Skip to main content

Advanced Search

Advanced Search

Current Filters

Filter your query


Publication Types


Area of Focus
Date Range

Area of Focus

Improving Health Care Quality

Issue Briefs


How Accountable Care Organizations Use Population Segmentation to Care for High-Need, High-Cost Patients

high-need, high-cost patients in hospital


  • Issue: New payment and care delivery models such as accountable care organizations (ACOs) have prompted health care delivery systems to better meet the requirements of their high-need, high-cost (HNHC) patients.
  • Goal: To explore how a group of mature ACOs are seeking to match patients with appropriate interventions by segmenting HNHC populations with similar needs into smaller subgroups.
  • Methods: Semistructured telephone interviews with 34 leaders from 18 mature ACOs and 10 national experts knowledgeable about risk stratification and segmentation.
  • Key Findings and Conclusions: ACOs use a range of approaches to segment their HNHC patients. Although there was no consistent set of subgroups for HNHC patients across ACOs, there were some common ones. Respondents noted that when primary care clinicians were engaged in refining segmentation approaches, there was an increase in both the clinical relevance of the results as well as the willingness of frontline providers to use them. Population segmentation results informed ACOs’ understanding of program needs, for example, by helping them better understand what skill sets and staff were needed to deliver enhanced care management. Findings on how mature ACOs are segmenting their HNHC population can improve the future development of more systematic approaches.


Five percent of the U.S. population has complex medical and behavioral or social needs, but this group accounts for 50 percent of the country’s health care spending.1 New payment and care delivery models such as accountable care organizations (ACOs) have prompted decision-makers at health care delivery systems to seek the best ways to meet these patients’ needs while controlling costs.2

To this end, many ACOs have used predictive modeling and risk stratification to sort their entire population into risk levels (such as low, medium, and high). ACOs typically linked their high-risk patients to the ACO’s general care management program. This approach has had mixed results, perhaps because high-risk patients have wide-ranging, heterogeneous needs, and different care management services benefit certain kinds of high-risk patients more than others.3

Fewer ACOs have taken the approach of subdividing (segmenting) this high-need, high-cost (HNHC) population into smaller subgroups with similar needs.4 The National Academy of Medicine and others have highlighted the importance of recognizing that all HNHC patients are not alike, and recommend segmentation of HNHC patients.5 It is theorized that segmentation will allow ACOs to better match patients to appropriate interventions, enabling them to provide higher-quality care and allocate limited resources more effectively. Interventions are most effective when they target the patients that they were intended to serve.6 For example, an intervention might include outreach to socially isolated patients with congestive heart failure (CHF); additional social support might improve their medical condition and avoid preventable emergency department (ED) visits.

Because few ACOs have tackled segmentation of HNHC patients,7 little is known about the best approach. To better understand the use of segmentation, we look beyond the few most visible efforts8 to explore how mature ACOs segment their HNHC adult population, as well as the challenges these initiatives face.


We completed interviews with 44 respondents: 10 national experts and 34 respondents from 18 ACOs. Most ACO respondents were medical directors, executives, care management program leads, clinician leaders, or data analytics leads. ACOs’ characteristics were balanced by region, type (Medicare Shared Savings Program [MSSP], Next Generation, Medicaid9), ownership type, and size of population served (see Appendix).

Population Segmentation Goals and Team Make-Up

In tackling risk stratification and segmentation, some ACOs’ goals are aspirational: improving patient outcomes, reducing costs, and achieving the Triple Aim.10 ACOs also hope to inform program management by improving their understanding of several elements: which patients are high cost, and why; which patients have needs that health care organizations could address; how to allocate resources, such as staff, to care teams; and how to help teams prioritize workloads. They also want to identify the needs of HNHC subgroups, identify any additional necessary training of care management staff, and determine manageable panel sizes for care managers or teams.

ACO teams conducting population segmentation typically include ACO chief medical officers, chief executives, population health leads, care coordination or care management program leads, data analytics leads, and practicing physician representatives (such as those from clinical leadership committees). To tailor care for the identified subgroups, teams add more frontline clinicians such as primary care physicians (PCPs), nurse care managers, social workers, care transition staff, and behavioral health providers.

