Abstract
Background: Care coordination helps patients with complex needs, but heterogeneity in its implementation is not understood. Latent class analysis (LCA) was used to describe types of care coordination in primary care using data from The Minnesota Care Coordination Effectiveness Study (MNCARES), a large representative observational study of Minnesota clinics. We also explore whether program types are associated with clinic, community, or patient characteristics.
Methods: Primary care clinics with care coordination participating in MNCARES were included in this exploratory analysis. Care coordinators responded to survey items about their programs’ approaches to addressing social and complex medical needs, communication, care coordination volume, and support and resources available for care coordination. LCA was used to identify and describe distinct types of care coordination using 42 survey items. Bivariate analysis compared types to clinic, community, and patient characteristics.
Results: Four types of care coordination emerged across 316 clinics: type 1 a well-supported social/medical approach, type 2 a high volume social/medical approach, type 3 a well-resourced complex medical needs approach, and type 4 an onsite low volume approach. Type 1 clinics were more likely to have medical and community service access and serve younger patients and those born outside the US. Type 4 clinics were more likely urban with less community service access and served older adults.
Conclusion: This novel LCA approach successfully identified 4 distinct types of care coordination used by participating clinics. These results will help researchers to learn which approaches to care coordination are most effective in which contexts and help clinics decide how to operationalize care coordination.
Introduction
For people with complex health needs, the health care system in the United States can be challenging to navigate. In addition to managing self-care, patients and families must also navigate various medical organizations and specialties, often at different locations.1 Many of these patients also have substantial social needs, thus also trying to navigate a disjointed array of social services.2,3 Care coordination, the deliberate organization of patient care often facilitated by at least one health care professional together with a patient,4 is a promising strategy to navigate these complexities in primary care and improve patient outcomes.5 The potential benefits of care coordination are easily apparent, but its effectiveness overall and in various contexts or settings remains unclear.6 This is in large part due to documented variability in how care coordination is implemented, resulting in measurement issues that make evaluation difficult.7
Care coordination programs vary widely in how medical and social needs are addressed, what type of patients are eligible, how care coordinators communicate, and more.8–11 As a result, the field lacks information about how to best organize and deliver care coordination in different contexts. Even efforts like the Agency for Health care Research and Quality (AHRQ) Care Coordination Measures Atlas have not been able to fully help care systems or researchers specify the various elements that comprise care coordination or which lead to better outcomes.12,13 Data-driven methods, like latent class analysis (LCA), may be able to provide insight into commonalities and differences between care coordination programs. Such information could be a necessary input for developing frameworks to specify different types of care coordination programs - a missing link needed to understand what care coordination does and how it relates to patient outcomes. Through LCA, researchers can identify whether there are naturally occurring clusters or types of a given approach based on the characteristics within those approaches.14 This bottom-up method has yet to be applied to care coordination to understand whether distinct types exist in primary care. However, similar cluster analyses have been successfully used to develop taxonomies in similar health care settings15,16 and LCA has successfully been used to describe types of health care strategies or interventions delivered in other settings.17,18
In this study, LCA is used to uncover that question among primary care clinics participating in the Minnesota Care Coordination Effectiveness (MNCARES), a large observational study with details on how care coordination is being implemented in hundreds of diverse clinics across Minnesota. Identified care coordination types are compared with other clinic characteristics such as geography, size, and care coordination patient demographics. Such information has the potential to allow future researchers to describe and study the effectiveness of various approaches to care coordination in various contexts or settings and support health care leaders as they make decisions about what approaches to care coordination to implement.
