Characterizing Nursing Home Care Team Communication via Text Messaging: A Social Network Analysis

Abstract

Purpose:

Communication breakdowns among healthcare teams contribute substantially to preventable adverse events in nursing homes. Although text messaging platforms are increasingly used to support care coordination, little is known about how these technologies are embedded within care team communication networks. The purpose of this study was to characterize patterns of text-message communication among nursing home care teams and to examine how network structures reflect underlying social system dynamics.

Methods:

Social network analysis was applied to text messages (n=5,092) linked to (n=585) nursing home-to-hospital resident transfers over 5 years (2015–2020). Message metadata and content were used to construct communication networks and to calculate network measures, including density, centralization (in-degree and out-degree), and reciprocity. Networks were qualitatively classified into communication models based on shared structural characteristics and interpreted using Rogers’ Diffusion of Innovations theory. All analysis and visualization was conducted using Python (3.13.9).

Results:

Three distinct communication models were identified. The Integrated model exhibited high message volume, high connectivity, and low hierarchy, consistent with mature adoption of text messaging as a general collaboration tool. The Hub-and-Spoke model showed moderate message volume with centralized information flow, reflecting protocol-driven, hierarchical communication. The Siloed model demonstrated high message volume but low role diversity, indicating niche use within specific professional roles rather than facility-wide coordination. Across models, individuals occupying central or bridging positions appeared to function as informal opinion leaders, influencing communication through frequent interaction rather than formal authority.

Conclusions:

The structure and quality of communication networks shape how information flows, influence is exercised through electronic messaging, and innovations are integrated into care processes. Social network analysis offers a rigorous approach for evaluating implementation and guiding strategies to support effective, team-based communication in nursing home settings. Future work should examine whether increasingly integrated communication networks are associated with improved resident outcomes.

Introduction

Deciding if and when a nursing home (NH) resident should be transferred to the hospital is a high-stakes, time-sensitive decision. NH staff are often faced with making this critical decision without a physician or other care provider physically present to offer consultation and guidance. Avoidable hospital transfers of NH residents result in unnecessary expenditures[1], physical and emotional distress[2, 3], and are often misaligned with resident values and preferences[4]. Reducing avoidable NH-to-hospital transfers requires effective communication between diverse team members, including physicians, nurses, social workers, and family caregivers, to balance potential benefits with risk, with careful consideration of “what matters most” to the resident[5]. Direct care providers (i.e., nurses and nursing assistants at the bedside) must combine data from multiple sources: subtle changes in physical assessment, history, patient and family preferences and then decide how and with whom to share this information. Communication breakdowns among healthcare providers contribute substantially to preventable adverse events[6]. Understanding how care teams are socially connected and how information flows between individuals is essential to reducing poor outcomes, including avoidable hospital transfers[7, 8].

Text messaging has emerged as a practical, low-cost communication tool in healthcare settings, including NHs. Staff increasingly rely on secure messaging platforms to quickly share resident updates, coordinate care, and seek guidance from clinical leaders or off-site providers[9]. Unlike traditional communication channels that can be time-consuming or difficult to access (e.g., phone, fax), text messaging enables rapid information exchange and real-time decision making. The immediacy of text messaging is especially valuable in NHs where timely responses to changes in a resident’s condition can prevent poor outcomes including avoidable hospital transfers[10]. As adoption of digital communication accelerates, understanding how text-based interactions shape clinical workflows and team collaboration has become essential[11].

Rogers’ Diffusion of Innovation theory provides a valuable framework for understanding how new practices and knowledge spread within healthcare settings, including NHs[12]. According to Rodgers, adoption of innovative behaviors, including clinical communication strategies, is shaped by characteristics of individuals, the nature of the innovation, and the structure of the social system in which communication occurs. Central to the diffusion process is the role of interpersonal communication channels and opinion leaders who influence other decisions to adopt and sustain new practices[13]. In long-term care environments, bedside nurses, advanced practice registered nurses (APRN), and administrative leaders serve as key influencers who can accelerate, or hinder adoption of best practices related to early detection, timely intervention, and coordinated decision-making to avoid unnecessary transfers.