Approaches to Population Segmentation

Most ACOs use both quantitative information, such as claims data, and qualitative data, including clinician assessments, to risk-stratify their population. This hybrid approach seems to offer the best compromise between consistent implementation and clinical salience. All 18 ACOs use claims data, utilization data, and/or reports from payers to risk-stratify their entire population. Sixteen ACOs also use limited clinical data elements from their electronic health records (EHRs) to inform risk stratification. In many of these, ACOs or third-party vendors employ an algorithm to analyze the available structured data and compute a numeric risk score. Based on this score, they typically classify their entire ACO population into low-, medium-, and high-risk groups. Several ACOs also identify a “rising risk” group. Some national experts and ACO respondents reported that numeric risk scores from vendors were not actionable because patients with the same risk scores could have wide-ranging needs, and the output lacked sufficient clinical context.

While all ACOs interviewed engage in whole-population risk stratification, some further segment their HNHC patients into subgroups. Some ACOs describe this process as sequential, with risk stratification preceding the segmentation of HNHC patients into smaller subgroups. Alternatively, the two efforts can occur as part of a single process. However, a few ACOs first identified patients with particular conditions or combinations of conditions, and then performed risk stratification and segmentation within those groups to determine which patients should receive more intensive and tailored care management.

Of the 13 ACOs reporting HNHC population subgroups, seven define their subgroups by incorporating clinical evaluation and risk assessment data that have been gathered in person from patients. Only four of these ACOs use data on patients’ social and behavioral needs in the segmentation process. Most ACOs identify these needs during patient assessments made while tailoring care management services for HNHC patients, rather than during segmentation.

There are numerous challenges to accurately and efficiently capturing data on social and behavioral needs for risk stratification and segmentation. One challenge is documenting meaningful social and behavioral health data in a discrete structured format in current EHRs. Systematic data on social needs are also scarce at both the population and individual patient levels. Given that social service agencies and community organizations already collect their own data on substance abuse, housing, and food programs, there is a need for improved data coordination between them and health care delivery systems.

Among ACOs that incorporate social and behavioral health needs into segmentation, some use a hands-on approach while others opt for more automated tactics. For example, Rio Grande Valley ACO, an MSSP with clinics in Texas and New Jersey, takes a hands-on approach (Exhibit 1). Its interdisciplinary clinical team employs a tool to categorize HNHC patients into subgroups based on four domains: the patient’s medical neighborhood; social support; medical status and trajectory; and self-management and coping skills, and mental health. Each subgroup is then assigned to an appropriate level of care management. In contrast, Montefiore ACO uses a highly automated approach to segmentation, incorporating claims and pharmacy data as well as indicators of patients’ psychosocial needs (Exhibit 2). Montefiore’s Next Generation ACO, an integrated hospital and physician entity in The Bronx, New York, serves 55,000 Medicare patients who typically receive medical care from Montefiore over their lifespan. Montefiore ACO has strong, in-house analytic capabilities and involves patients’ PCPs after segmentation is complete.

Although there is no consistent set of subgroups into which ACOs segment their HNHC patients, certain subgroups are common. These subgroups include frail elderly, advanced illness (palliative, hospice, and end-of-life care), transitional care, homebound, comorbid medical conditions (often including diabetes, CHF, or chronic obstructive pulmonary disease [COPD]), comorbid medical and mental health conditions, chronic care rising risk, disabled, and end-stage renal disease (ESRD). The national experts and ACO clinicians in our study cautioned against using single disease-focused segments, because they risk missing the underlying cause of a patient’s problems or fail to address comorbid conditions. ACOs identify frail elderly patients in a variety of ways: clinician referral, in-person clinical frailty assessments, in-house or vendor analyses based on diagnoses, claims-based utilization and patient demographics data, and frailty constructs such as the Johns Hopkins Adjusted Clinical Groups (ACG) System.11

Engaging PCPs to refine their segmentation approaches can increase the usefulness of results, as well as frontline providers’ willingness to use them. Involvement of primary care teams can help address PCPs’ initial skepticism and concern that an ACO is “interfering” in their patients’ care. ACOs use provider input to adapt algorithms to include variables that are particularly important to their population. For example, one interviewee said they “constantly solicit provider feedback,” noting that “three physicians found issues with the algorithm not accurately identifying patients with chronic kidney disease and some basic mental health issues.” Based on physician feedback, “we went back and layered GFR [glomerular filtration rate] values and PHQ-9 [Patient Health Questionnaire-9] data so these patients would be picked up in the high- and rising-risk categories.”12 A few ACOs have a team of clinicians that identifies important variables to include in their algorithms.

