Services targeting social determinants of health—such as income support, housing, and nutrition—have been shown to improve health outcomes and reduce health care costs for older adults.1–4 As health policy objectives increasingly prioritize cost containment, payers and providers have been considering opportunities to address social determinants within the health care system as a means of improving effectiveness and efficiency.5–7 Recent evidence has shown clinical benefits from screening for social needs and referring patients to social services.4 Models such as Accountable Health Communities for Medicare and Medicaid patients8 as well as state9,10 and local11 initiatives are further testing strategies for addressing social determinants in the context of clinical care. It has been recognized that sustained improvement in outcomes for people with complex health and social needs will likely require cross-sectoral partnerships at the community level to coordinate the diverse organizations that provide health care and social services.12
Research on public health systems suggests that patterns of collaboration among the distinct organizations providing health care, public health, and social services within a community are related to population health,13 and emerging evidence suggests that cross-sector collaboration could improve health for older adults.12,14,15 Enthusiasm for collaborative models may be tempered, however, by other recent studies showing limited impacts from collaborative community-based health programs on average,16 and lagging progress by multisector partnerships intended to improve regional health.17 In this context, more detailed information is needed to illuminate which features of collaborative networks are associated with potentially avoidable health care use and spending, to highlight potentially cost effective investments.
A growing body of literature identifies dimensions on which cross-sector collaborative networks vary, including the types of organizations involved,18,19 what they do together, the density of interconnections among network participants,20 and which organizations act as central brokers.21 Despite this relatively advanced understanding of the forms that networks might take, multiple systematic reviews22,23 have noted a gap in evidence linking properties of collaborative networks to outcomes. Understanding the role of health care organizations is particularly important, as models for coordinating health care with preventative attention to social determinants often envision a prominent role for health care providers and financing from health care payers.8,24
To address this evidence gap, we sought to identify features of collaborative networks of health care and social services organizations that are associated with avoidable health care use and spending for Medicare beneficiaries. We identified a diverse sample of 20 US communities, 12 communities with low levels of avoidable health care use and spending (high performers) and 8 communities with high levels of avoidable health care use and spending (low performers). We contrasted features of collaboration networks in high versus low-performing communities.
Study Design and Sample
We used network analysis to characterize collaborative ties among health care and social services agencies in 20 communities with consistently high (n=12) or low (n=8) performance across 3 outcomes expected to be sensitive to coordination among health care and social services: (1) hospitalizations for ambulatory care sensitive conditions; (2) risk-standardized hospital readmission rates; and (3) Medicare spending per beneficiary.25 Communities were demarcated by Hospital Service Areas (HSAs), geographical units in which a population generally receives most of its hospital care from the same hospital or set of hospitals.25 HSAs are similar in scale to counties, although they can follow different boundaries. Although our study was not focused exclusively on hospitals, we chose to use HSAs as geographic boundaries to facilitate clear identification of which hospitals should be approached for participation, since we anticipated that hospitals (as well as other health care providers) might be important partners in the networks.
To be classified as a higher performing community, an HSA had to perform in the top quartile on at least 2 of 3 indicator outcomes. To be classified as a lower performing community, an HSA had to perform in the bottom quartile on at least 2 of 3 indicator outcomes. We focused our sampling frame on HSAs in which at least 75% of the population was living in areas defined by the Census as urbanized or urban clusters. We excluded HSAs in the extreme quartiles of acute care hospital bed density, as supply can influence utilization.26 We excluded HSAs with >3 hospitals, eliminating the largest HSAs, to ensure that we could reasonably identify most of the relevant health care and social service organizations. Eligible HSAs were therefore midsized cities and defined regions within large metro areas. We stratified the 42 eligible communities according to household income and drew a purposeful random sample of 20 HSAs (Table 1): 12 higher performing HSAs and 8 lower performing HSAs, split between above-median and below-median income. Of the 20 HSAs included in the current network study, 16 had been involved in a qualitative study previously reported.14 As is common for that methodology,27 we included more HSAs from the higher performing group.