Social network analysis (SNA) is a research methodology used to examine the relationships and interactions among a group of individuals, organizations, or other entities, known as “nodes” or “actors”[14]. SNA has been used widely in healthcare to study organizational relationships within hospitals or care teams, knowledge sharing patterns within interprofessional teams, and to develop recommendations for improving patient care[15]. In addition, SNA can help identify areas for improvement in interprofessional care by identifying individuals who play key roles in facilitating or hindering decision-making and revealing gaps in communication[16, 17]. For example, a study using SNA found that teams with denser communication networks and more clustering demonstrated faster dissemination of information, suggesting that promoting interconnectedness within teams can enhance knowledge sharing and potentially improve care coordination[18].

SNA offers a rigorous empirical approach to operationalizing theoretical concepts by mapping real-world communication patterns among NH care teams. By analyzing text messages, this study will identify network characteristics such as centrality, density, and patterns of influence that align with Rogers’ constructs of opinion leadership, communication channels, and social system dynamics. Understanding which individuals occupy central positions, bridge communication gaps, or function as early champions of effective care coordination practices will provide insight into how innovations in clinical communication diffuse across the care team. The objectives of this study were to: (1) determine the feasibility of conducting analysis of social networks within NHs as care teams making decisions to transfer a resident to the hospital; and (2) describe and analyze the social network of NHs regarding text message communication centered on resident transfer events.

Methods

Study Design

This analysis was conducted using text messaging data collected from the Missouri Quality Initiative (MOQI), a 5-year (2016–2020) Centers for Medicare and Medicaid Services (CMS) demonstration project (1E1CMS331080) designed to reduce avoidable hospitalizations for long-stay NH residents[19]. NHs that participated in the MOQI project had access to a common set of resources including: 1) advanced practice registered nurses (APRN) embedded in each NH, 2) feedback reports on hospitalizations and transfers, 3) Interventions to Reduce Acute Care Transfers (INTERACT) tools[20], and 4) enhanced communication systems including secure text messaging. The text message dataset included message content, timestamps, senders, and recipients. The messaging application was accessible to health workers via desktop and mobile devices and supported features such as read receipts, response time tracking, the ability to add or remove participants from message threads, and multi-user accounts. NH were permitted to use the text messaging application for any clinical or operational communication purpose and were not restricted to transfer-related messaging.

We first determined the feasibility of SNA by calculating the total number of text messages sent within the resident transfer window (14 days prior to transfer) and the number of unique roles sending and receiving the messages by NH. NHs with sufficient message volume and variety of roles were selected for additional analysis. Examining the social networks within and between these NHs allowed us to gain a deep understanding of the social networks within the broader context of political, legal, and practice workflow factors that vary across facilities. NH names have been pseudonymized to protect confidentiality, and all analyses were conducted in accordance with the University of Missouri Institutional Review Board approval (2009109).

Measures

Feasibility Testing

Feasibility of SNA was assessed across NHs that participated in the MOQI project and used secure text messaging (n=12) using objective criteria to determine inclusion. Three thresholds were applied to ensure that the analyzed facilities possessed adequate data density (i.e., sufficient number of text messages exchanged) and structural complexity (i.e., at least two different people exchanging messages) to yield meaningful network analysis:

1. Volume: Total message count > 100 in the 14-day transfer window. This threshold ensured sufficient observational data to construct reliable directed graphs and compute stable network metrics.

2. Diversity: Presence of > 2 unique active roles (sender/receiver). A minimum of three distinct roles was required to permit analysis of interprofessional communication patterns, a core interest given the multidisciplinary nature of transfer decision-making.

3. Stratification: Non-zero variance in role participation. A variance of zero indicated complete dominance by a single role type, which would limit meaningful network analysis. Sender role variance was calculated as the statistical variance $\left(S^2\right)$ of messages across unique sender roles.

Although no universal “minimum viable” standard for social network analysis in secure messaging contexts was identified in the literature, the chosen criteria align with recommendations that networks contain at least moderate edge counts and node variance to avoid trivial or unstable results[21].

Construction of the SNA Matrix

For facilities meeting inclusion criteria, we constructed directed, weighted graphs $G=(V,E)$ where Nodes = $V$ showing the standardized care roles (e.g. “Director of Nursing”, “Advanced Practice Registered Nurse”, “Licensed Practical Nurse”) and Edges = $E$ showing the directed links weighted by the frequency of messages sent from one role to another.