Many ACOs ask the PCP or other clinical staff to review the results of their segmented high-risk patient subgroups. They allow clinical staff to add or remove patients, using their clinical judgment of who could benefit from enhanced care management. A medical director described how to engage frontline providers early in the segmentation process: the ACO must carve out time in the providers’ schedule “30 minutes a week for a month, where you pull them off the front line, they don’t see patients, the nurse sits down with them, and they look at the list.” Conversely, a few ACOs do not seek clinician input; for them, risk stratification and segmentation “happen behind the scenes.”

Some ACO and national expert respondents said it was important to communicate segmentation results to frontline clinicians in a transparent, accessible, and actionable way — such as a banner or button in the EHR that indicates the patient’s risk group. In at least one ACO, clinicians also can click the button to see the top 10 variables used to calculate the patient’s risk level. In another ACO, the patient’s risk score is “literally a flag in the electronic record with a pulldown tab to get in touch with the care manager.”

Even among ACOs pursuing population segmentation of HNHC patients, only a few go beyond preexisting care management programs to further tailor care to those subgroups. ACOs that tailor care to subgroups use existing disease-specific care management programs, such as a program for ESRD patients. They also create new or modify existing care management efforts based on the needs of various subgroups. Most respondents stressed the importance of keeping HNHC patients with their usual primary care practice while adding an enhanced layer of care management. That might mean embedding a care manager in the primary care site or using a care manager or care management team housed elsewhere in the organization. Tailoring care for subgroups typically includes addressing the care management team’s clinical backgrounds and care management skills, or the frequency, duration, and type of the team’s contacts (home visits or phone calls, for example).

The care management team usually adapts an enhanced care management approach for individual patients within a high-risk subgroup, based on in-person or telephone-administered risk assessments conducted by a nurse care manager or nurse care coordinator. At several ACOs, physicians and lead care managers are heavily involved in designing or identifying existing risk assessment tools that guide how care is tailored.

ACOs struggle to tailor care to HNHC subgroups when lack of funding limited their ability to hire enough care managers. Care management staff are sometimes so busy with current high utilizers that they lack resources to reach out to rising-risk patients. And many are frustrated with the lack of coordination among care management programs from different health plans and initiatives. As one ACO clinician observed:

[Care management is] siloed and business-driven, not patient-driven. Why do we have nurse care managers in primary care? Because someone’s paying us to do it in the [primary care demonstration] program. Why do we have nurse care managers doing discharge planning in the hospital? Because DRG [diagnosis-related group] payments make that a valuable activity from the hospital’s perspective. Why don’t we have nurse care managers managing our cystic fibrosis patient population? Because nobody pays for it.

Refinement of Population Segmentation Approaches over Time

National experts and ACO respondents stressed the need for ongoing feedback loops. To improve its utility, they either refined or completely replaced their segmentation approach over time. A few ACOs used continuous feedback loops that incorporated short-term process measures, such as chronic condition control and rates of emergency department utilization.

Respondents offered examples of improvements made to risk stratification and segmentation approaches after such assessments: incorporating new or more current data sources, such as EHR data; enhancing the collection of social and behavioral health data; modifying the care management team (hiring more social workers, for example, or reallocating care managers); and changing relationships with third-party vendors. Process refinement often includes better engagement of frontline clinicians as well as more oversight from formal physician advisory committees.

Challenges to Assessing Effectiveness of Population Segmentation and Care Management

Although care management informed by risk stratification and segmentation can help improve program management and some process measures, changes in cost or quality outcomes cannot necessarily be attributed to these efforts. Some study respondents noted improvements, including a decline in admission rates for particular conditions (CHF and COPD, for example); reduced ED visits; increased contact with patients who had not contacted the system in the prior two years; increased use of evidence-based preventive services; and improved patient self-confidence in their ability to manage their chronic conditions. ACO respondents also noted that population segmentation influenced program management goals.