Methods
Population and Design
Since 2010, Minnesota Department of Health (MDH) has certified primary care clinics in Minnesota as “health care homes” if they have a defined process for providing care coordination among other criteria (eg, electronic registry and performance). Certified clinics are required to document details of their approach in an application every 3 years but are allowed considerable latitude in approach. In 2020, participation in the MNCARES observational study was offered to all certified clinics in lieu of recertification. Of 415 certified primary care clinics, 36 were ineligible (eg, serving primarily pediatric patients, having fewer than 10 care coordination patients). Of the 379 remaining, 316 enrolled. The 63 clinics that were eligible but did not participate were primarily in urban locations (84%) and 40% were part of large health systems. The majority (n = 51) declined participation when initially approached and the remaining declined after initial acceptance but before providing any data or were nonresponsive to further outreach. The primary aims of MNCARES are to compare effectiveness of care coordination programs that do or do not have a social worker integrated in the care coordination team and to understand the care process and contextual factors associated with patient outcomes. The main study collected survey data from participating organizations, care coordinators, and patients. Survey data were merged with clinic patient data from electronic health records (EHR). The latter was aggregated by an independent data management organization that regularly collects clinical data for quality reporting across the state (Minnesota Community Measurement, MNCM). The exploratory analyses (data-driven rather than a priori hypothesis-driven) presented here use care coordinator survey data and patient demographic data from EHR. Additional information about the larger MNCARES study has been previously reported.11,19
Care Coordinator Survey
All 316 clinics enrolled in MNCARES (of the 379 eligible, 83%) were required to complete a survey about their current care coordination model. Study team members from MDH asked clinic liaisons to identify a care coordinator at each clinic best positioned to describe the current care coordination approach. If an identified care coordinator worked in multiple study-enrolled clinics, they were asked to answer applicable survey items separately for each clinic. In April 2022, these care coordinators were sent an invitation e-mail with a unique survey URL from the study e-mail address. Reminder e-mails were sent to nonresponders over 3 weeks. If no response was received, MDH and study liaisons worked to identify another care coordinator who would be able to provide responses. Because clinic participation in the care coordination survey was required to remain in the study in lieu of health care homes recertification, survey responses were sought from every participating clinic. Individuals who responded to the survey were not given an incentive but were sent a thank you e-mail signed by the MDH Director of Health Care Homes.
The survey was developed by a multi-disciplinary stakeholder team including patients, care coordinators, nurses, social workers, clinicians, researchers, and methodologists. The team identified a list of relevant constructs including clinic and care coordination structure, staff, referrals, services, communication, logistics, panel size, and clinic support using AHRQ’s Care Coordination Measurement Atlas as a guide.13 As no validated questionnaire existed to cover selected constructs, new survey items were designed following best practices in survey methods.20 The drafted survey was built in REDCap21 and was pilot tested in February 2022 by 10 care coordinators who worked in clinics not enrolled in MNCARES. Pilot questions asked for feedback on the survey itself, including the length, clarity, and appropriateness of questions. The pilot demonstrated acceptability with the exception of clarification of instructions for care coordinators that worked with more than one clinic and a few places where parenthetical descriptors were needed in question stems. The final survey was 79 items including items with ordinal and categorical response options and was designed to take about 20 minutes to complete.
A selection of 42 items related to the care coordination approach at a clinic were used in this exploratory latent class analysis (LCA). These items were selected and organized conceptually from the full survey by the above multi-disciplinary stakeholder team to comprehensively describe a care coordination approach. Items that were not selected described the health system, clinic, general patient population, or community beyond the care coordination program.11 Specifically, the survey items included in the LCA were close-ended items that documented the social and medical needs addressed by care coordination in clinics, the type of referrals to care coordination made, the type of care coordinator communication used, logistics related to care coordination in clinics, the volume of care coordination provided, and support and resources for care coordination.
Additional survey items asked responding care coordinators close-ended questions about their clinics and patient populations more broadly, including the use of EHR monitoring, availability of health care or community services available and patient financial constraints. These items were not included in LCA to create the care coordination typologies but were used in post hoc bivariate analyses.
Patient Demographic Data
Clinical demographic data were aggregated for patients receiving care coordination at participating clinics by MNCM, the independent data manager, as part of the EHR clinical dataset. These variables, including age, sex, race, ethnicity, primary language spoken, birthplace, and insurance payor, were mapped to clinic and use in post hoc bivariate analyses comparing across LCA types.