We identified health care and social service organizations from each HSA through a 2-phase process including: (1) systematic web-based research to identify common categories of health care and social service organizations present across most communities [eg, hospitals, health centers, Area Agencies on Aging (AAA), United Way, senior centers, health departments, housing agencies, and others], and (2) snowball sampling via email and telephone conversations to identify additional organizations that served older adults in each community. Given the wide range of relevant social services provided by specialized agencies, we identified more individual social service organizations than health care organizations in most communities. Within each organization or relevant department, we invited the individual with the highest seniority (eg, executive director, director, or manager) to complete a survey on behalf of the organization. When a specific contact person could not be identified, the web-based survey was emailed to a general organizational email address and/or a paper survey was mailed to the organization’s physical address.
To develop our survey instrument, we drew on prior organizational network surveys20,28 as well as previous qualitative research on collaborative relationships among health care and social service organizations.14 Employing a roster-based approach to elicit network ties,29,30 the survey presented respondents with a matrix listing all health care and social service organizations we had identified in their HSAs (average of 14.7 organizations per HSA), and asked the respondents to indicate how their organizations collaborated with others over the prior 12 months. Respondents could identify 7 types of collaborative ties: any collaboration, client referrals, sharing information about clients, cosponsoring activities (such as programs or advocacy), financial relationships, community needs assessments, and other. We tested the instrument in cognitive interviews31 with 8 leaders of health care and social service organizations in New Haven, CT (not a study HSA) and revised the survey accordingly. A blinded version of the network survey is provided in the Appendix (Supplemental Digital Content 1, http://links.lww.com/MLR/B753).
The survey was administered as a web-based survey from June 2017 to October 2017. Nonrespondents were contacted via email, phone, and postal mail, and were provided opportunities to complete the survey in paper form.
The survey data on interorganizational ties was used to construct network measures of cohesion and centrality using UCINET version 6.636.32 To describe the properties of each HSA’s network of health care and social service organizations as a whole, we calculated the following measures: (1) network density, the number of ties in the network as a proportion of all possible ties; (2) compactness, the degree to which nodes can reach each other through direct or indirect ties; and (3) centralization, the degree to which connections in a network are centralized around one or more nodes. We calculated whole-network measures for each specific type of collaborative tie reported in the survey (subnetworks). To incorporate information on strength of ties, we also calculated whole-network measures that were weighted according to the number of different types of ties reported between partners (multiplex networks). In addition to these whole-network measures, we calculated three organization-level (node-level) centrality measures: (1) betweenness centrality (the extent to which a given node falls along the shortest path between 2 other nodes), indicating a brokering or gatekeeping role; (2) in-degree centrality (ties received), indicating popularity or prestige; and (3) out-degree centrality (ties sent), indicating a tendency toward outreach. The centrality measures were normalized to permit comparison across networks. To visually explore patterns in the network data, we used Onodo (onodo.org) to create network maps called sociograms, in which the organizations are depicted as points and the relationship ties (eg, client referrals, sharing information, cosponsoring projects, etc.) are depicted as lines connecting the points.
We used t tests to compare the means of whole-network statistics (density, compactness, centralization) for HSAs in the high-performing (n=12) versus low-performing (n=8) groups. For comparisons of node-level normalized centrality measures, which are clustered within HSAs, we used single predictor regression models that permitted us to use clustered SEs.33 In these regression models, the normalized centrality measures served as our dependent variables, and a binary variable indicating membership in a high or low-performing HSA served as our independent variable. Analyses were conducted using SAS (version 9.4; SAS Institute, Cary, NC). As survey response rates varied across HSAs, in sensitivity analyses we reran all analyses excluding the 4 HSAs with a response rate of ≤50%. For node-level analyses of betweenness centrality and out-degree centrality, we only included organizations that responded to the survey, as these measures depend on outgoing ties nominated by the respondent. For analyses of in-degree centrality we included all organizations that were listed on the survey instrument, as this measure depends only on incoming ties nominated by other organizations in the network.
