Interdisciplinary communication is instrumental in optimizing care delivery and mitigating the risk of adverse events during the course of a patient’s care. Using social network analysis of a text messaging platform, we characterize text message-based communication relationships between various roles within the healthcare provider team over a 12-month period in an inpatient quaternary care setting. We demonstrate quantitatively that the frontline provider role exerts significant influence over network information flow, as well as a greater degree of influence in conducting the care of patients ultimately experiencing a clinical deterioration during their hospital encounter, relative to frontline providers engaged in the care of patients not experiencing a clinical deterioration. We also characterize networks relative to the timing of a clinical deterioration, noting that while individual participant influence may decrease, there are concomitant increases in network participation in the hours prior to a deterioration.

Timely and effective communication is a vital element to delivering care safely, with communication failures cited by the Joint Commission as a leading root cause of adverse events. [23] The final common pathway to mitigate the trajectory of the clinically deteriorating patient, whether recognized through derangements in physiologic monitoring, alterations in an early warning system, or simple pattern recognition is an escalation in care mediated by an attempt to mobilize resources through broadcasting concerns to available resources. While the use of effective tools including SBAR (situation, background, assessment, recommendation) and critical language have shown promise in facilitating verbal communication between healthcare team members in a highly complex medical environment, relatively less is understood concerning the implications of using text-based means of communication. [3] Indeed it has been posited that with this growth in the use of text message communication come vulnerabilities leading to alert fatigue, inappropriate replacement of critical verbal conversations with text-based communication, and misinterpretation of messaging due to lack of context. [24] One analysis by Hagedorn and colleagues of text messaging patterns between healthcare providers within a children’s hospital outlined a relatively dense and connected central network, comprised of a relatively small number of highly-connected network participants as well as a large number of network participants with relatively few connections. [25] We report here similar findings, but leveraging available message metadata enabled a granular understanding of text communication behaviors between providers related to the care of specific patients over a 12-month time period.

Among measures of centrality examined, eigenvector centrality was of particular interest as this measure accounts not only for the number of connections a node has, but for the connectedness of a node’s connections; therefore while any two nodes may demonstrate a comparable number of connections, a node connected to nodes of greater influence would also relatively higher measures of eigenvector centrality. [19] With higher measures of eigenvector centrality among frontline providers relative to their nursing and physician counterparts, we quantitatively demonstrated that the frontline provider role exerted significantly greater influence upon information flow through the secure text messaging network. Consistent with an inpatient clinical setting, we also demonstrate objectively that the frontline provider role more effectively serves as an intermediary directing the transit of information between other nodes (betweenness centrality) and greater efficiency in the dissemination of information across an entire network (closeness centrality) relative to both nurses and physicians. Interestingly, we also demonstrated relatively higher measures of eigenvector, betweenness, and closeness centrality among pharmacists in this network; indeed the role of the clinical pharmacist in the inpatient setting has been independently associated with significant reductions in resource utilization and mortality, along with improved medication safety practices in the inpatient setting. [26, 27] Their relative importance in this network may suggest an organizational prioritization of this role in the inpatient setting, or potentially the pharmacist’s preferred reliance upon secure text messaging relative to other communication modalities.

In addition to describing overall network behavior, the linkage of message metadata with hospital encounter features enabled the creation of individual, encounter-specific subnetworks. By creating these individual encounter-specific subnetworks, and further classifying these networks as related to the care of a patient ultimately experiencing a clinical deterioration, among frontline providers we demonstrate significantly higher eigenvector centrality values and a greater degree of influence in networks associated with the care of the clinically deteriorating patient, relative to frontline providers caring for patients in deterioration-free subnetworks. While frontline providers in these deterioration networks demonstrate greater connectedness to other influential nodes within a deterioration network, also demonstrated were significantly lower measures of closeness centrality in these same networks. These findings suggest that while frontline providers caring for the deteriorating patient may be members of a tightly-connected and highly influential group within a network dedicated to the care of the deteriorating patient, they also tend to be less centrally-located and further removed from other groups within that same network. Indeed, lower closeness centrality was observed in virtually all other roles caring for the clinically deteriorating patient, a finding potentially explained by greater multidisciplinary communication demands and resultant longer, more indirect communication paths. Betweenness centrality values were also zero for many nodes within the encounter-specific networks. As betweenness centrality quantifies the frequency with which a node lies on the shortest path between two other nodes, a betweenness centrality value of zero is consistent with a node that does not serve as an intermediary in any shortest path. Given the median betweenness centrality was 0 for all roles in both deterioration and deterioration-free networks, these findings suggest that the encounter-specific networks were relatively fragmented, with many nodes existing in isolated pairs or small clusters lacking intermediary connections.

As social networks are dynamic and subject to change with respect to both numbers, influence, and connectedness of participants over time, these changes may impact the ease by which information may traverse a network. Given the timing of messages was a known factor among patients experiencing a clinical deterioration in this study, we were able to quantitatively demonstrate changes in an individual’s subnetwork of healthcare providers as time approached a clinical deterioration. While we demonstrate an increase in both the number of participants (nodes) and the number of connections (edges) within a network as time approached a deterioration, we also demonstrated decreases in clustering coefficient after the first twelve hour epoch that subsequently increased in the twelve hours preceding a deterioration; while these initial decreases in clustering coefficient may represent a joining of disconnected nodes or a tendency toward one-to-many broadcast communications to disconnected recipients, subsequent increases in clustering coefficient approaching Time 0 in turn may reflect more sustained engagement of participants within network subgroups. We also demonstrated concomitant decreases in an individual node’s influence on subnetwork behavior (eigenvector centrality) over the same time window. As eigenvector centrality measures the influence of a node based upon its connections to other influential nodes within the network, decreases in network influence over time may be secondary to participants (appropriately) transitioning to alternative, synchronous forms of communication as a patient’s clinical status declines; future comprehensive studies incorporating the analysis of alternative means of communication between healthcare team members will be an important step in more fully characterizing interdisciplinary communication practices during the care of the clinically deteriorating patient.

This study was subject to several limitations. Text messaging is by no means the sole means of promoting interdisciplinary clinical communication, and in the setting of a clinically deteriorating patient requiring timely interventions, the importance of closed-loop, synchronous communication including telephone-based or “face to face” conversations escalate. Within this series alone, 47% of encounters were linked to text message exchanges; while a proportion of inpatient admissions may not have required interdisciplinary communication beyond face to face or telephone conversations during the course of care, identification of messages related to a clinical deterioration required end users to have actively appended a unique patient identifier to the message(s) thus serving as a barrier to complete capture of the message corpus. Furthermore, as investigators were not able to analyze the work schedules of healthcare team participants, this study design did not permit an understanding of network participation relative to their total time caring for patients in the inpatient setting. While this effort focused upon message metadata, unstructured message content was not analyzed as a part of this investigation; future efforts incorporating message content through natural language processing methods holds promise in identifying any objective relationships between changes in communication behaviors and a patient’s clinical state that can be leveraged to mitigate the risk of deterioration. [28]