Abstract
Background. Many out-of-care people living with HIV have unmet basic needs and are served by loosely connected agencies. Prior research suggests that increasing agencies’ coordination may lead to higher quality and better coordinated care. This study examines four U.S. interagency networks in AIDS United’s HIV linkage and retention in care program. This study explores changes in the networks of implementing agencies. Methods. Each network included a lead agency and collaborators. One administrator and service provider per agency completed an online survey about collaboration prior to and during Positive Charge. We measured how many organizations were connected to one another through density, or the proportion of reported connections out of all possible connections between organizations. Network centralization was measured to investigate whether this network connectivity was due to one or more highly connected organizations or not. To compare collaboration by type, density and centralization were calculated for any collaboration and specific collaboration types: technical assistance, shared resources, information exchange, and boosting access. To characterize the frequency of collaboration, we examined how often organizations interacted by “monthly or greater” versus “less than monthly.” Results. Density increased in all networks. Density was highest for information exchange and referring clients. When results were restricted to “monthly or greater,” the densities of all networks were lower. Conclusions. This study suggests that a targeted linkage to care initiative may increase some collaboration types among organizations serving people living with HIV. It also provides insights to policy makers about how such networks may evolve.
Linking people living with HIV (PLWH) to regular, ongoing primary care is vital for addressing the U.S. HIV epidemic (Mugavero, Amico, Horn, & Thompson, 2013). Doing so, however, is challenging as target groups are sometimes extremely underserved (Haley et al., 2014). Many out-of-care PLWH have unmet basic needs such as untreated mental illness, ongoing substance use, and housing instability that prevent them from seeking care (Rumptz et al., 2007; Sherer et al., 2002). This population is often served by a loosely connected system of care that includes nonprofit AIDS service organizations (ASOs), local health departments, and local clinics (Mugavero, Norton, & Saag, 2011). Reasons for this lack of integration may be that organizations are tasked with achieving their own missions and may not be coordinated or linked with other agencies to meet the needs of PLWH optimally. However, the existing needs of these individuals (e.g., housing and employment) are often beyond what these organizations can provide individually. Hence, such organizations must form substantive partnerships characterized by sharing resources and coordinating services across organizations (Bloxham, 1997).
In the literature, this type of coordination is called “interagency collaboration” or “interorganizational collaboration” and is defined as “mutually beneficial and well-defined relationships entered into by two or more organizations to achieve a common goal” (Mattessich & Monsey, 1992). It can be particularly important for the effective delivery of health services in an era where there is increased recognition of the relationship between health outcomes and the “up-stream” social factors that have historically been addressed by social support agencies (Dryfoos, 1994). In this setting, examples of such factors include homelessness, joblessness, and substance abuse. Previous studies have found that such collaboration boosts efficiency and performance through exchange of resources, knowledge, and social capital (Parmigiani & Rivera-Santos, 2011). Other benefits include less duplication of efforts, increases in each organization’s power and legitimacy, achieving a greater number of process outcomes, greater trust and value of partners, higher agreement about priority outcomes, rationalization of resources, having rare resources as part of the network, and provision of superior services (Jones, Thomas, & Rudd, 2004; Parmigiani & Rivera-Santos, 2011; Retrum, Chapman, & Varda, 2013). Most studies, however, have not examined population health outcomes (Retrum et al., 2013).
To date, few peer-reviewed publications have examined interagency collaboration among organizations serving PLWH using network analytic techniques. To our knowledge, none have examined networks in multiple geographies at more than one time point. Previous studies on these types of networks include a study of 30 agencies in Baltimore that suggests agencies are relatively well connected but collaborate on an ad hoc basis rather than through formalized agreements (Kwait, Valente, & Celentano, 2001). A later study of this network found that schools and corporations were often not a part of this network despite their important role in serving PLWH. Most organizations did not interact each month (Khosla, 2013). Previous research has also found that organizations not interacting with partners at least monthly were often on the periphery of interorganizational networks (Khosla, 2013; Thomas, Isler, Carter, & Torrone, 2007). Frequency is a property characterizing relationships between organizations that has been explored previously (Knoke & Yang, 2008; Kwait et al., 2001; Madlberger, 2008). In addition, earlier studies found that information exchange and client referrals were the most often observed collaboration types while resource sharing or technical assistance were least often observed (Provan, Nakama, Veazie, Teufel-Shone, & Huddleston, 2003; Thomas et al., 2007; Wright & Shuff, 1995). One study found that memoranda of understanding (MoUs) in these contexts, which formalized collaboration between organizations, to be largely ineffective. This study found that frontline staff were often not aware of MoUs and that MoUs did not necessarily lead to collaborations between signing organizations (Khosla, Marsteller, & Holtgrave, 2013).