A total of 189 of 294 organizations completed the survey, for an overall response rate of 64%. An average of 9.5 organizations responded per HSA. Most participants (n=165) completed the survey online; 22 completed a paper survey, and 2 completed the survey via phone. Across HSAs, response rates ranged from 45% to 94%. The average response rate in high-performing HSAs (69%) was greater than in low-performing HSAs (57%) (P=0.03), and social service organizations had a higher response rate (74%) than health care organizations (49%) (P<0.01). Hospitals had nearly the same response rate (50%) as health care organizations overall.
Across our sample, 30% of respondents represented health care organizations while 70% represented social services agencies (Table 2); the proportions of respondents representing health care and social services were not significantly different between high and low-performing HSAs (χ2P=0.55). The specific types of organizations most frequently represented in our sample were social services and elder services organizations that provided multiple types of services (eg, United Way, community services, councils on aging).
On average, organizations reported having collaborative ties with 8.4 organizations in their networks. The most common type of collaborative relationship was client referral (mean: 6.2 ties), followed by cosponsoring activities (eg, projects, advocacy) (mean: 3.9 ties), and sharing information about clients (mean: 3.7 ties). Working with partners in community needs assessments (mean: 2.2 ties) and financial relationships/contracts (mean: 2.1) were less common. Respondents reported a mean of 1.1 ties for “other” activities.
Examining a multiplex organizational network of 6 types of collaborative ties, we found that high-performing communities had organizational networks that were on average more highly interconnected than low-performing communities. Compactness in high-performing HSAs was 0.45, compared with 0.36 in low-performing HSAs (P=0.10) (Table 3). Density was also somewhat higher in high-performing HSAs (0.32) than low-performing HSAs (0.26) (P=0.17). The networks in high-performing HSAs were less centralized than low-performing HSAs (degree centralization of 0.60 in high-performing communities versus 0.66 in low-performing communities) (P=0.29).
Comparing the subnetworks representing specific types of collaborative ties measured in our survey, we found that subnetworks of cosponsoring activities such as projects or advocacy demonstrated the most pronounced differences between high and low-performing HSAs. Networks of cosponsoring activities in high-performing HSAs exhibited greater density (P=0.06) compactness (P=0.06) and centralization (P=0.05) compared with low-performing HSAs, results that attained or approached significance even with the small sample size of 20 HSAs. Subnetworks for the other types of ties were generally denser and more compact in high-performing HSAs relative to low-performing HSAs, although these differences did not reach statistical significance.
Positions of Specific Types of Organizations
Health care organizations were significantly more integrated into collaborative networks of health care and social service agencies in high-performing networks, whereas health care organizations occupied more peripheral positions in low-performing networks. Health care organizations (eg, hospitals, outpatient providers, home health) occupied positions of significantly greater betweenness centrality in high-performing HSAs (mean centrality: 0.04) as compared to low-performing HSAs (mean centrality: 0.01; P<0.01) (Appendix, Table 1, Supplemental Digital Content 1, http://links.lww.com/MLR/B753, to access the Appendix, click on the Appendix link in the box to the right of the article online). The same, significant associations held for in-degree centrality and out-degree centrality. Health care organizations’ ties to social service organizations (rather than other health care organizations) primarily drove this difference. Health care organizations in high-performing HSAs had ties with an average of 4.88 social service organizations, while health care organizations in low-performing HSAs had ties with an average of 3.10 social service organizations (P=0.07 for difference). Health care organizations in high-performing HSAs had slightly more ties with other health care organizations than health care organizations in low-performing HSAs (average of 2.88 vs. 2.62 ties) but this difference did not approach statistical significance (P=0.71). The relative positions of health care and social service organizations in 2 high and 2 low-performing networks are shown in Figure 1. These 4 HSAs were selected for visualization due to their high survey response rates.