To characterize the structure and dynamics of each network, we examined four properties. Density was calculated as the ratio of observed edges to the total number of possible edges, revealing how thoroughly connected the team was overall. A high density indicated a highly interconnected group where most roles communicated directly with one another. Because communication was directed, we distinguished between two forms of degree centralization. In-degree centralization measured the extent to which incoming messages concentrated on a few roles. Out-degree centralization, in contrast, measured how much outgoing communication concentrated in a few roles. Reciprocity captured the tendency for messages to be returned along the same path, reflecting the presence of genuine two-way dialogue rather than purely one-way reporting. Finally, intra-role volume described the percentage of all messages that stayed within the same role (e.g. nurse-to-nurse or physician-to-physician communication) highlighting the degree to which staff relied on peers at their own level rather than reaching across role boundaries. Together, these measures provided an understanding of how each NHs’ care team modelled communication behavior during transfer events. All analysis and visualization was conducted using Python (3.13.9).

Exploratory Analysis

We conducted additional temporal network analyses within the same 14-day pre-transfer frame. These exploratory analyses evaluated: (1) changes in communication structure as transfer time approached using time-to-transfer windows (<24h, 24–72h, >72h), and (2) structural differences across calendar periods (pre-2019 vs 2019–2020), including a COVID-era sensitivity split (pre-2019, 2019-February 2020, March-August 2020). For each stratum, we estimated density, in-degree centralization, out-degree centralization, hierarchy, reciprocity, and weighted clustering, plus normalized intensity metrics and permutation-based robustness checks.

Results

Feasibility Assessment

The feasibility assessment stratified the 12 facilities into our inclusion and exclusion criteria. Four facilities met the inclusion criteria while eight were excluded due to insufficient volume and lack of role diversity. See Figure 1. The four facilities that met inclusion criteria contributed 5,092 text messages linked to 585 unique transfer events during the observation period. Feasibility metrics for all participating NHs are shown in Table 1 demonstrating the high volume, role diversity, and variance of the included NHs compared to the excluded ones.

Figure 1:

Total Message volume (log scale) versus unique active roles. Red lines indicate the inclusion threshold. S1, N1, G1, and D1 appear in the upper-right quadrant, demonstrating sufficient data for SNA.

Table 1.

Feasibility metrics for all Nursing Homes

Facility	Total Messages (14d)	Number of Transfers with TM within 14d window	Avg TM/Transfer	Unique Roles	Sender Role Variance $\left(S^2\right)$	Status
S1	3670	189	19.42	14	3,410,179	Included
N1	552	71	7.77	9	54,535	Included
G1	480	57	8.42	6	87,855	Included
D1	390	57	6.84	3	51,487	Included
D2	41	15	2.73	2	220	Excluded
D3	40	9	4.44	1	0	Excluded
D5	26	4	6.50	1	0	Excluded
D4	25	2	12.50	1	0	Excluded
F1	16	5	3.20	6	16	Excluded
N2	12	2	6.00	2	12	Excluded
D6	8	1	8.00	1	0	Excluded
D7	3	1	3.00	1	0	Excluded

Nursing Home Resident Demographics

The facilities serve diverse populations of transferred residents varying by age, race, and Alzheimer’s Disease and Related Dementias (ADRD) compositions (see Table 2). Notably, G1 serves a predominantly black population (82.9%), compared to S1 and N1 who were predominantly White (>%80) during the study period. D1 has a mixed population with 72% White and 26.2% Black. Age differences were also observed with D1 serving older populations (mean age 76.5) compared to the younger cohorts in G1,N1, and S1 with mean age 70.2, 66.4, and 67.9 respectively. High proportion of female residents was also observed in this transfer sample for D1 (71%) and S1 (67.9%).

Table 2.

Demographic Characteristics of Nursing Homes (n=12).

