A Framework for an Intelligent Social Engagement Support System: Identifying and Addressing Challenges at Multiple Levels to Reduce Health Disparities

Abstract

Increasingly managed care organizations (MCOs) – health plans that offer covered health services through networks of healthcare providers and hospitals – are recognizing the role of social drivers of health on both health outcomes and unnecessary healthcare utilization (e.g., emergency department visits) and are taking actions. They are employing social care navigators (SCNs) and screening for social needs such as food, housing, and transportation. However, individuals with social needs can experience barriers to completing the screening and to receiving social care when referred. Additionally, the volume of people with social needs means that SCNs have high caseloads. Without guidance on how to prioritize their efforts, the people who have the highest social needs may not be identified or receive SCN support. To address the many problems, we developed a framework for an intelligent social engagement support system. This framework builds on the Health Care System domain of the National Institutes on Minority Health and Health Disparities’ Health Disparities Research Framework as a foundation, leverages learning health system and community-engaged approaches, and incorporates machine learning at key points to create an adaptable, multi-level framework that is intended to have numerous positive outcomes, including reducing health disparities in socially-driven health outcomes. We discuss how we are using this framework and how we envision it could be adapted to different MCO contexts.

Introduction

There is significant evidence of the role of social drivers of health (SDoH) such as lack of access to healthy food, inadequate housing, and insufficient social supports on health outcomes (e.g., incidence of acute and chronic diseases, complications from conditions)^1–5 and socially-driven healthcare utilization (e.g., emergency department, ED, visits)^6–10. Indeed, some estimates show up to 80% of health outcomes are attributable to social, economic, or behavioral factors². For instance, discrimination and the associated barriers to care have been linked to multiple mental health conditions, including anxiety and depression¹¹, as well as to cardiovascular health¹². In addition, children from lower socioeconomic backgrounds are more likely to have physical and mental health issues¹³. In another example, a study found that SDoH were associated with ED visits among older adults with multiple chronic conditions⁶.

Despite this evidence, the U.S. health system has struggled to adequately address these important factors. Several approaches are currently being used, and each have their own unique challenges. One of the most common approaches is healthcare provider-initiated social needs screening, which tends to center on certain primary care visits (e.g., pediatric) and hospitalizations. Although there are advantages to this approach such as meeting people where they are at and normalizing social needs as an important part of one’s health, there are also a number of barriers, including a majority of healthcare providers not being formally trained in SDoH^14,15, limiting their awareness and ability to adequately address SDoH; heterogenous opinions among health providers of integrating social care into medical practice^16–19, which could lead to uneven implementation; resource limitations that result in unsustainable caseloads of people who need social care^20–23; and logistical challenges such as coordinating with other providers and community-based organizations (CBOs)^20,24–30.

In a promising approach that takes the burden off individual healthcare providers, some managed care organizations (MCOs) are taking the lead to try to identify and address social needs³¹. MCOs are health plans that offer covered health services through a network of doctors, hospitals, and other healthcare providers³². They are increasingly recognizing the potential to improve health outcomes and reduce healthcare spending due to unnecessary healthcare utilization such as emergency department (ED) visits³³. MCOs are taking actions such as employing social care navigators, including social workers, care coordinators, nurses, and community health workers, and screening members for social needs, including food, housing, utilities, interpersonal violence, and transportation^33,34. A common model being followed is the Centers for Medicare and Medicaid’s (CMS) Accountable Health Communities (AHC) Model, which highlights the importance of screening, referral, and navigation support³⁵. While these are relatively new programs with few published evaluations, there are two common pain points described in the literature where guidance is needed to optimally direct resources and that, without this guidance, may prevent the programs from achieving their full potential – (1) identifying individuals with social care needs^14,36–39 and, (2) supporting people in receiving the services for which they were referred^27,29,37,40.