One specific type of social service agency, the AAA, exhibited significantly higher betweenness centrality than any other type of organization (Fig. 2). The mean betweenness centrality of AAAs (based on having any type of collaborative tie) was 0.12, double the value of the next highest category of Adult Protective Services (P=0.05) and significantly higher than any other of the 11 types of health care and social service organizations that we examined (P<0.01). This result held across both high and low-performing HSAs, reflecting the consistently central positions of AAA in the networks of health care and social service organizations serving older adults across our study HSAs.
In sensitivity analyses to assess the potential impact of low response rates on our results, we reran analyses without 4 HSAs that had response rates of ≤50% (3 of these HSAs were high performing and 1 was low performing). In an analysis of the remaining 16 HSAs, the direction and significance of the results were largely unchanged. However, the precision of estimates for the whole-network measures increased, making the comparisons presented above in Table 3 more statistically significant. The high-performing HSAs had greater density (P=0.01), greater compactness (P<0.01), and lower centralization (P=0.05) relative to low-performing HSAs.
The integration of health care and social services is increasingly recognized as an important way to promote healthy communities. Our results indicate that more cohesive interorganizational networks of health care and social service organizations are associated with lower levels of potentially avoidable health care use and spending among Medicare beneficiaries. This finding suggests that interorganizational relationships may strengthen services for older adults in a community in ways that help prevent avoidable health care use and spending. One specific type of interorganizational collaboration measured in our study, cosponsoring projects, was more closely associated with high performance than other collaborative ties. Examples of cosponsoring projects documented in prior, qualitative research14 include joint projects to meet the needs of individuals frequently seen in hospital emergency departments, community fall prevention programs, and joint advocacy for legislative change. Cosponsoring projects involves a high degree of mutual commitment, as organizations must agree on a priority project and invest resources together. In addition to practical benefits, cosponsorship ties may indicate that an organizational network embodies more collaborative experience and history, enabling the network to support more effective collaboration across a range of activities. Weiner and colleagues34 observed that effectiveness of community health coalitions “depends not only on promoting interorganizational collaboration on specific community health activities but also (and perhaps more important) on encouraging participating organizations to change their priorities, practices, and interrelationships.” Cosponsorship ties may signal that organizations are willing to change their priorities and practices to achieve results that will make the system of organizations more effective as a whole, which is consistent with prior qualitative work.14 Subnetworks for other types of ties, including client referrals, sharing information, and financial contracts, also demonstrated higher mean density and compactness in high-performing HSAs compared with low-performing HSAs, although the differences did not reach statistical significance, potentially due to our small sample size of 20 HSAs. These differences could still be important, however, and may signal value of lower commitment interactions such as client referrals or information sharing. To put the networks documented in our study in context, prior studies of health-related multisector networks have reported network densities in the range of 0.03 to 0.42,35,36 depending on the collaborative tie measured, which is close to the range observed in our study.
We found that health care organizations were significantly more engaged in high-performing networks than low-performing networks, as measured by centrality. Prior research suggests that hospital involvement in public health activities can increase the scope of public health services available to a local population,37 and that interorganizational collaboration on community health needs assessment enhances hospital investment in community health.38 Our results extend this earlier work to demonstrate that hospital engagement with multisector networks is associated with reduced levels of avoidable health care use and spending, measures of key importance to health care systems that face increasing responsibility for the costs of care provided.
At the same time, health care organizations rarely occupied the most central brokerage positions in their networks; this role was most often assumed by the AAA, which is consistent with earlier findings.39 The central position of AAAs is in line with the mission of these agencies, established under amendments to the Older Americans Act in 1973, which is to coordinate services for older adults across most communities in the US. Consequently, as policy makers and health care managers engage in efforts to foster cross-sector partnerships, the AAAs could be leveraged as brokers.
