new academic director of the Executive Master of Healthcare Leadership program at Brown University; Michele Wolfsberg; Patty Kelly-Flis; Andrea Galgay, director of ACO Development at the R.I. Primary Care Physicians Corp.; Catherine Hunger; Dr. Ed McGookin, chief medical office at Coastal Medical; Dr. Beth Lange, pediatrician; Charlotte Crist, managing director, clinical programs, at Blue Cross & Blue Shield of Rhode Island; Jason Lyon, behavioral health administrator at the R.I. Executive Office of Health and Human Services; Melissa Lauer, health information technology specialist at R.I. Executive Office of Health and Human Services; Dr. Pano Yercararis, co-director of the Care Transformation Collaborative-RI; Suzanne Campbell, Care Transformation Collaborative-RI; Sandy Curtis; Betsy Dennigan, URI College of Nursing; Elaine Fontaine, director of data quality and analytics at R.I. Quality Institute; Dr. Ailis Clyne, medical director, R.I. Department of Health; Scott Young, director, business strategy and consumer services at R.I. Quality Institute; Liz Fortin, program director of community health teams for CTC-RI; JoEllen Golberg, senior clinical practice advisor, R.I. Quality Institute; Deb Hurwitz, co-director of the Care Transformation Collaborative-RI; Brenda Jenkins, senior program administrator at Healthcentric Advisors; Marti Rosenberg, project director of Rhode Island’s State Innovation Model grant; Mark Grey, OHIC; and Garry Bliss, Accountable Entity program director at Integra.

What follows is a summary drawn from the meeting’s minutes, which offers insights into the discussion of how to identify and track high-risk patients as an integral part of the overall efforts to transform the health care delivery system in Rhode Island.

It represents a deep dive into health care policy. But, by making the conversation more transparent, ConvergenceRI seeks to allow broader participation and community engagement in the discussion.

What was said
Libby Bunzli, who facilitated the meeting, began with a discussion of concerns about the previous work done under the auspices of OHIC.

“One of the major concerns from the previous work is that there was a lot of mismatch from the lists that providers received from various payers to identify high risk patients,” Bunzli said, according to the minutes of the meeting. “The OHIC High Risk Patient Identification Workgroup brought together a variety of stakeholders to look at reducing the complexity of managing these lists. Our consultants at Bailit Health did interviews and research and provided a good understanding of the environment.”

At the same time, Bunzli continued, “The SIM Integration and Alignment initiatives identified a need to align high-risk assessment. SIM recognized the opportunity to align rather than duplicate to be more effective. This is focused on identifying high-risk patients among a patient panel, but there was also a focus on social and environmental determinants of health. When we measure high risk we want to include the social determinants in that assessment. We also recognized that there is a difference in measuring risk for pediatrics and adult medicine, and that the outcomes may take longer to see with children.”

In terms of process, Bunzli explained, “SIM will now be taking over from this group, with [Bunzli serving] as the consistent lead and we will pick up from where we left off.”

Why the working group was moved from OHIC to SIM has never fully explained.

Detailing what the consultant Bailit found
Bunzli then went over the Bailit process and survey results. Most were using claims data, and a variety of software was used.

Bailit had laid out a basic framework for aligning high-risk assessment, according to Bunzli:

1. Establish goals, specify population of interest and engage providers. It was made clear during interviews that it is essential to bring together a variety of stakeholders and include providers as the ones who understand which elements go into the risk question.
2. Define high risk – there are many ways this could be defined, so a unified definition is required.
3. Assess data sources and purchase or develop risk prediction tool. There could be a variety of data sources already present in the state that could work well for this. It was identified as very important to explain why the patient is high risk and have some way to zero into what is driving the risk score.
4. Apply prediction tool to population and stratify. Depending on the tool, may need another step to assess “impactability.” There may be different levels of risk, but also a different level of whether an impact can be made.
5. Determine specificity and sensitivity of prediction tool and make necessary adjustments. The interviews made clear that it was important for the provider to make adjustments on the list and determine whether they are really high risk or identify if there is the ability to have an impact, because they should not be held accountable for everything if there is not ability to impact.
6. Work flow – share data and determine appropriate care management interventions.

