Identifying and Stratifying Members
After selecting the eligible population for a care management program, States must consider how to identify eligible members and how best to target their resources (e.g., care managers, interventions, funding). Many States provide more intense care management services to the most high-risk or high-cost members. States can implement a risk stratification tool or a predictive model to categorize members in high-, medium-, and low-risk categories. States also consider which members are most "impactable" by using individual-level tools, such as health assessments, the Patient Activation Measure,n or other measurement and screening tools. To apply their interventions most effectively, States can use a combination of population-based tools and individual-level tools.
States can identify eligible members using a variety and combination of methods, including claims data, physician referral, self-identification, hospital discharge planners, predictive modeling, and referral from other State agencies, non-profit organizations, and social services. When using claims data to identify potential members, States should ensure that they have the necessary means to run the claims data internally or that their vendor employs the necessary staff.
Oklahoma identifies and enrolls members into the care management program in the following ways:
- Calls by Medicaid beneficiaries to the SoonerCare Helpline.
- Referrals from physicians, nurses, family, or friends.
- Claims and encounter data, including inappropriate emergency room (ER) utilization.
- Medicaid eligibility workers.
Stratification allows States to direct program resources most appropriately. To stratify members, States can develop an internal tool or rely on predictive modeling software.
State-specific stratification tool. To stratify members, States have used disease, comorbidity, health assessments, and health care costs, as well as other criteria (e.g., age, gender). State staff can develop a stratification methodology relying on claims and demographic data.
To determine the appropriateness of the selected variables and methodology, States should test the data periodically and depend on their familiarity with the program's population to assess the stratification tool's accuracy. Analyzing the member population to derive a unique stratification model typically proves more accurate for predicting risk than using a generic model. States should identify a segment of the population where costs are modifiable and analyze factors that predict high-cost or high-risk.
States also should consider their program model and interventions to determine which behaviors they can target. For example, if the program employs interventions to target ER utilization or medication adherence, ER utilization rates or pharmacy claims data should be included in the stratification tool.
Checklist: Primary Steps in Stratification
√ Identify target variables.
Indiana program staff used 2 years of retrospective claims data to derive a predictive model for stratifying members to low- and high-intensity care management interventions. They compared the predictive value of several different approaches for using claims-based predictors in a particular year to forecast total claims paid in the following year. Of the many predictors modeled, the final algorithm involved three:
- Total net Medicaid claims costs in the past year.
- Medicaid aid category.
- Total number of unique medications filled in the past year.
Indiana also allows its nurse care managers to reassign some members to low-risk interventions, even after the stratification model identifies them as a high risk. Based on their experience, nurse care managers can also graduate members from high-risk interventions if self-care knowledge and skills are sufficient or to reassign members for whom care management is unlikely to improve their health status.
North Carolina staff developed their stratification tool internally. To design the process, Community Care of North Carolina (CCNC) met with care managers who decided, based on their experience, to target individuals with a high number of ER visits or inpatient claims. To communicate information on high-risk members to care managers, CCNC sends case identification reports that list individual high-risk members. Care managers review these reports to identify members currently in care management, members that have appeared on the list multiple times, and new members. Based on these reports, care managers prioritize outreach and care management quarterly.
Patient Activation Measureo
The Patient Activation Measure (PAM), developed by Dr. Judy Hibbard and colleagues at the University of Oregon, assesses the knowledge, skills, and confidence for managing one's own health and health care. The PAM segments members into one of four progressively higher activation levels:
Reliable and valid for use with patients managing a wide range of chronic illnesses, the PAM has been found to be helpful wherever a patient has a significant role to play in managing his or her health. The PAM can be used for the following purposes:
Predictive modeling. States use predictive modeling to identify and stratify members for care management programs. Predictive models can improve program efficiency by identifying members who are or might become a high risk. Predictive modeling is defined as the "process of analyzing currently available data to prospectively identify specific individuals who are at high-risk of having adverse outcomes in the near future."p States can employ predictive modeling in a variety of ways, including to:
- Identify eligible members for inclusion in care management interventions.
- Focus on especially high-cost members.
- Help target appropriate populations effectively.
- Conduct needs assessments and quality improvement.
- Provide "actuarial" risk information to inform financial decisions.