Our results should be interpreted in light of several limitations. First, health care use and spending in the study HSAs was likely influenced by a combination of factors, not solely by interorganizational collaboration. Still, collaborative networks did vary significantly between the high-performing and low-performing groups, suggesting that the interorganizational collaboration is involved in performance. Second, our study measured interorganizational relationships, but did not collect data on the capabilities of individual organizations (eg, service quality or managerial capacity). Individual organizations that are better managed and/or provide better quality services might also be better at forming collaborative relationships with other organizations in their communities, or have extra resources made available from efficient operation to do so. Third, we sampled only 20 communities so our results might not generalize to all settings, and the small sample size may have prevented identification of some meaningful distinctions. However, our sample was diverse in terms of geography and economic context, and the identification of some relationships that attained statistical significance despite the small sample attests to the strength of the associations. Finally, survey nonresponse may have influenced our network measures. Our sensitivity analysis excluding the lowest response rate HSAs produced the same results as our main analysis with greater precision, however, suggesting that low response rates tended to bias our results toward the null.
Future research to build upon our findings using larger, more generalizable samples could focus on testing the impact of partnerships by specific types of organizations. The time and resources required to identify and collect data from all organizations involved in caring for older adults in a given community limits the scale of a network study of the type we report here, but larger scale studies could measure partnerships reported by certain key organizations in the community, such as AAA, hospitals, or health care practices. National-scale data on partnerships by such key organizations could be linked with health outcomes.
In conclusion, our results suggest that greater interorganizational collaboration among health care and social services agencies could contribute to lower rates of preventable health care use and spending. Active cross-sectoral collaboration by health care organizations appears to be particularly important to variation on these measures; health care organizations facing risk-based reimbursement models may find collaboration an appealing area for investment. Productive partnerships are likely to require substantial commitment though, requiring organizations to align priorities and resources. Lower commitment modes of partnership, such as referring clients or sharing information, did not distinguish high-performing networks. Efforts to foster effective partnerships could build on established brokers, such as AAA, which are already positioned centrally in most networks.
1. Taylor LA, Tan AX, Coyle CE, et al. Leveraging the social determinants of health
: what works? PLOS ONE. 2016;11:e0160217.
2. Shier G, Ginsburg M, Howell J, et al. Strong social support services, such as transportation and help for caregivers, can lead to lower health care use and costs. Health Aff. 2013;32:544–551.
3. Gusmano MK, Rodwin VG, Weisz D. Medicare beneficiaries living in housing with supportive services experienced lower hospital use than others. Health Aff. 2018;37:1562–1569.
4. Berkowitz SA, Hulberg AC, Standish S, et al. Addressing unmet basic resource needs as part of chronic cardiometabolic disease management. JAMA Intern Med. 2017;177:244–252.
5. Daniel H, Bornstein SS, Kane GC. Addressing social determinants to improve patient care and promote health equity: an American college of physicians position paper. Ann Intern Med. 2018;168:577.
6. Joynt Maddox KE. Financial incentives and vulnerable populations—will alternative payment models help or hurt? N Engl J Med. 2018;378:977–979.
7. Gottlieb L, Sandel M, Adler NE. Collecting and applying data on social determinants of health
in health care settings. JAMA Intern Med. 2013;173:1017–1020.
8. Alley DE, Asomugha CN, Conway PH, et al. Accountable Health Communities—addressing social needs through Medicare and Medicaid. N Engl J Med. 2016;374:8–11.
10. Gaskin DJ, Vazin R, McCleary R, et al. The Maryland health enterprise zone initiative reduced hospital cost and utilization in underserved communities. Health Aff. 2018;37:1546–1554.
11. Murphy SME, Hough DE, Sylvia ML, et al. Going beyond clinical care to reduce health care spending: findings from the J-CHiP community-based population health management program evaluation. Med Care. 2018;56:603.
12. Towe VL, Leviton L, Chandra A, et al. Cross-sector collaborations and partnerships: essential ingredients to help shape health and well-being. Health Aff. 2016;35:1964–1969.
13. Mays GP, Mamaril CB, Timsina LR. Preventable death rates fell where communities expanded population health activities through multisector networks. Health Aff. 2016;35:2005–2013.