AIDS United is an organization that seeks to end the U.S. AIDS epidemic through a broad portfolio of activities including grant making, policy, and formative research. In 2010, AIDS United launched Positive Charge (PC), a program to link out-of-care PLWH to care in five U.S. locations. As a requirement of PC, each lead grantee agency had at least two collaborating partner agencies with which grant funds were shared. All networks had at least four collaborating agencies. Lead agencies were ASOs, academic centers, and an HIV special needs insurance plan designed for PLWH. Collaborating agencies were often other ASOs, local departments of public health, health care providers, and social services providers. Grant amounts ranged from $1.1 million to $2.2 million. The lead agency was responsible for disbursing funds to partners and managing the overarching activities of the project and partners to meet intervention goals. Due to the structure, grantees anticipated that this project would foster tighter interorganizational collaborations in their geography, particularly through information exchange and referrals. Grantees did not know the number of connections formed across organizations (measured by density), whether those connections were primarily due to a small number of well-connected organizations or distributed throughout the network (measured by network centralization), how often organizations would interact (measured by frequency), and the type of collaboration between those organizations.
This article aims to address gaps in the literature on interorganizational collaboration among agencies that serve PLWH by describing how organizations collaborated to provide linkage to care services. This article explores the extent to which funding formal collaboration may alter the density, network centralization, and frequency of various collaboration types (overall, referrals, shared resources, information exchange, and technical assistance).
Method
Setting and Design
Positive Charge took place at five locations around the United States: multiple regions of the state of North Carolina; San Francisco/Oakland of California; Chicago, Illinois; New York City, New York; and multiple cities in the state of Louisiana. However, only four locations are included in this article because data are unavailable for one network. The projects began in July 2010 and ended in August 2013. All five grantees and partners sought to link out-of-care PLWH to regular medical care and support their clients to stay retained in care (Table 1). Their programs were tailored to address the needs of out-of-care individuals in the local area, which was severely affected by HIV. Each location had a lead organization and collaborated with local partners to implement PC. Each PC program had a variety of partner types including community-based organizations (CBOs), CBOs that focused exclusively on PLWH (ASOs), clinics, academic institutions, and state or county health departments (Table 2). All five lead agencies had received a grant from AIDS United to support these linkage and retention programs. The PC programs and their evaluation have been described in greater detail elsewhere (Jain et al., 2016; Kim et al., 2014; Maulsby et al., 2015). While basic information about the networks of implementing organizations has been presented (Jain et al., 2016), this study builds on prior work by examining collaboration at all locations by type (collaboration, access, information exchange, and technical assistance) and frequency before and during intervention.
Positive Charge Program Profiles.
Positive Charge Sites and Partners.
Note. CBO = community-based organization; ASO = AIDS service organization.
Measures
We calculated density and network centralization. Density measures the proportion of the actual versus possible number of network ties. The term network ties refers to a reported collaboration between two organizations. A network tie is reciprocal when both organizations report collaborating with one another. A network tie is nonreciprocal if one organization reports collaboration with another organization but the other organization does not report this collaboration. Density ranges between 0 and 1. In this context, density is used to explore the extent to which PC implementing organizations are connected to one another. Density’s companion measure, network centralization, quantifies the extent to which network connections result from one organization’s connections to the others (high network centralization) or whether all organizations are connected to one another (low network centralization; Freeman, 1979). Due to the structure of the networks explored in this article, network centralization is used to investigate whether network density is primarily due to the lead organization’s connections or more distributed across the network.