Lessons learned
Bunzli then provided an overview of the lessons learned as part of the analysis conducted by Bailit:

• Limitations to analytic models
• Important for the tool to have claims plus clinical data
• Timely, high quality data is more effective
• No gold standard to incorporate social determinants of health

In terms of “impactability,” Bunzli said the lessons learned included:
• Predictive analytics with hands-on review for impactability
• Start small with predicting something more simple and then move beyond
• Include clinicians in tool construction
• More specific on results the better

While the use of the term “impactability” has apparently found a home within health care analytics, the word itself is somewhat impenetrable in terms of what it means and how it is being used. The folks on the SIM working group would do well to translate the term into something more coherent, more transparent and less wonkish.

In terms of social determinants of health, Bunzli said that the lessons learned included:

• Try to include publicly available data
• Use validated patient-reported instruments
• Inconsistencies among different notes
• [Develop a strategy on how best] to understand family indicators.

Strategy options
Bunzli then outlined a series of potential goals and strategy options for the group around what might be most useful for identifying high-risk patients. They included:

• Develop a consensus definition of high-risk in Rhode Island
• Create a variable data bank of important predictors of risk
• Define a standardized method for capturing variables in the social determinants of health
• Leverage state and federal databases
• Develop an APCD/HIE platform for identifying high-risk patients.

A more fundamental question, whether or not identifying high-risk patients will lead to better public health outcomes, does not appear to have been asked and answered.

Working group discussion of proposed strategies
The minutes of the meeting revealed a robust discussion around the strategies and their implementation.

CRIST: Good strategies, but waiting for the shoe to drop on how we will do that. Can see the consensus definition being a challenge across the state, especially with payers each coming up with their own assessment and use of complex tools with different risk models. Foresee it as a challenge to ask all the payers to identify high risk the same way.

MCGOOKIN: In the context of the social determinants of health, which is seen as very important in high-risk definition – is it truly necessary to develop a consensus definition? It will depend on what the organization’s capacity is to respond to the high-risk factors. Once you screen you are ethically obligated to respond to the screens and what you do will depend on your capacity. With the payers there is a risk that payers may want to enforce those rules universally.

GALGAY: The term consensus may not be the right way to describe this. We agreed that there may be consensus on the categories of information that go into the high-risk assessment, but unrealistic to ask that [UnitedHealthcare] change their national calculation.

Maybe more flexibility built into the definition will help it work across more scenarios, and allow a variety of diverse practices to determine how to apply the methodology.

LANGE: CTC has a consensus definition for high risk to be used for reporting. This is for adults only. PCMH Kids is still working on one. It is based upon a practice’s information in their EHR, which may not include hospital utilization. There could probably be some better practices and a consensus definition may help people not waste their efforts and be more effective in identifying high-risk patients.

HOLLMAN: Patient characteristics even though one may be in a limited population, probably run across all variety of patients. We do not have agreement on the non-claims based information that could be used, and this could be an opportunity to test what we have available.

Working together to come to an idea of best practices may be very helpful, especially in scenarios with complex cases that may seem impossible to improve. Would a list of best practices for addressing some of the high-risk factors assist in some of those scenarios? How is the list then used? Is there a requirement attached there in how it is used? We do not see imposing any specific requirements, rather having it available as a guidance tool.

Guidance is tricky because there may be additional factors that may impact a specific patient. Thundermist’s list of indicators were very impressive, and we may find out what really matters. We ask them to do a lot of questionnaires, which frustrate the patients, but maybe there is a way we can arrive at a shorter assessment, 4-5 questions, so providers can identify a real underlying issue with less burden to the patients.

BUNZLI: Another advantage would be looking at high risk on a population level. For example if we have a standardized social determinants of health screening, we may identify major issues, such as a housing challenge, in a more systematic way, we could have justification for changes on a systemic level.

[In] identifying pre-Medicaid risk for those populations. [Bunzli said] she would reach out to Medicaid to find the language being used for high-risk in geriatric population, and reach out to get the CTC definition.

CAMPBELL: Also be interested in looking at practices that use the “How’s Your Health” assessment for high risk. This can help determine if patients feel like they can handle management of their condition. How’s Your Health seems to fit in that additional question category that covers information beyond the standard data and gets to social determinants of health and other indicators.

There is also an opportunity with clinical/claims data to help identify rising risk early enough to address the issue before it becomes high risk. The groups are different whether you are talking Medicaid, geriatric, kids, and different cases where money is spent. There is a functional capacity assessment questionnaire in a national quality measure.