14. Brewster AL, Brault MA, Tan AX, et al. Patterns of collaboration among health care and social services providers in communities with lower health care utilization and costs. Health Serv Res. 2018;53(S1):2892–2909.
15. Brewster AL, Kunkel S, Straker J, et al. Cross-sectoral partnerships by area agencies on Aging
: associations with health care use and spending. Health Affairs. 2018;37:15–21.
16. Fry CE, Nikpay SS, Leslie E, et al. Evaluating community-based health improvement programs. Health Aff. 2018;37:22–29.
17. Siegel B, Erickson J, Milstein B, et al. Multisector partnerships need further development to fulfill aspirations for transforming regional health and well-being. Health Aff. 2018;37:30–37.
18. Bevc CA, Retrum JH, Varda DM. New perspectives on the “Silo Effect”: initial comparisons of network structures across public health collaboratives. Am J Public Health. 2015;105(S2):S230–S235.
19. Hogg RA, Varda D. Insights into collaborative networks of nonprofit, private, and public organizations that address complex health issues. Health Aff. 2016;35:2014–2019.
20. Retrum JH, Chapman CL, Varda DM. Implications of network structure on public health collaboratives. Health Educ Behav. 2013;40(suppl 1):13S–23S.
22. Hearld LR, Bleser WK, Alexander JA, et al. A systematic review of the literature on the sustainability of community health collaboratives. Med Care Res Rev. 2016;73:127–181.
23. Roussos ST, Fawcett SB. A review of collaborative partnerships as a strategy for improving community health. Ann Rev Public Health. 2000;21:369–402.
26. Delamater PL, Messina JP, Grady SC, et al. Do more hospital beds lead to higher hospitalization rates? a spatial examination of Roemer’s Law. PLoS One. 2013;8:e54900.
27. Curry LA, Spatz E, Cherlin E, et al. What distinguishes top-performing hospitals in acute myocardial infarction mortality rates? Ann Intern Med. 2011;154:384–390.
28. Provan KG, Milward HB. A preliminary theory of interorganizational network effectiveness: a comparative study of four community mental health systems. Admin Sci Q. 1995;40:1–33.
29. Borgatti SP, Foster PC. The network paradigm in organizational research: a review and typology. J Manage. 2003;29:991–1013.
30. Wasserman S, Faust K. Social Network Analysis
: Methods and Applications, 1st ed. Cambridge, NY: Cambridge University Press; 1994:857.
31. Beatty PC, Willis GB. Research synthesis: the practice of cognitive interviewing. Public Opin Q. 2007;71:287–311.
32. Borgatti SP, Everett MG, Freeman LC. Ucinet 6 for Windows: Software for Social Network Analysis
. Harvard, MA: Analytic Technologies; 2002.
33. White H. Maximum likelihood estimation of misspecified models. Econometrica. 1982;50:1–25.
34. Weiner BJ, Alexander JA, Shortell SM. Management and governance processes in community health coalitions: a procedural justice perspective. Health Educ Behav. 2002;29:737–754.
35. Provan KG, Nakama L, Veazie MA, et al. Building community capacity around chronic disease services through a collaborative interorganizational network. Health Educ Behav. 2003;30:646–662.
36. Fried BJ, Johnsen MC, Starrett BE, et al. An empirical assessment of rural community support networks for individuals with severe mental disorders. Community Ment Health J. 1998;34:39–56.
37. Hogg RA, Mays GP, Mamaril CB. Hospital contributions to the delivery of public health activities in US metropolitan areas: national and longitudinal trends. Am J Public Health. 2015;105:1646–1652.
38. Carlton EL, Singh SR. Joint community health needs assessments as a path for coordinating community-wide health improvement efforts between hospitals and local health departments. Am J Public Health. 2018;108:676–682.
39. Wholey DR, Gregg W, Moscovice I. Public health systems: a social networks perspective. Health Serv Res. 2009;44(pt 2):1842–1862.