We assessed the frequency of each type of collaboration. To do so, categorical responses were dichotomized to reporting collaboration at least monthly versus less frequently as done by a previous study (Khosla, 2013). Resource sharing was not included because one of its dimensions (sharing staff across organizations) cannot be explored in terms of frequency. Similarly, the survey questions on overall collaboration did not include or seek information on frequency.
Data Collection
Project data were collected 1 to 2 years following project initiation to allow for completion of any start-up activities. Project start dates ranged from July 2010 to August 2011. Data collection start dates ranged from summer 2011 to summer 2012. The organizations included in the PC networks were identified using two methods: document reviews and interviewing staff at the project’s lead agency. All key informants participated in the online survey. To gain a breadth of responses, there were two individuals surveyed and interviewed within each organization 1 to 2 years following project initiation. Each individual was surveyed and interviewed one time. One of these individuals was in a “service” role, or directly interacted with clients, and the other was in an administrative or leadership position and often supervised those in “service” roles. The survey questions drew on the list of organizations in PC networks and asked which organizations were working together and in what capacity at the time of the survey and 6 months prior to project implementation. The researchers sent online surveys three times by e-mail and then contacted the study coordinator at the lead agency for help in gathering these survey responses. The institutional review board at the Johns Hopkins School of Public Health determined this work not to be human subjects research.
The questions were adapted from two sources. We adopted a checklist approach where respondents checked off organizations they had worked with, which is a common method for collecting network data (Kwait et al., 2001; Provan et al., 2003; Thomas et al., 2007). Reliability and validity testing was not conducted, but these questions have strong face validity, were reviewed by various experts in the field, and have been used in previous studies (Kwait et al., 2001; Messeri & Kiperman, 1994). Respondents were asked how often they conducted a list of activities, and these activities were classified into collaboration types as shown in Table 3. Respondents could specify that they conducted each of these activities “never,” “several times a year or less,” “once a month,” “several times a month,” “once a week,” “several times a week,” “once a day,” “more than once a day.” Respondents could also select “not applicable” or “don’t know.”
Constructs Included in Each Collaboration Type.
The survey also included a question on whether or not staff were shared across organizations; this question did not ask about frequency of staff sharing.
Analysis
Network analysis relied on survey data, which were converted into matrices. The matrices included the lead and partner organizations as headers in each row and column. “1” indicated collaboration was reported between the organizations named in the corresponding row and column and “0” indicated no collaboration was reported. These matrices were then loaded into UCINET to calculate measures of network centralization, average node degree, and density during and prior to PC (Borgatti, Everett, & Freeman 2002).
Sociograms are charts that plot relationships between entities, in this case organizations. We created sociograms of PC networks 6 months prior to PC and during PC (Figures 1-4). Each dot represents an organization and is labeled with the acronym for the organization’s name (Table 2). The sociograms were created using UCINET’s companion software, NetDraw (Borgatti et al., 2002). For PC networks with completed surveys from all involved organizations, nonreciprocal ties were included in the analysis. For networks where completed surveys were not available from all organizations, ties between organizations where data from one organization were missing were assumed to be bidirectional.
Louisiana Positive Charge network sociograms. (a) 6 months prior to Positive Charge; (b) during Positive Charge.
Chicago Positive Charge network sociograms, during and 6 months prior to Positive Charge.a
North Carolina Positive Charge Network sociograms. (a) 6 months prior to Positive Charge; (b) during Positive Charge.
San Francisco/Bay Area Positive Charge network sociograms. (a) 6 months prior to Positive Charge; (b) during Positive Charge.
Results
The organizational networks described in this analysis include CBOs, ASOs, clinics, health departments, academic institutions, and county jails. We approached 28 organizations about completing the survey. Of those, 24 (86%) responded with at least one completed survey. Organizations missing responses included an academic institution and ASO in the North Carolina PC network, a clinic in the Louisiana PC network, and a CBO in the San Francisco/Bay Area network. Between one and three completed surveys were received from all other organizations.
Density
All types of collaboration, as measured by density, increased between the 6 months prior to PC implementation and during the project (Table 4). In three of the four locations, overall density of collaboration increased during PC. The one location lacking an increase in density, Chicago, already had a density of 1 (all organizations were connected to one another) prior to PC (Figure 2). The sociograms representing the North Carolina, Louisiana, and San Francisco/Bay Area networks prior to PC have fewer lines connecting organizations than during PC. This increase in density is visually represented by the increased lines in the sociograms (Figures 1-4).