States typically contract with an external vendor for predictive modeling. In working with a predictive model vendor, States should understand how the tool works and what information will be available should the contract end or be awarded to another vendor. Although most predictive modeling software is proprietary, States should be involved in the identification and stratification algorithms or criteria used to identify high-risk members to ensure that the model is working appropriately for their program. As with the State-developed tools described above, program staff should rely on their knowledge of the program members and provide feedback to the vendor on the predictive model's success.
Pennsylvania's vendor has a proprietary risk stratification methodology to assign members to three tiers, each associated with a different level of intervention. Members can move among the tiers, if necessary, and each member receives an assessment every 3 months. Self-management is a component of all three tiers.
- Level One. Educational mailings, nurse call line, audio health libraries. (The member receives a call from a nurse care manager at weeks 6 and 26.)
- Level Two. All Level One services plus more frequent calls from a nurse care manager.
- Level Three. All Level One and Level Two services plus in-person visits and a more intensive call schedule.
Many States use health assessments, in addition to claims data, to stratify members across risk categories. Health assessments provide information to care management program staff to gauge a member's health condition, create treatment plans, and evaluate programs. Health assessments can be administered by the member's primary care physician, a nurse care manager, or they can be self-administered. In many instances, assessments occur the first time program staff contact identified members. Staff should use this time to actively engage members in the program. Some States have included motivational interviewing with the administration of the assessment for this purpose.
Lessons Learned: Health Assessments
States have used assessment tools that are publicly available, purchased, created by the State, or adapted by the State to fit its needs. Commonly used assessments include the following:
- EuroQol. The EuroQol (EQ-5D) survey asks five questions about mobility, self-care, daily living, pain, and depression. Initially developed for administration in an in-person interview format, EQ-5D has been adapted for telephonic delivery.q
- Short Form Health Survey. The Short Form Health Survey comes in three versions: 36, 12, and eight questions. Its eight domains encompass physical function, role limitations due to physical functioning, general health perceptions, bodily pain, social functioning, energy/vitality, role limitations due to emotional functioning, and mental health.r
- Patient Assessment of Chronic Illness Care. A brief, validated patient self-report instrument, the Patient Assessment of Chronic Illness Care (PACIC) gauges the extent to which members with chronic illness receive care that aligns with the Chronic Care Model. Developed by Improving Chronic Illness Care staff, the PACIC asks users to respond to 20 questions that yield results on patient activation, delivery system design, decision support, goal-setting, problem-solving, and follow-up.s
- Patient Health Questionnaire. The Patient Health Questionnaire (PHQ) is a tool designed to help primary care physicians diagnose depression as well as select treatment options. Versions of the PHQ include two-question, eight-question, and nine-question surveys. Several Learning Network States, including Iowa and Indiana, use the PHQ in their care management programs.t
Selecting Opt-In versus Opt-Out
An important consideration in designing an enrollment and outreach strategy is whether the program is opt-in or opt-out. Opt-in programs notify members of their eligibility for the program; members then must actively choose to enroll. In opt-out programs, members are enrolled automatically but have the option to disenroll themselves. Both enrollment mechanisms have advantages and disadvantages.
Opt-out programs generally have higher member enrollment than opt-in programs. By easing the enrollment process, the opt-out model allows members to try the program even if they are hesitant. States have found that once members participate in the program they rarely choose to disenroll or opt out. However, many of the members in opt-out programs might not be actively "engaged." In Texas, an average 21 percent of members opt out of the program, with COPD having the highest opt-out rate at 40 percent and CHF having the lowest rate at 0 percent. Illinois has a lower rate of opt-outs, with only 3 percent of the population choosing not to participate.
With an opt-in program, a stronger likelihood exists that the vendor will be able to engage most of the members successfully. In this model, States might assume that all of the members enrolled want to improve their health conditions through program participation. In Kansas, 21 percent of eligible members have opted into the care management program. Washington's two care management vendors have estimated opt-in rates of 22 percent and 44 percent.
Iowa encountered difficulty recruiting people for its opt-in asthma program. The State initially identified 1,312 high-use asthmatics for program outreach but were able to reach only one-third of them through telephone calls. Deciding to revise its outreach strategy, Iowa sent letters to the high users, but only 17 responded and enrolled. Iowa realized that to enroll 250 members it would have to open the program to the entire population of asthmatics, which led to the enrollment of 266 members with asthma.
