Recommendations

  1. Identify specific individuals within your chosen population using multiple approaches.
  2. Develop real-time patient identification methods.

1. Identify specific individuals within your chosen population using multiple approaches.

Below is guidance on how to begin using data to identify individuals and questions to discuss with primary care physicians who see patients in your population segment.

Generate a list of patients who fit the description of your chosen population segment, using at least two of the following methods.
  • If your chosen population is defined by high utilization, review your available health care utilization data. If possible, generate a list of the top-utilizing patients in the past year. (Examples of a utilization threshold include four emergency room visits and two hospitalizations in the past six months; three or more inpatient visits in the past year). Then, review the previous year’s data, to see if last year’s top-utilizing patients also had high utilization two years ago; if possible, repeat to look for the same pattern three years ago. Reviewing 2-3 years’ worth of data minimizes the risk of identifying patients whose utilization would decrease with no additional services, which is the challenge of regression to the mean.
  • If your population  with complex needs and high costs is defined by a clinical threshold – e.g., chronic conditions, the experience of homelessness, or another characteristic captured in patient files – generate a list of patients who meet the threshold. Flags related to specific health or social conditions can be very useful here. (Examples of clinical thresholds are patients discharged from the hospital with five or more medications, patients with four or more chronic conditions, or patients with no fixed address). Note that the goal is to screen in those patients with complex needs and high costs and are a good fit for your enhanced care program. Challenges and obstacles to good health outcomes are characteristics of those who would fit the program, and should not be grounds for exclusion; these methods should not be used to screen in patients with low levels of need.
  • If you survey patients about their self-reported health status, identify those patients who rate their health as fair or poor. A relevant threshold is patients who rate their health as fair or poor.
  • If your organization has access to predictive models that aim to identify patients at risk of high utilization in the future, run the model every 3-6 months to continually identify potential patients.
  • Stratify patients according to relative need or risk of high utilization in the future. Patient stratification is the determination of levels of need within a population segment. For example, teams may establish definitions related to high, medium, and low needs. Stratification can help your team to prioritize patient lists according to need.
Meet with primary care clinicians to review the list of patients who are in the population segment with complex needs and high costs.
  • Ask them to share what they know about the patients on the list, and specifically ask them to identify the patients they are most concerned about.
  • Questions for primary care providers include:
    • Who is on a steady health decline trajectory?
    • Who, without more intensive assistance now, is going end up in the emergency room or the hospital?
    • Who keeps you up at night?
    • For whom do you need some extra intelligence (eyes and ears) in the home?
  • Try to identify themes or characteristics that would place an individual patient in the chosen population segment.
Marry what you learned in Step 1 with what you learned from primary care physicians.
  • Have the discussions with the primary care physicians refined your understanding of the criteria for the population segment?
  • Have the discussions led you to recognize specific flags that will help you to identify individual patients who are good candidates for your program?
  • Create an easy to use form for clinicians and staff to use to refer to and verify the individual meets the criteria for the enhanced care program. Review Cambridge Health Alliance’s Complex Care Management Triage Tool.

Here are a variety of ways to identify potential patients. Remember, it is best to use a combination of quantitative and qualitative approaches – ideally, three of the following six methods.

Quantitative approaches
Use clinical information systems to identify individuals (e.g., diagnosis flags).
Pros
  • Clinical data are generally easily available from an electronic health record system or other clinical data systems (such as pharmacy data, chronic condition registry).
  • Clinical data can generate lists of patients with multiple chronic conditions.
  • If social or behavioral health needs are captured in the data system, clinical data may be able to list patients with certain flagged social or behavioral health needs.
Cons
  • This method will not predict which individuals will be high cost in the future.
  • Additionally, for many people, it is the social determinants of health, such as their current living situation, that are the key drivers of high utilization.
  • Social determinants of health are generally not included in claims or clinical data.

Considerations

The utility of this method largely depends on your access to clinical data. The following challenges are obstacles to identifying patients through claims data:

  • Your team does not have frequent and consistent access to claims data;
  • Your team does not have access to information technology and data analytics support;
  • Claims data are of poor quality or incomplete.
Create a utilization threshold to identify patients with a certain number of visits to the emergency department or inpatient stays in the past year.

Examples of utilization threshold include four emergency room visits and two hospitalizations in the past six months; three of more inpatient visits in the past year.

Pros
  • Claims data that tally hospital-based service utilization are easily obtainable for large health care systems.
  • When the data are obtained frequently (without lagging by three or more months), it allows teams to attempt to find and meet potential patients soon after they encounter health challenges.
Cons
  • Utilization data review detects patients with high health care costs in the past, but it cannot predict which patients will continue to have high utilization in the future.
  • Utilization data are vulnerable to regression to the mean, meaning that the cost curve will decrease over time, even if patients receive “usual care” (no special intervention or enhanced services). This means it is imperative to identify individuals who are likely to have persistently high costs, who can be impacted by care redesign.
  • There is generally a time lag between getting the data and the events for the individuals.
  • For many individuals, the social determinants of health and their current living situation are key drivers, but this information is typically excluded in claims data.

Use a predictive model.

A predictive model is a statistical method wherein patient characteristics, past utilization, past events, or other data are compared against a theoretical combination that aligns with 100% chance of high utilization in the future. The model will generate a relative risk score for each patient, indicating the risk of high future utilization.