HOLLMAN: [I am] concerned about us as a group coming up to a consensus without having any real background in the evidence base, and it might be beneficial to hire a medical student to do 40 hours of research to help collect that information. CTC had shared a lot of information from research in this area, so perhaps they have collected some materials already that could be useful.

The social determinants of health
Bunzli then directed the working group’s conversation toward a discussion of the questionnaire prepared by the CMS Accountable Health Communities.

BLISS: These are validated-ish questions. CMS has developed a 10-question tool that deals with the five core areas that every recipient of the grant must ask about: housing, quality of housing, transportation, food insecurity, utilities, and interpersonal violence.

These questions are being tested out in some areas of the community. They believe that the questions are too complicated and the reading level is too high. It is expected the wording will change some.

There is a supplemental list of questions which may be asked as well, however you cannot get to the additional questions in the CMS platform if you did not test positive for one of the core needs. Integra has informed that they need to know soon what platform should be used. No one knew a platform would be created by CMS, but now people are not intending to use the CMS data platform for the long-term future. They expect to switch to the Health Leads platform. They are working with United Way 211 to build the database.

CMS is still saying that in 5 years, their data platform will be shut down. CMS will be impenetrable for anyone that is not a grantee. Does not make sense with the way we are building a statewide database. Have not figured out yet how to allow access from others – can work it in some way where people do not have to test positive to get to the other questions.

[There are also] some reservations about it being a federal database. Starting with the CMS platform and then moving to Health Leads will allow for them to catch up to what CMS determines.

FONTAINE: Would Health Leads or Integra be open to the databases feeding the [statewide health information exchange] for care coordination? When you build a parallel play database not integrated into the EMR, that’s a brick wall to providers. For example, Kidsnet has been a challenge over the years. Worthwhile having a community discussion with the people have dabbled in pieces of this to help understand what may be sustainable over time.

Integra proposal and budget includes now that the survey would interact with Epic and Cerner. Health Leads has experience with doing bi-directional feeds with other EHRs.

The Health Leads tool in the slides are the questions developed for Thundermist with Health Leads. They put the associated ICD codes because it was important to monitor these things on a population level.

With the Z codes [which indicate the reason for the encounter], there may be the ability to come up with a definition and consensus for ourselves with the Z codes.

GROUP: [unidentified comments in the minutes] The group urged caution about relying on claims data. The time lag is an issue. Also need to think about the frequency of these screening tools. How often do you use them? There are a lot of implementation questions still to ask. We should look at the way some of these assessments have been used in Massachusetts.

Is there an idea to try to build a centralized platform to support the assessment, and is there a greater reason for us to reach a consensus that could be productive at the end?

LAUER: If there is some sort of consensus and it makes sense to build it centrally rather that at each individual entity, it is something the state would pursue?

YERCARARIS: Makes more sense to try to use some of the existing work, such as RIQI’s bi-directional interfaces, rather than build externally where it maybe more expensive/resource intensive.

BUNZLI: Where possible it makes the most sense to leverage any existing resources possible. Also makes sense to tier the questions and have skip patterns – if someone can drive to work you do not need to ask the detailed questions because you can assume they can shower, prepare meals, etc.

We might actually be able to find a very refined method to put something in place. If there are positives people might want to drop down into more detailed questions. We can identify the categories from the experts in this, such as academic researchers.

CLYNE: All of the questions speak to current conditions, so there is a fundamental question of whether we also look at past circumstances, which could indicate risk or have led to current conditions.

Who is asking these questions? How are we engaging with people? Probably the people you are not seeing – is it homeless centers, health homes, schools, etc. Also good not to hand them a piece of paper. This goes back to our need for full integration, physical health/behavioral health/social services. There is also this concern not to ask the question if there is no way to address the issue.

In Oakland, Calif., there is a piloted method of providing a list, and then having the person tally the number of issues they are having so the provider has less to react to specifically but can make an appropriate referral. However, need a warm hand-off to behavioral health. Not everyone is skilled at asking the questions in a way that will have the patient comfortable to open up.

The how will probably impact the what. It’s about training, interoperability, practice impact. How will we make this part of our health system as a whole, both with providers and with patients.

The assessment needs to be accessible across the system so that individuals do not need an in-depth rescreening.