In hospitals, “sentinel events” are events that carry with them a significant risk of unexpected death or harm.  It is estimated that ⅔ of such sentinel events result from communications failures during the handoff of a patient from one provider to another (e.g. during a nursing shift change).

In a recent paper, a team of researchers from the University of Arizona and Carnegie Mellon University outlined how social network analysis can help analyze and prevent such errors.  The research team examined seven medical-surgical patient care units from three acute care urban hospitals in the U.S. Southwest, with a total of 226 nurses. One set of data reflected information that the day shift gave to the oncoming night shift, collected in nurse pairs as follows:

Day nurse 1 > Night nurse 1   never gave info (0)
Day nurse 1 > Night nurse 2   often gave info (3)
Day nurse 1 > Night nurse 3   constantly gave info (4)
etc. for all night nurses

Day nurse 2 > Night nurse 1   seldom gave info (1)
Day nurse 2 > Night nurse 2   seldom gave info (1)
Day nurse 2 > Night nurse 3   constantly gave info (4)
etc. for all night nurses

then etc. for all day nurses

The team also collected a second set of data that measured data the night shift received from the day shift.

You may recognize this as network (graph) data – an edge list (list of connections) where the edge weight is the extent of information flow (the number in parentheses).  From these data, the research team was able to measure characteristics of the network, such as:

• • Degree distribution:  The distribution of the number of connections per individual
  • Density:  Another measure of connectedness, focusing on actual links as a proportion of total potential links
  • Hierarchy:  A measure of the extent to which information flow is two-way
  • Centrality:  The number of individuals whose connections make them central to the network
  • Cliques:  Small unconnected sub-networks
  • Isolates:  Individuals not well connected

The team identified 12 such metrics, out of the 80 provided by the software they use, ORA (skim this 1800 page user guide to get a sense of these metrics).  The characteristics of a nursing network help determine the speed with which information traveled throughout the network (diffusion).

Next the team looked at outcome measures for the patients in the seven medical-surgical units, metrics such as:

• • Falls
  • Adverse drug events
  • Patient satisfaction scores in four areas

The team found that strongly connected nursing networks fostered reduced falls. On the other hand, somewhat contrary network characteristics – the presence of cliques and isolates – were those that reduced adverse drug events.  

From a statistical perspective, though, perhaps this was not so surprising.  With 7 different units being tested, 12 network metrics and 7 patient satisfaction scores, there is ample scope for spurious correlations to occur by chance.  In any case, the visualization and description of the nurse communication network is a valuable tool for hospital management.