NH	Included	Total Certified Beds	Total Transfers	Age Mean	Age SD	% Female	% American Indian/Alaska n Native	% Asian	% Black or African American	% Hispanic or Latino	% Other/Multiracial	% White	% ADRD Unknown	% ADRD Middle	% ADRD Early	% ADRD Late
S1	Y	189	212	67.9	12.3	67.9	0	0	0	0	0	97.2	73.6	15.6	7.5	1.9
N1	Y	282	149	66.4	15.6	51.7	0	0	16.8	0	0	83.2	61.1	8.7	18.1	9.4
G1	Y	120	117	70.2	12.6	44.4	0.9	0	82.9	0.9	0	15.4	88	8.5	0.9	2.6
D1	Y	321	107	76.5	12.5	71	0	0.9	26.2	0	0.9	72	67.3	21.5	9.3	0.9
D2	N	240	38	80	11.4	78.9	0	0	57.9	0	0	42.1	50	21.1	26.3	2.6
N2	N	120	20	62.6	11.7	55	5	0	5	0	0	90	80	0	15	0
D3	N	250	15	80.1	6.8	53.3	0	0	0	0	0	100	13.3	66.7	0	20
F1	N	150	13	65.7	12.5	53.8	0	0	0	0	0	100	61.5	15.4	7.7	15.4
D5	N	240	8	81.8	7.5	75	0	0	0	0	0	100	62.5	37.5	0	0
D6	N	190	6	80.5	21.6	100	0	0	0	0	0	100	66.7	16.7	16.7	0
D4	N	180	4	76	6.7	100	0	0	0	0	0	100	100	0	0	0
D7	N	152	3	70.3	15.4	66.7	0	0	66.7	0	0	33.3	66.7	33.3	0	0

SNA Network Diagrams

The four included facilities exhibited distinct social network topologies. In Figure 2, we display the communication structures of the four facilities. Node size corresponds to the message volume of the individual sender/receiver, and edge color and width represents the total messages flowing to and from the nodes. S1 (top-left) shows a dense, multi-hub structure. N1 (top-right) and G1 (bottom-left) show more distinct primary hubs. D1 (bottom-right) shows a sparse structure dominated by specific ties to APRN roles. The network volume of messages (edges) in the network graph is displayed as a total beside the NH title. Due to single messages being sent to multiple recipients, this number is higher than the total message volumes displayed in Table 1.

Figure 2.

Social Network Typologies from 4 Nursing Homes.

SNA Quantitative Measures

In Table 3, we show the quantitative network metrics for each NH. The network analysis for NH S1 revealed a highly dense, less centralized model suggesting a “flat” hierarchy where multiple roles communicate directly with each other frequently during transfer events. The low intra-role volume further demonstrated high inter-role collaboration. N1 and G1 exhibit moderate density but higher out-centralization. This structure (N1 and G1) indicates a communication network where specific roles act as central dispatchers of information. D1, while technically dense due to low node count (3), is unique in its 94.8% intra-role volume. This shows that nearly all communication during the transfer window occurs within a single role group with minimal interdisciplinary engagement through the text message platform.

Table 3.

Quantitative Social Network Metrics.

Facility	Nodes	Edges	Density	In-Centralization	Out-Centralization	Reciprocity	Intra-Role Vol (%)
S1	13	143	0.917	0.181	0.181	0.937	6.8%
N1	8	33	0.589	0.469	0.633	0.788	33.8%
G1	6	20	0.667	0.160	0.640	0.600	10.5%
D1	3	5	0.833	1.000	1.000	0.800	94.8%

Exploratory Results

Within the 14-day frame, window volumes were sufficient for pooled temporal SNA: <24h (n=2,136), 24–72h (n=1,029), and >72h (n=2,098), with stable node counts (18–19). Network density was similar in the two windows nearest transfer (0.602 and 0.605) and lower in >72h (0.512). Reciprocity remained high across windows but declined farther from transfer (0.893, 0.886, 0.834 for <24h, 24–72h, and >72h, respectively). Out-degree centralization was highest in >72h (0.515 vs 0.361 and 0.356), suggesting more concentrated outbound communication in the more distal pre-transfer period. Weighted clustering remained low and stable across windows (~0.034–0.038). Permutation-based z-scores supported non-random structure across all windows, particularly for reciprocity and centralization. For example, reciprocity z-scores were 6.96 (<24h), 5.87 (24–72h), and 6.80 (>72h). Per-case window coverage was limited: among 411 transfer cases with messages in the 14-day frame, only 49 had all three windows represented (25 with >=3 messages in each window), reinforcing that pooled/facility-level interpretation is more stable than within-case trajectory inference.