Network Density by Positive Charge Network, Collaboration Type, and Frequency.
Density by Type
Collaboration related to access, information exchange, technical assistance, and resource sharing also increased during this time period as measured by density (Table 4). For two networks, however, the density of resource sharing networks did not change—Chicago and North Carolina. Also, the density of access networks decreased slightly in Chicago, where all organizations reported collaborating to boost access for clients during the 6 months prior to PC (Figure 2).
Comparing networks for the four types of collaboration (information exchange, technical assistance, increasing access, and resource sharing), the density of information exchange networks was higher than any other type of collaboration. Network collaborations to increase access, such as referrals, most often had the second highest density followed by technical assistance. Across all four networks, resource sharing had the lowest density and did not increase as much as other types of collaborations during PC implementation.
Table 4 displays the densities of three types of at least monthly collaboration: access, information exchange, and technical assistance. When survey responses on these three types of collaboration were restricted to monthly or more often, all four networks had lower densities (Table 4).
Network Centralization
In this context, connections were initially driven by a lead grantee, which controlled access to funds. Despite this structure, network centralization decreased in Louisiana and San Francisco (Figure 3, Table 5). North Carolina’s network centralization did not change. Chicago’s network centralization did not change, but its network centralization was already 0, the lowest possible value (Figure 2, Table 5).
Network Centralization by Positive Change Network and Type of Collaboration.
Network Centralization by Type
With the exception of North Carolina’s PC network and access in San Francisco/Bay Area of California, network centralization was lower during PC compared to 6 months prior across all four types of collaboration: shared resources, access, information exchange, and technical assistance. These measures also increased in density; hence, the increase in density is distributed across the network and not associated with one or two organizations.
We also examined network centralization by collaboration frequency. Specifically, we examined network centralization for monthly or more frequent reported interactions in access, information exchange, and technical assistance (Table 5). Across all sites, network centralization was similar or higher.
Discussion
Changes in Density and Network Centralization
Across all PC locations, general and all four specific types of interorganizational collaboration increased over time as measured by density. Previous longitudinal studies have found similar results (Provan et al., 2003). In North Carolina, in particular, this statewide project created connections between geographically distant organizations that did not previously collaborate.
There were two networks where density increased and network centralization decreased. This finding means that the increased connections, evidenced by higher density, were not necessarily due to greater connection to the lead grantee (Figures 1 and 4, Table 5). Strong connections to the lead grantee were expected because of the grant structure; hence, this finding suggests that such a grant structure may drive connections among partner organizations in addition to connections to the lead organization.
Interpreting Collaboration by Type
In all networks, information exchange had the highest density of any collaboration type. Access was the second densest, followed by technical assistance and shared resources. This hierarchy was consistent in the 6 months prior to as well as during PC implementation. Previous studies have also observed this pattern where information is exchanged most often, but shared resources (staff, materials) are hardly ever exchanged (Kwait et al., 2001; Thomas et al., 2007). One reason for this hierarchy in collaboration types may be that while these organizations are tightly connected, informal collaborations that do not require sharing tangible resources such as staff and funds between organizations are more likely to form. In other words, those occurring on an “as needed” basis versus formal connections are simpler. Formal connections between organizations are possibly ineffective (Khosla et al., 2013). Another reason for this often observed hierarchy may be that client referrals and exchange of information, which are possible to exchange outside of formal agreements, reflect ties primarily at the service delivery level while formal agreements and joint programs may suggest more interactions at the administrative level. The latter require a long-term and administrative commitment, which may take more time, planning, resources, and effort (Gans & Horton, 1975).
However, the observed increase in density of one particular collaboration type, access, may lead to improved HIV care because increased interactions have the potential to enhance services for a population without building any new infrastructure or training more personnel (Jones et al., 2004; Parmigiani & Rivera-Santos, 2011). It seems promising that collaboration may boost service availability for PLWH given that previous studies have suggested that receiving ancillary services boosts engagement in care (Conviser & Pounds, 2002). A client may need to be referred to another organization for services. Strong network collaborations have the potential to alleviate this situation but may be or may not be preferable to colocated, “wraparound” services (Jain et al., 2016; Khosla et al., 2013).