Analysis
Latent class analysis is an analytic method that identifies common response patterns reflecting theorized unobservable latent constructs and groups records with similar response patterns into mutually exclusive classes or types.14 For this work, LCA was selected over a cluster analysis approach for the compatibility with the survey item response option data type (categorical/binary) and the theoretical assumption that care coordination approaches are driven by latent multifaceted factors. Given the exploratory nature of this work, the ability to assess the probabilistic class categorizations provided by LCA was also considered a strength to evaluate the certainty of our final model-based groupings. LCA was used to identify and define distinct types of care coordination approaches using the above set of 42 survey items related to care coordination approach. Each survey item was dichotomized into categories that were most clinically meaningful and interpretable, as determined by the above-described multi-disciplinary stakeholder team through a series of meetings. Missingness within the survey items was quantified and described. Items with no response were grouped into the reference level (zero) during dichotomization. Presence of multicollinearity within the dichotomized analytic data were evaluated using a correlation matrix. Pairs of variables with r > |0.6| were reviewed by the multidisciplinary team and a qualitative judgment was made as to whether both were meaningful within the final typologies.
Four unique LCA models were run with the specified number of classes ranging from 2 to 5. Model fit statistics (Akaike Information Criterion [AIC], Bayesian Information Criterion [BIC])22,23 and predicted class population shares were used to select the final model, prioritizing parsimony and entropy while avoiding small classes (<10% of clinics).24 In addition to statistical indicators of best choice, identifiability/interpretability, and actionability/accessibility of the classes from the clinical perspective was considered when selecting and describing a final model.25 Descriptive summaries of the resulting classes were used to describe each typology’s defining characteristics. A subset of characteristics was chosen for descriptive display, organized by conceptual domain through a series of iterative meetings with the aforementioned multi-disciplinary stakeholder team. The team selected items that most distinguished between groups and were more conceptually distinct from one another for ease of visual interpretation. The entire list of items is provided in the Appendix. This team also qualitatively reviewed these characteristics with the goal of identifying a case description for each type identified empirically.
Lastly, bivariate tests were used to identify association of care coordination types with other clinic, health system, or community characteristics from the care coordinator survey or clinical patient demographic data.26 All analysis was performed in R using the poLCA package.27
Results
Clinic and Care Coordinator Characteristics
All the 316 clinics provided care coordination survey data (100% response rate). The majority of the clinics were in an urban geography and were part of a large health system (Table 1). Care coordinator-reported clinic access to various medical and social services was varied, with most having ready access to pharmacists, but fewer having access to community health workers. Approximately one quarter of care coordinators reported patients in their clinics having financial constraints that limited access to services. See Table 1 for additional details.
Descriptive Characteristics of Minnesota Primary Care Clinics with Care Coordination Participating in the MNCARES Study, n = 316 Clinics
Two hundred and twelve care coordinators completed surveys represented the above 316 clinics. Responding care coordinators were most likely to have a nursing degree (n = 163, 77%), serve as a care coordinator at only 1 study clinic (n = 151, 71%), work on average 21.6 hours per week as a care coordinator and have worked as a care coordinator in the study clinic for an average of 4.8 years (Table 2).
Descriptive Characteristics of the 212 Responding Care Coordinators Representing 316 Clinics Participating in the MNCARES Study
Care Coordination Types
Latent class analysis fit statistics and class distribution suggest 4 distinct types of care coordination exist in these data. As compared with the other models (2, 3, and 5 classes), 4 classes or types provided the best simultaneous minimization of AIC and BIC,22,23 maintained a high entropy (94%), and resulted in nontrivial predicted class membership population shares (15% to 48%).