Respondents noted numerous challenges to quantitatively assessing the effectiveness of current risk stratification, segmentation, and care management approaches. These include:

  • Regression to the mean.13
  • Small sample sizes of high-risk subgroups within an ACO, resulting in insufficient statistical power to assess effects on outcomes.
  • Limited actionability of claims data because of the time required for health plans to process claims, as well as claims’ lack of clinical nuance.
  • Cost of integrating EHR data when ACO medical practices use different EHR platforms.
  • Difficulty of establishing causality when ACOs participate in simultaneous initiatives, such as same-day appointments or efforts to reduce readmissions and increase access to urgent care clinics.

Exhibit 3 summarizes respondents’ collective advice to ACOs new to population segmentation.

Vendors’ claims of achieving savings can be hard to validate, as some respondents reported. One ACO physician said “they did not provide the statistical analysis that [would let] me know for sure that they’re not just reporting regression to the mean.” Another ACO physician noted that both vendors and ACOs “face immense pressure . . . to come up with any data that supports their work.” This respondent stated it is unrealistic to expect “you could hire a turn-key solution from the outside and drop it on top of existing practices and within a year have a positive outcome.”


In this report, we described how 18 mature ACOs approach population segmentation and tailor their resources. While all the ACOs in our sample risk-stratify their entire population to identify high-need, high-cost patients, only two-thirds segment the HNHC patient population into smaller subgroups to identify those with similar needs. Most have in place a sequential process, with risk stratification preceding the segmentation of HNHC patients into subgroups. A few first identify patients with particular conditions, or combinations of conditions, and then perform risk stratification and segmentation within those groups. This latter approach is similar to one taken by Denver Health.14

Similar to the results of prior research,15 our study finds that algorithms based solely on claims data do not capture sufficient information on clinical, behavioral health, or social needs. On the other hand, prior research documents the challenges of solely relying on patient-completed health risk assessments or clinician judgment to identify individual patients for care management.16 Like others,17 we find that hybrid approaches — using both quantitative and qualitative data to segment a population and identify patients most likely to benefit from care management — offered the best compromise between consistent implementation and clinical salience. Although there are no consistent sets of subgroups into which ACOs further segment their high-risk patients, ACO respondents in our study frequently identify certain subgroups. High-risk subgroups sometimes correspond to categories supported by their existing care management programs, in part because of funding and expediency. Others adapt existing programs or create new ones for some subgroups. ACOs use their segmentation results to help determine manageable patient panel sizes, as well as how to allocate staff resources and workforce training to their care management teams.

Although our qualitative sample has good variation by ACO and respondent characteristics, we cannot generalize from our study to all ACOs, or even to all mature ACOs.

Challenges and Emerging Opportunities

Respondents identified several challenges to population segmentation and resource tailoring, as well as potential strategies to address them. Ongoing needs include:

  • Improving the availability of current, accurate data on patients’ clinical, functional, social, and behavioral health needs.
  • Strengthening analytic and clinical resources.
  • Improving the evaluability of segmentation and care management programs.

Limited availability of current and accurate data. ACO respondents reported the need for timely, high-quality clinical data that can capture patients’ current risk factors more accurately than claims data; this sentiment has been described by others.18 Using the most recent patient information recorded in the EHR might allow the segmentation results to more accurately reflect the current needs of the patient, particularly compared to using claims data.

ACOs also struggle to capture data on their patients’ social and behavioral health needs that can systematically be used in the segmentation process. Although clinicians may already record social and behavioral health needs in a text field in the patient’s record, these data cannot be readily used in an algorithm that stratifies patients by risk.

ACOs could especially benefit from tailoring enhanced care management services to patients’ functional status. Frail people with poor functional status, for example, are challenged by carrying out activities of daily living, and drive higher costs over time. To tailor services, however, ACOs would need to create new structured data or access existing data. For example, ACOs could work with their EHR vendors to develop a standardized assessment of social and behavioral health needs, including functional status. Moreover, health care delivery organizations and government and social service programs (for example, corrections, foster care, or the Supplemental Nutrition Assistance Program) could enter into data-sharing agreements. These collaborations could help ACOs determine which patients need particular services.