Of the 13 States in the initial Medicaid Care Management Learning Network, five operate opt-in programs; eight run opt-out programs.
States use various techniques successfully to enroll members in their care management programs: outreach materials, hospital enrollments, physician referrals, consumer incentives, and coordination with other organizations. Many States choose to employ a combination of these techniques to make certain they target all eligible members.
With most programs, once eligible members have been identified, program staff often send letters and information regarding enrollment. Typically the member is required to sign and return the application or call a number to enroll (opt-in program) or disenroll (opt-out program). With opt-in programs, States should expect a low response rate with their first round of mailings, but members who respond generally will prove the most motivated. To maximize program enrollment, program staff might follow up with additional mailings or telephone calls to non-respondent members. In sending member outreach information, ensuring that materials are developed at an adequate level of health literacy is important so members can understand them easily. Please go to Section 4: Selecting Care Management Interventions for additional information on health literacy.
Many States, including North Carolina, Indiana, Illinois, and Rhode Island, place nurse care managers in hospitals to reach out to potential members on discharge. These nurses enroll and provide more information to members who were hospitalized for chronic conditions covered in the State's care management program. By targeting eligible members on discharge, the State hopes members will be more inclined to participate in interventions that might improve the health problems that caused their recent hospitalization. States have been successful at increasing enrollment by reaching members at these "teachable moments" to encourage them to make a difference regarding their health care decisions.
Lessons Learned: Enrolling Members
Most States employ physician referrals to identify potential members for enrollment in a care management program. In general, if a physician provides care to a Medicaid beneficiary who might qualify for enrollment, the physician will alert the program, whose staff then will contact the individual. By offering the option for physician referrals, States have been successful at increasing enrollment and reinforcing the importance of care management through physician support. Physician referrals actively engage the provider in the program, which can lead to greater physician buy-in. Please go to Section 2: Engaging Stakeholders in a Care Management Program for more information on strategies to engage providers.
Coordination with Other Organizations
Other organizations, such as community health centers, long-term care facilities, or other government agencies might provide a venue for enrollment efforts. Serving a high percentage of Medicaid beneficiaries, community health centers understand the population's cultural and social environment. Programs can work with these organizations to enroll members in care management programs.
As a technique to encourage member enrollment in a care management program, some States are considering offering members a small incentive. Understanding that the effectiveness of incentives is still being debated, many States have chosen to test the usefulness of incentives for increasing member participation. Programs can use a variety of financial and non-financial incentives, such as cash, gift cards, small gifts, lotteries, or redeemable credit, to encourage enrollment in a care management program. Please go to Section 4: Selecting Care Management Interventions for additional information on consumer incentives.
Kansas' vendor offers consumer incentives of $10 or less. The vendor establishes relationships with new members by taking fruit baskets staff have assembled to in-home visits. As an incentive to attend orientation meetings, program staff deliver $10 grocery store gift certificates. In addition, the vendor has established an agreement with a local YMCA to provide membership scholarships. Currently, 26 care management members participate in the initiative with the YMCA. The vendor tracks members' gym usage to ensure that the care management members on scholarship are using the gym.
Systematically planning which populations to cover and how to identify members for enrollment will help maximize the potential impact of a State's care management program. By identifying the population, program staff can move forward with tailoring interventions to meet member needs. With successful identification, enrollment, and stratification, the program will reach the appropriate populations.
n. Hibbard JH, Stockard J, Mahoney ER, et al. Development of the Patient Activation Measure (PAM): conceptualizing and measuring activation in patients and consumers. Health Serv Res 2004; 39(4):1005-26.
p. Center for Health Care Strategies. Using risk-adjustment for payment and predictive modeling for care management in Medicaid: A WebExchange. Princeton, NJ March 2005.
q. Available at: EurpQol. EQ-5D. http://www.euroqol.org. Accessed November 16, 2007.
r. Available at: SF-36. http://www.sf-36.org. SF-36.org. Accessed November 16, 2007.
s. Available at: Robert Wood Johnson Foundation. Improving Chronic Illness Care, Patient Assessment of Chronic Illness Care, http://www.improvingchroniccare.org/index.php?p=PACIC_Survey&s=36. Accessed November 16, 2007.
t. Available at: Pfizer Inc. Patient Health Questionnaire (PHQ) Screeners. http://www.phqscreeners.com. Accessed November 16, 2007.