When reviewed for collinearity, 5 pairs of items had correlations from r = 0.60 to r = 0.75. None had a correlation r > 0.75. The multidisciplinary team reviewed these pairs and determined qualitatively that each variable in the pair indeed represented distinct characteristics, thus all items were retained. After reviewing response patterns for all 42 items across the 4 identified care coordination program types, the multi-disciplinary stakeholder team organized items into domains. These domains were the care coordination program approach to addressing social needs, approach to addressing complex medical needs, care coordination communication, the volume of care coordination, and the support and resources available for care coordination. Two to 3 items that best differentiated the types qualitatively were then selected for presentation by the team. The response patterns for all 42 items overall and by type are provided in the Appendix. Figure 1 shows the overall proportion of the 316 clinics endorsing a selected item (light gray dot) compared with the proportion of clinics in a given type endorsing the same item (black dot). A thick black line denotes when more clinics in a type endorsed an item compared with clinics overall. A thin light gray line denotes when fewer clinics in a type endorsed an item compared with clinics overall.
Proportion of care coordination programs endorsing select survey items by type and domain, n = 316 clinics.
Care Coordination Program Type 1 – Well-Supported Social/Medical Approach
Care coordination program type 1 represents the largest number of programs (n = 153 clinics, 48% of the sample). Based on response patterns, the multi-disciplinary stakeholder team described these programs as having a strong social needs approach with the majority using social needs assessments to identify patients as eligible for care coordination and providing social needs-related referrals (eg, financial needs assessments and referrals). However, only approximately half of the care coordination programs in type 1 had a social worker on the care coordination team. The programs in this type were also described as having a focus on complex medical needs with a high proportion providing mental health needs assessments and referrals and facilitating specialty care services. The team also described programs in this type as being well-integrated into the primary care team as care coordinators regularly communicated in-person with clinicians, had a high volume of care coordination patients, felt their role was valued by clinicians and were satisfied with resources available for care coordination.
Care Coordination Program Type 2 – High Volume Social/Medical Approach
Care coordination program type 2 represents the second largest group (n = 67 clinics, 21% of the sample). In reviewing response patterns, the multi-disciplinary team described this set of care coordination programs as similarly focused on serving patients with social needs and offering social needs referrals, like type 1, but as more likely to have a social worker directly on the care coordination team. These programs also similarly focused on complex medical needs, but were less integrated into primary care, with less direct communication with primary care and perceiving clinicians as less likely to value the care coordination role. They are not often onsite or work with many clinicians and had larger patient panel sizes.
Care Coordination Program Type 3 – Well-Resourced Complex Medical Needs Approach
Programs in type 3 represented the smallest number of clinics (n = 46 clinics, 15% of the sample). The care coordination programs in this type were less likely to have a strong approach to addressing social needs among care coordination patients but did identify patients for care coordination based on complex medical needs. Care coordinators in these programs were less likely to be regularly onsite at a given clinic, work with fewer clinicians and have smaller patient panel sizes, but feel they have adequate time and resources available for care coordination programs.
Care Coordination Program Type 4 – Onsite Low Volume Approach
Lastly, programs in type 4 represented 50 clinics (16% of the sample). After reviewing response patterns, the multi-disciplinary team described this program type as likely providing essential medically focused care coordination with minimal additional social, behavioral, or complex medical needs support. The average care coordination program in this type was less likely to coordinate social needs through things like providing social needs assessment or referrals and was least likely to have a social worker on the care coordination team. The average care coordinator in this type of program manages a smaller patient panel and works with only a few clinicians and are less likely to perceive adequate time or support for care coordination but are likely onsite at clinics.
Care Coordination Types Related to Clinic, Community, and Patient Characteristics
Tables 3 and 4 show associations between care coordination program type and clinic, community, and patient characteristics. Care coordination programs in type 1 are more likely to be in clinics with reported ready access to community services (P < .01), pharmacists (P < .01), and surgical (P < .01) and other medical specialty services (P = .02). The programs in this type are also more likely to serve patients who are younger (13% were 18 to 34 years old), born outside of the United States (39%), do not speak English (21%), or are uninsured (5%). Care coordination programs in type 2 are most likely found in urban clinics (P < .01) with community health workers more readily available (P < .01). Care coordination patients in programs of this type are more likely to be older (63% 65+ years old) and White (74%) compared with care coordination patients in other types. Care coordination programs in type 3 are more likely to be in a rural clinic (P < .01) including small, satellite, or newer clinics with fewer patients, staff, and services. Patients in this type of care coordination program were likely to be born in the United States (95%) and be insured with a non-Medicaid insurance (94%). Care coordination programs in type 4 were likely to be in urban clinics (P < .01) with less access to specialty medical services (P < .01), community health workers (P < .01), or community services (P < .01). Care coordination patients in this type of program are likely to be older (60% 65+ years old), Asian (46%), or have Medicaid insurance (6%).