Resource-intensive processes. Though many mature ACOs do their risk stratification and segmentation in-house, others lack the technical infrastructure, funding, and workforce to do so. ACOs without in-house analytic capabilities often find the risk stratification and segmentation process to be a “heavy lift,” and some relied on third-party vendors to support their work.

Involving frontline clinicians in the segmentation process was a time-intensive activity, but one that could make the overall process more efficient. Involving frontline clinicians reportedly makes them more likely to accept the results of segmentation, which in turn affects whether patients accept enhanced care management services. Clinician input also helps tailor services to patients’ needs. To reduce the burden on busy clinicians, some ACOs seek this input from a select subgroup of knowledgeable physicians, as well as from other clinical staff.

Improving the evaluability of segmentation and tailored care programs. A very large ACO may be able to quantitatively evaluate its own program,19 but small and medium ACOs often lack adequate sample sizes of HNHC patients. Methods for real-world evaluations of such programs across health delivery organizations exist,20 but we first need a better understanding of what population segmentation looks like on the ground. We hope this paper adds to a growing knowledge base.

The complex financing of health care in the United States also complicates ACOs’ abilities to evaluate their programs. ACOs find themselves torn between meeting the reporting requirements and quality measure goals of different payers and programs and analyzing data for internal evaluations of program impact. Furthermore, some respondents note that payer initiatives’ concern for annual costs influence ACOs. It leads them to apply that narrow, short-term focus to their internal evaluations of segmentation and tailored care programs, instead of considering the impact on multiyear costs or broader population health outcomes. If ACOs could move beyond these short-term requirements, they might focus more on true population health by segmenting along the lifespan to address the root causes of patients’ needs.21


How We Conducted This Study

We studied Medicare ACOs and a few Medicaid ACOs operating under Centers for Medicare and Medicaid Services authority that had been in place for at least three years, or that had a long history of risk contracting before becoming an ACO. We wanted to hear from well-established health care delivery organizations that had developed incentives to control costs. We did not interview representatives from Medicare Advantage plans because they are typically not health care providers, and they face a variety of local issues that affect how they interact with their network and local payers. We focused on their approaches to risk stratification, segmentation, and tailoring care to their adult patient population. The New England Institutional Review Board (NEIRB) determined that this study was exempt from NEIRB review (WO-1-20071-1).

Sample Identification

Before interviewing ACO respondents, we interviewed national experts knowledgeable about risk stratification and segmentation; we identified them based on our literature review and referrals from experts in the field.22

We used two data sources to identify ACOs for interviews. The National Association of Accountable Care Organizations (NAACOS) provided us with a list of the 50 “most mature” ACOs participating in NAACOS activities and events. We emailed the contact for each ACO, explaining the purpose of our study, and asked the following: whether they pursued risk stratification and segmentation; whether they used that information to decide how to deliver care to high-risk subgroups; and whether they would be willing to put us in touch with the individual who led those efforts, for a potential interview. To reach ACOs in regions not captured by volunteers from the NAACOS’ list, we purposively identified additional ACOs from Becker’s Hospital Review.23

Semistructured Interview Content

We used two separate protocols with parallel content that was tailored to either national expert or ACO respondents. We asked national experts about their experiences with, and views of, ACOs’ approaches to risk stratification, segmentation, and tailoring of health care resources. Within these three areas, we explored a variety of topics:

  1. Terminology ACOs use for risk stratification and segmentation.
  2. How ACOs define their target population for segmentation.
  3. Types of staff participating on the teams conducting population segmentation.
  4. Segmentation goals.
  5. Description of processes and data sources, and involvement of third-party vendors in population segmentation.
  6. Whether and how social support and behavioral health needs are incorporated into risk stratification and segmentation.
  7. How clinicians are involved in population segmentation.
  8. How clinicians have reacted to risk stratification, segmentation, and output.
  9. Strengths and weaknesses of population segmentation approaches.
  10. How, if at all, ACOs assess or consider patient interest in care management as part of the segmentation process.
  11. How, if at all, they assess and refine their risk stratification and segmentation approaches over time.
  12. How they used segmentation results to tailor care, and if they try to evaluate health outcomes.
  13. How respondents would approach risk stratification and segmentation if they could focus on long-term, multiple year outcomes rather than annual outcomes.
  14. Advice for ACOs or other entities interested in segmenting their HNHC population and tailoring care to resulting subgroups.