Given the importance of SDoH to health outcomes and the existing challenges faced by attempts to integrate social care into the health system, there is an urgent need for more theoretical and empirical research to guide these efforts and ensure that the people most at risk receive the care and services that they need. Here, we developed a theoretical framework to address these challenges and provide formal processes for ongoing engagement, evaluation, and improvement that could be used to develop or improve existing MCO social care programs. While empirical testing is still needed, we argue that implementing this data-driven framework that directs resources where they are needed most has the potential to decrease disparities in socially-driven health outcomes, as well as reduce unnecessary healthcare utilization to ease the burden on the U.S. healthcare system.

Methods

We leveraged the Health Care System domain of the National Institutes on Minority Health and Health Disparities’ (NIMHD) Health Disparities Research Framework as a foundation⁴¹, utilized learning health system⁴² and community-engaged approaches⁴³, and incorporated machine learning methods, to build a framework for an intelligent social engagement support system. This framework aims to continuously identify and address the current challenges with the existing approach to social care, working at multiple levels – individual, interpersonal, community, and societal – to, ultimately, improve health outcomes and reduce health disparities in socially-driven health outcomes.

Results

Table 1 shows how the Intelligent Social Engagement Support System Framework addresses challenges identified in the literature and the intended outcomes at each level of the NIMHD Disparities Research Framework⁴¹ and Figure 1 shows the key components and workflow for the intelligent social engagement support system. In this section, we go through the major components of the Intelligent Social Engagement Support System Framework: member-centered intelligence (individual member level); knowledge and resource management (social care navigator-member interaction level); community-based organization activation (community level); and ethics, equity, and governance (societal level). The presentation of these components will highlight the community-engaged approach and consideration of multiple levels of influence that are central to this framework.

Table 1.

Summary of how the Intelligent Social Engagement Support System Framework addresses identified challenges and the intended outcomes at each level of the NIMHD Disparities Research Framework.

	NIMHD Disparities Research Framework Levels of Influence
Individual Members	Interpersonal Social Care Navigator-Member	Community Community-based Organizations	Societal
Identified Challenges	High prevalence of unmet social needs38 Barriers to multiple stages for receiving social care: ▪ completing screening36,44,45 ▪ receiving referral33,46 ▪ accessing care at CBO46,47	High caseloads make it impossible to reach everyone and limits the time social care navigators have to build relationships with members48–50	Are not well-integrated into the healthcare system14,49	Approaches that rely only on manual efforts and target only one level of influence (e.g., individual) are limited in their societal benefits51,52
Targeted Innovation	Member-centered Intelligence Identifying barriers at multiple stages to improve system (i.e., learning from failure points) Following up with members at high risk for unnecessary ED utilization who do not complete the screening Algorithm to identify members with social needs that have characteristics that may prevent them from successful engagement for proactive contact from a social care navigator	Knowledge and Resource Management Algorithm to prioritize those at high risk for unnecessary ED utilization and reduce caseloads at the screening stage Formalize knowledge in the system to learn from unsuccessful social care efforts Algorithm to identify members with social needs that have characteristics that may prevent them from successful engagement for proactive contact from a navigator	Activating Community CBOs engaged throughout the system such as in understanding member barriers at multiple stages and to connect with communities with a large population of high-risk members Formalize knowledge in the system to learn from unsuccessful social care efforts	Centering Ethics, Equity, and Governance Assessing and correcting for bias in training data Human-centered approach to system development Rigorous evaluation of system Leverage interdependencie s at each level and targeted technological innovation to create a fair and scalable framework
Intended Outcomes	More members screened for social needs More members referred for social services More members receiving social services Lower rates of unnecessary healthcare utilization (e.g., ED visits) Better health outcomes	More time spent with high-risk members Better relationship with members	Continually improved social engagement support system Better communication and continuity of care between healthcare and social care providers	Reduce health disparities Establish adaptive long-term interventions Inform policy

Acronyms: CBO = Community-based organization; ED = emergency department

Figure 1.