Relationship Between Care Coordination Program Type and Clinic and Community Characteristics, n = 316 Clinics
Relationship Between Care Coordination Program Type and Care Coordination Patient Characteristics, n = 12,676 Care Coordination Patients
Discussion
Latent class analysis uncovered 4 distinct types of care coordination programs in primary care defined by their social and medical needs approaches, communication, volume, and support for care coordination. These types ranged from ones that focused on social and complex medical needs and communicated closely with primary care clinicians to those that were less well-resourced and focused less on social or complex medical needs. This contrast with other descriptions of care coordination in the literature that use broader categories like the Framework for Care Coordination in Chronic and Complex Disease Management7,28 or the Social Ecological Model (S.E.M.)29 or have developed their own broad domains. In a recent systematic review, authors identified broad care coordination approaches by S.E.M. levels (eg, “service linkage” in the individual level, “interprofessional care” in the organizational level, and “multisectoral care” in the systems level).30,31 The characteristics that distinguished types in the current analysis could be reorganized to align with these broader categories but add finer grain detail about the specific approaches taken. For example, within the individual level “service linkages” approach, current type 1 and 2 programs were more likely than Types 3 and 4 to specifically provide financial needs and mental health service referrals and coordinate specialty care services. The current groupings empirically describe, in detail, how care coordination services are delivered in primary care and set the stage for understanding which approaches to care coordination may be most effective and in which settings. Other reviews have acknowledged the using theoretical frameworks to describe care coordination often do not lead to development of measurement tools to compare with effectiveness, an important next step for the field.32
The larger MNCARES study had an a priori aim to compare the effectiveness of 2 approaches to care coordination – a medically-focused approach or a medical-social approach distinguished by the absence or presence of a social worker on the care coordination team. This is one way to easily categorize care coordination in terms of the approach to addressing social needs.11 The exploratory, data-driven LCA presented here reveals that 2 of the 4 identified types focused on assessing or addressing social needs in addition to medical needs (Types 1 and 2). However, in type 1, only half of programs had a social worker on the care coordination team. A recent review organized approaches to social workers providing care coordination into categories including activities with patients, activities with other practitioners, activities that link the two, and cross-cutting activities.33 These authors posit that social workers are necessary for implementing these care coordination activities or at least are particularly well-positioned to do so. However, social worker on the care coordination team was only one of the social needs-related factors that distinguished between care coordination program types in this analysis. Future research should compare whether differences in clinical and patient-centered outcomes exists between various social-medical approaches and the other types more broadly.
Another key learning from this work was the difference in care coordination communication with the primary care team and how well resourced or supported care coordination was across types. Types of programs differed in terms of the degree and timing of communication between care coordinators and primary care clinicians, whether care coordinators were onsite, if clinicians were perceived as valuing the care coordinator role, and whether the responding care coordinator was satisfied with resources available for their work. For example, type 1 programs had care coordinators who interacted with primary care clinicians often, were onsite, and felt clinicians valued their role. Conversely, type 2 programs had care coordinators who were less likely to be onsite (about 50%), worked with a higher number of clinicians but interacted directly with them less frequently and did not view clinicians as strongly valuing the care coordination role. Early qualitative findings from the MNCARES study suggested these factors, especially support of the care coordination role, as critical for the success of a program.9 Communication and colocation have been priority interests in health care research, particularly as it relates to team-based care.34 These findings in combination allow for future research to test whether these factors in care coordination programs lead to better patient outcomes and to understand how to promote these factors, if effective.