Pros
  • Well-validated predictive models can be useful in identifying subsets of high-risk patients who may be at risk of high utilization in the future.
  • Predictive models provide a relatively complete view of health care costs.
  • Many organizations use predictive models.
Cons
  • Predictive models do not register social determinants of health, limiting their ability to predict many individuals who need the support of enhanced programming.
  • Predictive models rely on the completeness of the data they analyze. Claims data has gaps, as individual patients are enrolled and dis-enrolled from health care insurance plans, and these data gaps reduce the accuracy of predictive modeling.
Qualitative approaches
Ask primary care providers to refer their patients to your enhanced program.
Pros
Cons
  • Primary care providers may have a difficult time identifying patients at the highest level of need or cost, and may be tempted to refer large groups of undifferentiated patients with complex needs.
  • Primary care providers will need support in making appropriate referrals, including:
    • Clear definitions of the patient characteristics that the enhanced program is set up to serve; and
    • Shared understanding of program goals.
  • It is unlikely that referral alone will result in identifying patients at risk of high future utilization.
Ask patients to rate their health status, identifying those who rate their health as fair or poor.
Pros
  • This method has been validated by the Centers for Disease Control and Prevention (CDC) as a good predictor of both mortality and high health care costs.
  • Self-rated health status surveys are relatively easy to collect initially and over time.
  • Some organizations already collect this information.
Cons
  • Many health care organizations do not have systems in place to elicit and collect this data.
  • Many organizations have not trained their workforce in how to administer or interpret the single-item health status survey.
Use a qualitative threshold based on an understanding of the level of self-efficacy, life trauma, or life conditions (e.g. poverty, social support, cognitive function).
Pros
  • Thresholds based on an understanding of the level of self-efficacy, life trauma, or life conditions (e.g. poverty, social support, cognitive function) surface many of the root causes of patients’ high utilization and are a good complement to electronic health records and claims data, which rarely include social determinants of health.
  • An individual’s feeling of self-efficacy significantly contributes to their ability to manage their health and life circumstances, is positively correlated with health outcomes and negatively correlated with utilization.
Cons
  • Obtaining this information is time and resource consuming.
  • Many organizations do not recognize and have not prioritized the skill it takes to elicit information about life circumstances and social determinants of health from patients.
  • Current care teams may not have the capability and skill to learn about social determinants of health from patients.
Tips and guidance for identifying patients who are a good fit for your program:

 

2. Develop real-time patient identification methods.

Establish a process for event notification from the emergency department, hospital, or other community organizations.
  • Create a flag in the electronic health record to signal that a patient is present.
  • Using PDSA cycles, test staff going to the hospital or being at the hospital to meet with individual patients during their visit to the emergency room or stay in inpatient care.
  • Stay attuned to emerging technology; for example, PatientPing. 
  • Find partners to explore technology solutions, such as the Oregon Health Leadership Council’s Emergency Department Information Exchange. 
  • The Winnipeg Regional Health Authority in Winnipeg, Canada leveraged their centralized data infrastructure to develop a real-time identification method: metge-photo
     “Our team partnered with the data providers to build the system to pull utilization data. Surprisingly, it was not too hard to do. We sent an email and invited key data stakeholders to a meeting. We developed the utilization database, pulling data essentially from the emergency department and from hospital admissions. We pulled that data into a system, which generates messages through faxes, believe it or not, back to the emergency department, should one of these individuals be present.” Colleen Metge, MD, Director, CHI Evaluation Platform Division Quality & System Performance, Winnipeg Regional Health Authority
Using PDSA cycles, test using a triage coordinator to find appropriate individuals by reviewing up-to-date patient information, including:
  • Hospitalizations and emergency department visits.
  • Stays in skilled nursing facilities.
  • Participation in services provided by community partners, such as homeless shelters.
  • Referrals from primary care providers (PCPs).
Support primary care and other providers to make appropriate referrals in real time on a week-to-week basis.
  • Continue to develop a shared understanding of the best patients for the program and encourage PCPs to refer as they become aware of them.
  • Create an easy-to-use referral mechanism, perhaps a template facsimile form.
  • Communicate directly with primary care providers within one week of the referral to update them on your team’s progress connecting with the patient; this will provide value to the PCP and engage them in a more real-time process.
Tips and guidance for real-time patient identification methods:
  • Develop methods to identify potential patients at opportune times, such as during a hospitalization or ED visit, when their motivation to participate in enhanced programming maybe higher than usual.
  • Co-locating care management staff of community health clinics, behavioral health clinics, and social service organizations in the hospital can be an effective way to identify potential patients, and is a practical and constructive way of collaborating across different organizations.
  • Create coordinated, responsive, and easy communication methods between those who refer patients and care management staff.
  • Partner with another organization that serves a high volume of your chosen population segment. Partnership methods around patient identification can be as simple as patient referral, or as sophisticated as developing a shared medical record to quickly identify individuals for the enhanced care program.

jeffbrennerupdated

“The Camden Coalition has tiered the problem at two different levels. We have a model of care for the most extreme patients, the outliers, and then a model of care for the next tier down of hospitalized patients. We are focused on hospitalized patients or patients using the emergency room. Both tiers are data driven. Every morning we wake up with a list of folks from the Health Information Exchange, a regional list, so we know how often they’ve been back in the last couple of years and the last six months. Anyone with two or more admissions in six months, with medical and social and behavioral complexity, is eligible for the higher tier, high utilizer intervention, and anyone with one admission or emergency room use is eligible for the next tier down. We don’t take referrals from anyone so the entire intervention is data-driven, coming from these lists every morning. We have a full-time team up in the hospital, and every morning they wake up with the list and they go right to the bedside, enroll patients, and offer the opportunity to enroll. Our interventions right now are being studied, so this is a randomized control trial. The care managers consent the patient, walk out, randomize them, and then walk in and let them know if they’ve been enrolled. If they get enrolled, it’s a 90 to 100-day intervention, intense, wrap-around case management. It looks a lot like programs such as an ACT team or a PACE program.”Jeffrey Brenner, MD, Executive Director, Camden Coalition of Healthcare Providers