The key components, workflow for the intelligent social engagement support system, and select outcomes.

Member-centered Intelligence

At its core, the health information system underlying this framework tracks member behavior and engagement history, identifies risk of social needs and disengagement points along the social care continuum, proactively addresses member barriers, and learns from engagement failures. This health information system addresses two key problems with existing social care programs – (1) not all patients are screened or fully complete the screening³⁶ and (2) individuals with identified social needs often do not receive social care^29,40. Both publicly available macro-level data – environmental, census, and geographic data, as well as proxy information that could be risk factors – and micro-level, individual member data – administrative data, claims, demographic data, diagnoses, symptoms, history of utilization – are stored in a social needs repository. New information will be continually added to this repository throughout the system, including after screenings and when barriers are identified through investigations of failure points (FPs), where members fail to engage with the social care system (see Figure 1). Importantly, having the social care navigators in-the-loop is expected to keep the new micro-level data quality high, as they will speak to and follow-up with members to fill in gaps as needed.

Two algorithms will be trained on data in the social needs repository – (1) to categorize members into high and low risk of non-emergency ED visits and (2) to identify characteristics that may prevent successful engagement with community-based organizations. The first algorithm is meant to target resources, particularly social care navigators (see more below), in following-up with the members most likely to have unmet social needs based on their risk of unnecessary ED utilization (i.e., socially-driven healthcare utilization)6. The second algorithm is meant to alert social care navigators to barriers that a member with unmet social needs is likely to have, so that these can be proactively addressed through personalized interventions to increase the likelihood that the member will receive the social care to which they are referred (see more below).

We are using a predictive modeling approach for the risk stratified model that begins with a discrete time survival model of whether a socially-driven ED visit occurred within a given month⁵⁶. We plan to compare this base model to several others, including regularized logistic regression, tree-based models, and support vector machines and select the model with the best balance of predictive accuracy, interpretability, and computation time. We are prioritizing interpretability and transparency in our modeling approach over computational complexity to ensure that it is understood and trusted by social care navigators. We are using the same approach for the second algorithm, adjusting for data sources and variable selection as needed.

At the member-level, there are several intended outcomes, including more members screened for social needs, more members with social needs receiving community-based social services, lower rates of unnecessary healthcare utilization (e.g., ED visits), and improved health outcomes.

Knowledge and Resource Management

This framework aims to formalize knowledge in the social needs repository that currently exists, in part, with social care navigators or that could be collected by these individuals about why members are not receiving the social care to which they have been referred. The social engagement support system uses this information, and information on member risk-level, to intelligently guide resources to the members that need it the most. Part of this is also using multiple modes of communication to engage the members, including text, phone, and in-person events to ensure reach and access, as well as to accommodate personal preferences^53,54. The framework addresses the key challenges of incredibly high caseloads for social care navigators that can keep them from reaching out to the highest need members and limit the time that they have to build relationships with these members. Existing research suggests that relationships with social workers can improve engagement with social care²⁹. It is, therefore, critical, to identify ways to help social care navigators to prioritize the cases that will most benefit from follow-up and additional time and effort. Ultimately, managing knowledge and resources should improve the interactions between social care navigators and members.

The intended outcomes of the knowledge and resource management component of the framework include improved social care navigator-member relationships. In addition, we hypothesize that there will be additional benefits at other levels, including increased likelihood that members will connect with needed resources (member benefit) and reduced unnecessary ED visits since additional resources will be targeted to support members at the highest risk for this type of visit (health system benefit). This also highlights the interrelated nature of the framework, where elements targeted to one level of influence may also have benefits at other levels.

Community-based Organization Activation

Community-based organizations are a critical part of the social care system but are often an underutilized resource of knowledge of the target communities. In this framework, engaging the community-based organizations at all points of the intelligent social engagement system will not only better integrate them into the health system and recognize their critical role, but will also learn from their knowledge of the communities and the barriers people from these communities face to getting social care to improve the system and increase the number of people with social needs receiving services. Ultimately, this should lead to better communication and continuity of care between healthcare and social care providers.

