Contributions of social network analysis for the modeling of organizational resilience in health services

Abstract

In order to survive, socio-technical systems display resilient performance (RP) to cope with variabilities. RP is emergent, arising from interwoven networks that involve several types of interactions between people, technologies, and work organizations. This thesis explores the role of the networks of social interactions in RP, in the realm of health services. In particular, the thesis addresses the problem of how to measure the contribution of individual actors to RP, considering their centrality in social networks as a proxy of that contribution. Although previous studies pointed out the role of social interactions in the rise of RP, they did not provide tools nor a corresponding theoretical framework for the assessment of organizational resilience based on social interactions. In order to address this gap, the present thesis is structured into three articles. The first article proposes an approach for the identification of key players that support RP, based on a composite resilience score (RS) for each actor comprised of the three most common metrics used in network analysis at the individual level, namely in-degree, closeness, and betweenness, in addition to two non-network attributes of actors - availability and reliability. The RS might be calculated for each actor, in four networks related to the four abilities of resilient systems, namely monitor, anticipate, respond, and learn. A global RS for each might also be calculated, as the total of the RSs from the ability-based networks. The second article presents an approach to developing and interpreting multilayer networks in light of resilience engineering. Layers correspond to the four abilities of resilient systems. Two multilayer networks were developed: one considering that actors are 100% available and reliable (work-as-imagined) and another considering suboptimal availability and reliability (work-as-done). The multilayer networks were analyzed through actor-centered (Katz centrality, degree deviation, and neighborhood centrality) and layer-centered metrics (inter-layer correlation, and assortativity correlation). Data for both papers 1 and 2 were gathered from the same 34-bed intensive care unit of a large teaching hospital in Southern Brazil. Finally, the third paper presents an approach for assessing the match between task risk and actors’ contribution to resilient performance, measured by the RS developed in the first paper. The law of requisite variety (LRV), which states that a complex controller (i.e., actors who have a high RS) is necessary for coping with a complex process (i.e., high-risk tasks), is the theoretical lens for analyzing that match. Cluster analysis divided the actors into first-order and second-order resilience groups, even though the clusters did not differ regarding the task risk. Based on the LRV and considering that the performance of the ICU is more often than not successful, the findings suggest that even the actors at the second-order resilience cluster reached a minimum threshold of effective social interactions. Data for the third paper was gathered from a six bed-cardiac intensive care unit located at the same hospital where data was gathered for the previous two papers.