Data Collection

We interviewed national experts in early 2017 and ACO respondents in mid-2017. On average, we interviewed two respondents per ACO. Interviews lasted from 60 to 90 minutes. We audio recorded and transcribed all interviews. Characteristics of our respondents are summarized in the Appendix.


We developed our initial code dictionary based on our literature review24 and refined it based on themes that emerged from respondents’ comments.25 We coded the interview transcripts using Atlas.ti qualitative analysis software (version 7.5.10), meeting weekly to verify coding and minimize researcher bias.


1. Steven B. Cohen and William Yu, The Concentration and Persistence in the Level of Health Expenditures Over Time: Estimates for the U.S. Population, 2008–2009, Statistical Brief No. 354 (Agency for Healthcare Research and Quality, Jan. 2012); and Peter Long et al., Effective Care for High-Need Patients: Opportunities for Improving Outcomes, Value, and Health (National Academy of Medicine, 2017).

2. Long et al., Effective Care, 2017; Jose F. Figueroa and Ashish K. Jha, “Approach for Achieving Effective Care for High-Need Patients,” JAMA Internal Medicine 178, no. 6 (June 2018): 845–46; David Blumenthal et al., “Caring for High-Need, High-Cost Patients — An Urgent Priority,” New England Journal of Medicine 375, no. 10 (Sept. 8, 2016): 909–11; and J. Lester Feder, “Predictive Modeling and Team Care for High-Need Patients at HealthCare Partners,” Health Affairs 30, no. 3 (Mar. 2011): 416–18.

3. John Hsu et al., “Bending the Spending Curve by Altering Care Delivery Patterns: The Role of Care Management Within a Pioneer ACO,” Health Affairs 36, no. 5 (May 2017): 876–84; and Jonathan Stokes et al., “Effectiveness of Case Management for ‘At Risk’ Patients in Primary Care: A Systematic Review and Meta-Analysis,” PLoS ONE 10, no. 7 (July 17, 2015): e0132340.

4. Martha Hostetter and Sarah Klein, “In Focus: Segmenting Populations to Tailor Services, Improve Care,” Quality Matters (e-newsletter), Commonwealth Fund, June 26, 2015; Long et al., Effective Care, 2017; and Karen E. Joynt et al., “Segmenting Potentially High-Cost Medicare Patients into Actionable Cohorts,” Healthcare 5, no. 1–2 (Mar. 2017): 62–67.

5. Long et al., Effective Care, 2017; and Susan L. Hayes et al., High-Need, High-Cost Patients: Who Are They and How Do They Use Health Care?: A Population-Based Comparison of Demographics, Health Care Use, and Expenditures (Commonwealth Fund, Aug. 2016).

6. Gerard F. Anderson et al., “Attributes Common to Programs That Successfully Treat High-Need, High-Cost Individuals,” American Journal of Managed Care 21, no. 11 (Nov. 2015): e597–e600.

7. Long et al., Effective Care, 2017; Melinda Abrams et al., “Overview of Segmentation of High-Need, High-Cost Patient Population” Presentation, National Academy of Medicine, Jan. 19, 2016; and Dana Jean-Baptiste, Ann O’Malley, and Tanya Shah, Population Segmentation and Tailoring of Health Care Resources: Findings from a Literature Review, Working Paper 58 (Mathematica Policy Research, Dec. 2017).

8. Clemens S. Hong, Andrew S. Hwang, and Timothy G. Ferris, Finding a Match: How Successful Complex Care Programs Identify Patients (California HealthCare Foundation, Mar. 2015); Hsu et al., “Bending the Spending Curve,” 2017; Tracy L. Johnson et al., “Augmenting Predictive Modeling Tools with Clinical Insights for Care Coordination Program Design and Implementation,” eGEMS (Generating Evidence & Methods to Improve Patient Outcomes) 3, no. 1 (July 2015): 1181; and Sabine I. Vuik, Erik K. Mayer, and Ara Darzi, “Patient Segmentation Analysis Offers Significant Benefits for Integrated Care and Support,” Health Affairs 35, no. 5 (May 2016): 769–75.