Lastly, each of the 4 types were related to clinic, community, and patient characteristics. Care coordination programs defined as type 1 – the well-supported social and medical needs approach – were more likely to be in clinics that served patients who were younger and more likely to be born outside the US. Conversely, care coordination programs defined as type 4 – the onsite low volume approach – were more likely to be in urban clinics with less access to medical and community resources and serve patients who were more likely to be older and identify as Asian. Care coordination programs in type 3 – the well-resourced complex medical needs approach – were most likely found in rural clinics and programs in type 2 – the high volume social/medical approach – were more likely found in urban clinics and serve patients who identified as white (74%), though also served a higher proportion of patients who identified as Black or African American (15%) in comparison to other types. Results suggest that different care coordination approaches may arise organically to meet the needs or constraints of certain systems and settings. Others have found relationships between community characteristics and the investment in care coordination broadly, for example larger and urban hospitals with higher volume were more likely to invest in care coordination.35 Our findings extend this work by comparing 4 types of care coordination distinguished by different approach domains.
This work represents an important step in studying and improving the effectiveness of care coordination and has several notable strengths as well as some key limitations. The data presented here represent complete capture of all primary clinics certified as health care homes that were enrolled in the MNCARES study. However, they are not necessarily representative of clinics providing care coordination in other geographies. Surveys were also completed by a single care coordinator at a given clinic. It is possible other types may emerge in other settings or if including more perspectives on a given clinic’s care coordination approach (eg, clinicians, clinic leaders, patients). While care coordinators were asked to provide unique survey responses for each clinic they served, analysis did not account for correlation in responses from a given care coordination potentially influencing class membership probability. Although a methodological constraint, this data structure reflects pragmatic realities of staffing in geographically and organizationally linked areas, in which staff may provide care across multiple programs. In addition, although these data represent a detailed picture of how care coordinators perceived implementation in their clinics, it is a cross-sectional snapshot. Clinical care is constantly evolving and reacting to the surrounding context and patient needs. Our results, showing care coordination type associated with clinic, community, and patient population characteristics, highlight the importance of understanding the community in which care coordination exists. Lastly, the survey data forming the basis of this LCA uses single items or sets of items to capture complex constructs.
Future research could extend this work by developing theory-driven taxonomies15,16 of care coordination approaches and supporting measurement tools to understand how clinics move between types over time and with changing contexts and patient needs. This and future efforts will help to unpack what each care coordination type means and develop specific recommendations for clinics as they implement care coordination in the most effective and efficient ways for their context. Carefully measuring the component parts of how care coordination is implemented in primary care is critical for identifying and testing factors leading to the success of different approaches. This novel use of LCA offers a promising, data-driven approach to categorizing and comparing 4 different types of care coordination currently being used in primary care practices. This approach sets the groundwork for laying out a future framework for clinics at all stages of care coordination, whether seeking to initiate care coordination or grow an existing program. Most importantly, it will be important to understand how different care coordination types affect overall care and outcomes for patients with complex chronic health conditions in various settings.
Appendix
Response Patterns for All 42 Items Included in the Latent Class Analysis Determining the Four Types of Care Coordination Present in MNCARES Study Clinics, n = 316 Clinics
Notes
This article was externally peer reviewed.
Funding: This work was funded by the Patient-Centered Outcomes Research Institute® (PCORI®) Project Program Award (IHS-2019C1-15625). All statements presented in this article, including its findings and conclusions, are solely the responsibility of the authors and do not necessarily represent the official views of PCORI®.
Confict of interest: Besides working for health systems or other organizations being supported by the above PCORI award, the authors have no conflicts of interest to declare.
To see this article online, please go to: http://jabfm.org/content/38/3/500.full.
- Received for publication August 23, 2024.
- Revision received December 10, 2024.
- Accepted for publication January 6, 2025.