Ethics, Equity, and Governance

Ethics, equity, and governance – from the approach to algorithm development to the stakeholder engagement at all steps – are core values of this framework. While the algorithms are targeted to two key points in the system where they could have maximum benefits, without careful assessment for and correction of any biases, these algorithms could create a less fair and equitable system that, when scaled up, do more harm than good. Thus, we are developing a modeling paradigm that can be adapted to new target domains efficiently to ensure scalability through the accurate propagation of the machine learning-based engagement system. This includes testing for historical bias, applying corrective filters through which to train the production model, and ensuring the model does not re-calibrate with differential predictions across protected classes when re-trained and scaled.

In addition, we take a human-centered approach to system development, evaluation, and governance that promotes ethical practices. We are engaging stakeholders including health plan leadership, social care navigators, and community-based health organizations in the development of the intelligent social engagement support system to ensure that it is trustworthy, useful, and usable. We will adjust plans based on stakeholder feedback, especially regarding concerns that have ethical implications. We will continue to engage stakeholders in the governance of the system.

These components are critical to the societal-level of the framework. The intended outcomes include reducing health disparities and informing policies such as the CMS Section 1115 Medicaid Waiver, which expanded states’ ability to address health-related social needs³¹.

Discussion and Conclusions

The proposed framework builds on existing frameworks and concepts to offer a multi-level view for how to take a member-centered, targeted approach to improve the identification of social needs and the number of individuals with social needs who receive care, with the ultimate goal of improving health outcomes and reducing health disparities. This approach answers the call made by Veinot et al. in 2019 to “level up” health informatics interventions to enhance health equity⁵⁵. It fills an important gap in thinking about how a population health information system can work with social care navigators or other healthcare professionals to provide targeted support to multiple levels. Ultimately, it offers a flexible theoretical framework that can be adapted to multiple contexts.

Currently, we are in the process of applying this framework within the context of one Medicaid MCO (Maryland Physicians Care, a Maryland Medicaid health plan) by enhancing their existing social information technology infrastructure with a set of machine learning models for risk identification, an engagement support system to maximize members’ use of social supports, and a continuous qualitative and quantitative improvement process to establish a learning health system. The algorithm and system development is currently underway, and we will be evaluating whether the framework achieves the intended outcomes in the coming years.

In addition, the framework can be adapted to other MCOs who may be at different points in the development of their social needs programs or have differing information system capabilities. We envision this adaptation would include consideration of the specific context of the MCO such as where they are in their social care program development, their resources, and the specific challenges faced in implementing the social care program (e.g., availability or quality of data). While the framework accounts for the common challenges described in the literature and for our partner, it is possible that some organizations may not have some of the challenges but may have others. An implementation science approach⁵⁷ could help to adapt the framework to the context. Specifically, identifying the key challenges for the organization’s social needs program and then using evidence-based strategies at the multiple levels of influence to address these challenges. In this way, the framework could be personalized to the MCO’s needs and support their intended outcomes. However, there are certain elements that we argue are of universal benefit such as thinking of the social engagement support system as a learning health system where information about member’s lack of engagement or disengagement is constantly being gathered and used to improve the system. Formalizing this information could also advance the science of engagement through publication of the common barriers in different contexts and effective methods of addressing those barriers⁵⁸.

To conclude, there is an urgent need to identify and meet social needs to improve health outcomes, reduce health disparities, and reduce the operational and financial strain on the healthcare system from unnecessary healthcare utilization. There is great potential for a framework that addresses multiple levels – from individual to societal – in improving the supports for engagement in social care programs. Our proposed framework leverages existing frameworks and approaches to provide targeted innovations that address key barriers and achieve intended outcomes.