9. Centers for Medicare and Medicaid Services, “Accountable Care Organizations (ACOs): General Information” (CMS, last updated Dec. 4, 2018).

10. Donald M. Berwick, Thomas W. Nolan, and John Whittington, “The Triple Aim: Care, Health, and Cost,” Health Affairs 27, no. 3 (May/June 2008): 759–69.

11. Johns Hopkins University, “The ACG® System Version 11.0 Technical Reference Guide” (JHU, Dec. 9, 2014).

12. Kurt Kroenke, Robert L. Spitzer, and Janet B. W. Williams, “The PHQ-9: Validity of a Brief Depression Severity Measure,” Journal of General Internal Medicine 16, no. 9 (Sept. 2001): 606–13.

13. Graham Upton and Ian Cook, Oxford Dictionary of Statistics (Oxford Paperback Reference, 2008). Regression to the mean is the phenomenon that if a variable, such as health care expenditures, is extreme on its first measurement, it will tend to be closer to the average on its second measurement; if it is extreme on its second measurement, it will tend to have been closer to the average on its first.

14. Johnson et al., “Augmenting Predictive Modeling,” 2015.

15. Johnson et al., “Augmenting Predictive Modeling,” 2015; Jean-Baptiste, O’Malley, and Shah, Population Segmentation, 2017; J. Frank Wharam and Jonathan P. Weiner, “The Promise and Peril of Healthcare Forecasting,” American Journal of Managed Care 18, no. 3 (2012): e82–e85; and Richard H. Bernstein, “New Arrows in the Quiver for Targeting Care Management: High-Risk Versus High-Opportunity Case Identification,” Journal of Ambulatory Care Management 30, no. 1 (Jan.–Mar. 2007): 39–51.

16. Christine Vogeli et al., “Implementing a Hybrid Approach to Select Patients for Care Management: Variations Across Practices,” American Journal of Managed Care 22, no. 5 (May 2016): 358–65.

17. Hong, Hwang, and Ferris, Finding a Match, 2015; Johnson et al., “Augmenting Predictive Modeling,” 2015; and Clemens S. Hong, Allison L. Siegel, and Timothy G. Ferris, Caring for High-Need, High-Cost Patients: What Makes for a Successful Care Management Program? (Commonwealth Fund, Aug. 2014).

18. Craig Schneider et al., “Reflections on the Pioneer ACO Program,” Presentation, AcademyHealth Annual Research Meeting, June 28, 2016.

19. Hsu et al., “Bending the Spending Curve,” 2017.

20. Jelena Zurovac et al., Effectiveness of Alternative Ways of Implementing Care Management Components in Medicare D-SNPs: The Brand New Day Study (U.S. Department of Health and Human Services, Office of the Assistant Secretary for Planning and Evaluation, Jan. 2014); and Deborah Peikes et al., “Effects of Care Coordination on Hospitalization, Quality of Care, and Health Care Expenditures Among Medicare Beneficiaries: 15 Randomized Trials,” Journal of the American Medical Association 301, no. 6, (Feb. 11, 2009): 603–18.

21. Joanne Lynn et al., “Using Population Segmentation to Provide Better Health Care for All: The ‘Bridges to Health’ Model,” Milbank Quarterly 85, no. 2 (June 2007): 185–208.

22. Jean-Baptiste, O’Malley, and Shah, Population Segmentation, 2017.

23. Brooke Murphy, Erin Dietsche, and Emily Rappleye, “100 ACOs to Know in 2016,” Becker’s Hospital Review, Aug. 18, 2016.

24. Jean-Baptiste, O’Malley, and Shah, Population Segmentation, 2017.

25. Matthew B. Miles, A. Michael Huberman, and Johnny Saldaña, Qualitative Data Analysis: A Methods Sourcebook (Sage Publications, 2014); and Johnny Saldaña, The Coding Manual for Qualitative Researchers (Sage Publications, 2016).

Publication Details



Ann S. O’Malley, Senior Fellow

[email protected]


Ann S. O’Malley et al., How Accountable Care Organizations Use Population Segmentation to Care for High-Need, High-Cost Patients (Commonwealth Fund, Jan. 2019).