Network Analysis Subtleties in ICU Structures and Outcomes

To the Editor:

We were extraordinarily pleased to read “The Structure of Critical Care Nursing Teams and Patient Outcomes: A Network Analysis” conducted by Dr. Costa and colleagues (1). This is a timely study using methodologic approaches to measuring structure in complex healthcare systems, such as critical care teams. In this letter, we feel there are additional approaches Dr. Costa’s team can consider, which we believe will improve the quality of the following network analysis in critical care.

The excellent way Dr. Costa and colleagues created connections among nurses has an unfortunate potential risk of building a high-density network, which may lack structural information, such as k-core and betweenness (2, 3). This Michigan team defined a connection (tie) between two nurses as they provided direct care for the same patient during the patient’s ICU stay. In this way, nurses caring for one patient within a period (the patient’s ICU stay) will form a complete subnetwork, within which all nurses are interconnected. The complete subnetwork has less structure information. Such a phenomenon becomes even worse (i.e., almost all nurses are interconnected in the nurse network) when 1) the patient’s ICU stays are prolonged (e.g., over 30 d) and 2) each nurse cares for a majority of patients in the ICU. As a consequence, most nurses will have the same values of k-core and betweenness (2), respectively. The downside here is disabling the exciting opportunity of investigating associations between network structure and mortality risk.

Understanding the evidence to validate that nurses are randomly assigned to a patient, regardless of their mortality risk, would augment this fine work. Currently, it is hard to determine if the low mortality risk is because of core and high-betweenness nurses or the strategies used to assign nurses to patients. If the majority of nurses are assigned to care for a higher percentage of low-mortality-risk patients than that of high-mortality-risk patients, then they will have more connections in the nurse network, and they have the potential to be core and high betweenness. Generally speaking, there are a larger number of low-mortality-risk patients than high-mortality-risk patients in the neurosurgical and surgical ICUs, so nurses caring for a higher percentage of low-mortality-risk patients have more connections. Therefore, the finding would be that nurses caring for patients with a higher percentage of low mortality risk have more connections in the network, so they are core and high betweenness.

To let researchers understand such a complicated situation deeply, we provide an example. Assume we have a scenario in which 50 nurses from group A and 50 nurses from group B cared for both high-mortality-risk and low-mortality-risk patients. Nurses in group A cared for 90% of patients with low mortality risk and 70% of patients with high mortality risk. Nurses in group B cared for 70% of patients with low mortality risk and 90% of patients with high mortality risk. In this hypothetical scenario, a low-mortality-risk patient was cared for by more nurses in group A than those in group B. Assuming there were 920 patients, 900 of them were low mortality risk, and 20 were high mortality risk. Nurses in group A would care for 810 patients with low mortality risk and 14 patients with high mortality risk, whereas nurses in group B would care for 630 patients with low risk and 18 patients with high risk. Based on the way Dr. Costa and colleagues built the nurse network, nurses in group A had more dense connections, and thus they are potentially core and high betweenness. An explanation of the finding would be that because group A nurses cared for more patients with low mortality risk, they were core and high betweenness. In short, if Dr. Costa and colleagues can provide the percentages of low- and high-mortality-risk patients cared for by high core and betweenness nurses, then it will improve the quality of this already high-value paper.

Dr. Costa and colleagues used the number of high-betweenness or core nurses involved in individual patient care rather than the ratio of those nurses in their finding, which we think provides some but limited evidence on effective staffing interventions. For instance, beyond being cared for by more high-betweenness nurses, a patient with low mortality risk can also be cared for by more low-betweenness nurses. Moving forward, we believe that a study focusing on the percentages (percent of core and high-betweenness nurses of all nurses caring for a patient), instead of the raw numbers, can supply more comprehensive suggestions to ICU staffing.