Primary 14-day period strata had comparable volume: pre-2019 (2,958 messages, 212 cases, 140 residents) and 2019–2020 (2,305 messages, 200 cases, 137 residents). Relative to pre-2019, the 2019–2020 network showed lower density (0.567 vs 0.657), lower reciprocity (0.856 vs 0.935), and greater centralization (in-degree 0.340 vs 0.239; out-degree 0.457 vs 0.363), with higher hierarchy (0.103 vs 0.050). Weighted clustering remained low in both periods (0.032–0.039). Permutation results again indicated meaningful departure from random mixing in both periods (e.g., reciprocity z=8.40 pre-2019; z=4.61 in 2019–2020). In COVID sensitivity analyses, all 3-way strata met feasibility thresholds by volume/case count: pre-2019 (n=2,958; 212 cases), 2019-February 2020 (n=1,850; 154 cases), and March-August 2020 (n=455; 47 cases). The March-August 2020 stratum demonstrated the lowest reciprocity (0.763) and highest hierarchy (0.191) among 3-way periods.

Discussion

This study used SNA of text message communication to characterize how NH care teams communicate around resident transfer decisions. We identified three distinct communication models: Integrated, Hub-and-Spoke, and Siloed. Together these models illustrate meaningful variation in how the same communication technology (i.e., text messaging), embedded within local social systems, shape information flow and influence care coordination practices. By empirically mapping communication patterns, this work demonstrates how SNA can operationalize core theoretical concepts including communication channels, opinion leadership, and social system dynamics in real-world clinical settings.

Communication Models as Expressions of Social System Dynamics

Viewed through a diffusion of innovations lens, the observed communication models represent distinct social system configurations that condition who communicates with whom, how influence is exercised, and how clinical communication technologies are ultimately used. The Integrated model (S1), characterized by high volume, high connectivity, and low hierarchy, most closely aligns with Rogers’ description of a well-integrated social system that facilitates rapid diffusion of innovations[13]. In this configuration, communication is distributed across roles, with multiple individuals occupying central or bridging positions. The density and redundancy found in this model are theorized to promote shared understanding, collective problem solving, and resilience to information loss. From a diffusion perspective, this model suggests a mature stage of technology adoption in which text messages function as a general-purpose collaboration channel rather than a task-specific reporting tool.

The presence of multiple highly connected actors may reflect the emergence of informal opinion leaders, that is, individuals who influence practice not through formal authority but through frequent interaction and trusted relationships. This interpretation aligns with prior research showing that interdisciplinary NH care teams, when supported by active leadership and positive information flow, are better able to engage staff in shared decision making and collaborative problem solving[22]. Other studies further reinforce this relationship, demonstrating that well-functioning teams with stable membership and clear coordination are associated with higher quality care whereas poorly supported or fragmented teams are less effective[23, 24]. Together these findings suggest that the structure and quality of communication networks rather than simply the presence of teams, shape how innovations, practices, and coordination efforts are adopted and sustained in long-term care settings.

In contrast, the Hub-and-Spoke model (N1 and G1) reflects a more centralized communication structure with messages primarily flowing through a small number of key individuals. This pattern is consistent with a hierarchical social system in which information is filtered upward through formal chains of command. Within Rogers’ framework, such systems may slow diffusion by limiting peer-to-peer exchange, even while maintaining efficiency for protocol-driven tasks. The central actors in these networks occupy positions of high betweenness and degree centrality, suggesting that they function as gatekeepers of information rather than the bridges between disconnected groups. While this configuration may support accountability and role clarity, it may also constrain processes increasingly recognized as critical for complex care coordination in NHs like lateral learning and shared decision-making[25].

The siloed model (D1) presents a different theoretical case: high message volume coupled with low role diversity and limited cross-role connectivity. This pattern suggests that the technology has been adopted within a narrow professional sub-group, mostly advanced practice registered nurses, rather than diffusing across the broader care team. From a diffusion perspective, this reflects partial or reinvention-driven adoption in which innovation is repurposed to replace existing tools (e.g., fax machines, or pagers) without fundamentally altering the social system. Although communication intensity is high, the lack of bridging ties may limit the spread of information and reduce opportunities for interdisciplinary coordination. This finding underscores that volume alone (i.e. the number of text messages exchanged) is insufficient to indicate meaningful adoption or clinical impact.