A key difference between this analysis and previously conducted analyses on HIV organization interagency networks is that they covered all HIV organizations in an area (a “census”) rather than a network bounded by a project as this study does. As a result, perhaps this study found stronger ties since the relationship between organizations is not one of competing for resources but seeing the lead agency as a source of funds/resources.
Dichotomization by Frequency
Three of the collaboration types (access, technical assistance, and information exchange) were also examined by frequency, specifically dichotomized to monthly or more versus less than monthly collaboration. After dichotomization, density dropped for all three types of collaboration, suggesting areas for improved coordination within these networks. Previous studies of Baltimore-based organizations that serve PLWH have also found that collaboration density dropped by nearly half within their networks when dichotomized to monthly (Khosla, 2013, 2014).
Previous research points to possible reasons: Studies have found that network characteristics such as incentives, performance appraisals, and trust between organizations may drive collaboration (Daley, 2009; Greene et al., 2002). In North Carolina, a study of two HIV prevention networks found that HIV prevention agencies were mostly working in isolation and any collaboration was driven primarily by influential individuals (Thomas et al., 2007). These characteristics were not assessed in our study.
Strengths and Limitations
This study is descriptive and cannot measure whether PC resulted in increased interagency collaboration nor does it describe outcomes resulting from network collaboration. Also, this analysis can only speculate as to whether and how interagency collaboration affected the health outcomes of PC participants. Rather, the aim of this work was to describe the network, collaboration between organizations, and internal structural change. Due to the grant structure, organizations involved in this study did anticipate changes in connections across organizations. However, the extent to which the changes observed aligned with their plans and proposed activities could was not assessed. Another limitation of this study is that the individuals completing the surveys may not have been aware of all collaboration activities. If so, certain collaborations may not have been captured, which would result in slightly lower density measurements. Also, no data were available for 4 out of the 28 organizations surveyed; for these 4 organizations, any ties reported by other organizations were assumed to be reciprocated. This imputation may have increased reported density slightly. Furthermore, questions on collaboration 6 months prior to PC were asked retrospectively, possibly introducing recall bias. Social desirability bias is also possible as participants may have understood collaboration to be viewed favorably.
This study offers a wealth of data and similar insights gained from across four different U.S. geographies. Also, unlike previous studies that examined organically arising networks, these networks stem from a shared project. Yet findings from both types of studies are similar. Furthermore, many previous studies pointing toward similar insights took place at one time point in one location (Khosla et al., 2013; Kwait et al., 2001; Thomas et al., 2007). Despite this study including multiple locations and two time points, the results strongly parallel those of previous works, suggesting that these results are highly generalizable.
Conclusion
The results from this study suggest that similar grants can boost collaboration across organizations. It also provides insights to policy makers about how such a network may function and what to expect when establishing this type of a network. One direction for further exploration is the extent to which the availability of increased funds through a single, central organization is an effective driver of increased collaboration. Other directions include the extent to which collaboration is sustained after the program has ended and what effect interorganizational networks have on client health outcomes. In addition, if integrating a network of organizations can increase the quality of services, what is the ideal configuration to maximize service quality?
Overall, this network analysis provides a picture of some early, innovative HIV retention in care programs that aim to serve a highly underserved U.S. population. While it is a descriptive analysis, it covers a wide geography and echoes findings of past research despite taking place in the context of a grant. It also elucidates types of network interactions that most easily occur (e.g., referrals and information exchange) and those needing greater support, such as administrative level functions (e.g., technical assistance and resource sharing).
Footnotes
Authors’ Note
Johns Hopkins Bloomberg School of Public Health had a relationship only with AIDS United (not with Bristol–Myers Squibb).
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This evaluation project was supported by a grant from AIDS United to Johns Hopkins Bloomberg School of Public Health. The overall Positive Charge Project was supported by a grant from Bristol–Myers Squibb to AIDS United. During this time, K. Jain was supported by the National Institute of Allergy and Infectious Diseases (T32 A1050056-12).