Opinion Leadership and Network Positions

Across models, SNA illuminates how influence is enacted through network position rather than formal role alone. Individuals occupying central or bridging positions, especially in the integrated model, are well positioned to act as opinion leaders who accelerate diffusion of effective communication practices. The nursing staff role, for example, connect otherwise disparate roles, enabling information to circulate more broadly and supporting alignment around shared goals. Conversely, in more centralized or siloed networks, opinion leadership may be concentrated or constrained, potentially reinforcing hierarchies rather than facilitating system-wide change. These findings extend prior work by demonstrating how opinion leadership can be empirically identified through communication behavior, rather than inferred from job titles.

Implications from Exploratory Analysis of Temporal Communication Network Dynamics

These temporal extensions add context to the primary cross-sectional network findings. Communication near transfer appears more distributed and reciprocal, whereas communication farther from transfer (>72h) appears more outbound-centralized, consistent with an earlier triage phase transitioning to broader bidirectional coordination closer to transfer decision points. Comparison of pre-2019 versus 2019–2020 suggests structural tightening of communication patterns during later years with lower density and reciprocity with higher centralization and hierarchy. This is directionally consistent with pandemic-era effects on communication workflows, including possible disruption during turbulent periods. The COVID-era sensitivity split further suggests a plausible breakpoint signal in March-August 2020, but this interval is smaller than earlier strata and should be interpreted cautiously. Overall, these exploratory analyses strengthen inference that communication structure around transfer decisions is temporally dynamic and context-sensitive, while also confirming that robust interpretation is strongest at pooled and facility-levels rather than per-case longitudinal trajectories.

Implications for Health Information Technology Implementation

Taken together, these findings suggest that the impact of clinical communication technologies depends not only on their availability but on how they are embedded within local communication networks. Implementation strategies that focus on individual adoption or message volume may overlook critical structural features such as connectivity and role diversity that shape how information flows and innovations diffuse. Interventions that intentionally cultivate cross-role communication, reduce unnecessary gatekeeping, and support emergent opinion leaders may be more likely to promote integrated and sustained use. For NHs in particular, where care coordination is challenged by staffing constraints and role fragmentation[26], understanding communication models offers a pathway to more targeted implementation and evaluation. SNA provides a rigorous, scalable method for assessing whether a technology is functioning as a collaborative infrastructure or merely digitizing existing silos.

Limitations and Future Directions

This study examined communication patterns within a limited number of facilities and findings may not generalize to all NHs. The analysis was limited to communication occurring on the text messaging platform during the 14-day transfer event window. Missing data in the network may represent face-to-face interactions or phone calls, not a lack of communication. Additionally, network structure alone cannot capture content, tone, or clinical appropriateness of communication. Future work should integrate SNA with qualitative analysis to better understand how network positions relate to decision-making, trust, and care outcomes. Longitudinal analyses could further elucidate how communication models evolve over time and whether shifts toward more integrated networks are associated with improved resident outcomes.

Conclusion

By mapping real-world text-message communication, this study demonstrates how SNA can translate diffusion theory into measurable network characteristics revealing how clinical communication technologies are adopted and used in practice. The identification of Integrated, Hub-and-Spoke, and Siloed models highlights that diffusion is not uniform, even within similar care settings. These insights advance both theory and practice showing how social structure shapes the diffusion and function of digital communication among NH care teams. The ways in which different actors interact throughout the decision-making process could identify key decision makers and influencers, enhance communication efficiency, facilitate multidisciplinary collaboration, understand social dynamics and bias, optimize workflow and resource allocation, support training and professional development, and improve patient-centered outcomes.

Highlights.

• Social network analysis of 5,092 text messages and 585 nursing home-to-hospital transfers.

• Three network typologies emerged: Integrated (highly connected), Hub-and-Spoke (centralized), and Siloed (role-specific)

• Informal opinion leaders, not formal authority, drove communication patterns in electronic messaging networks

• Network structure may influence care coordination, quality, and innovation adoption in nursing homes.