Exploring the impact of different multi-level measures of physician communities in patient-centric care networks on healthcare outcomes: A multi-level regression approach

Abstract

A patient-centric care network can be defined as a network among a group of healthcare professionals who provide treatments to common patients. Various multi-level attributes of the members of this network have substantial influence to its perceived level of performance. In order to assess the impact different multi-level attributes of patient-centric care networks on healthcare outcomes, this study first captured patient-centric care networks for 85 hospitals using health insurance claim dataset. From these networks, this study then constructed physician collaboration networks based on the concept of patient-sharing network among physicians. A multi-level regression model was then developed to explore the impact of different attributes that are organised at two levels on hospitalisation cost and hospital length of stay. For Level-1 model, the average visit per physician significantly predicted both hospitalisation cost and hospital length of stay. The number of different physicians significantly predicted only the hospitalisation cost, which has significantly been moderated by age, gender and Comorbidity score of patients. All Level-1 findings showed significance variance across physician collaboration networks having different community structure and density. These findings could be utilised as a reflective measure by healthcare decision makers. Moreover, healthcare managers could consider them in developing effective healthcare environments.

Figure 1. Conceptual multi-level regression models for patient-centric care network.

Figure 2. A multi-level regression model for analysing patient-centric care network.

Figure 3. Construction of a physician collaboration network.

In a hospital (say H1), patient Pa1 is visited by Ph1, Ph2 and Ph4 physicians, patient Pa2 is visited by Ph2, Ph3 and Ph4 physicians, and physician Ph3 and Ph4 visit patient Pa3. This patient-physician network is depicted in the panel (a). The corresponding physician collaboration network for this patient-physician network is demonstrated in the panel (b). In this physician collaboration network, there are network connections with weight 1 between Ph1 and Ph2, between Ph1 and Ph4, and between Ph2 and Ph3 because they visited a common patient. The weight of the links between Ph2 and Ph4 and between Ph3 and Ph4 are two as they visited two common patients.

Figure 4. Illustration of network communities in an abstract social network.

A network is said to have ‘community structure’ if the nodes of that network can easily be divided into sets of groups such that each set of nodes is densely connected internally. Each set of nodes is called a ‘community’. In the above network, there are three communities (i.e. C1, C2 and C3). Any node of these communities has more links with other nodes of the same community compared to the number of links with other nodes from other communities.

Figure 5. An example of the construction of a representative physician collaboration network and extraction of its communities from the research dataset of this study.

In this physician collaboration network, 28 physicians visited 57 times to 10 patients. The label for each actor represents a system generated unique ID for each actor. Each physician ID ends with the suffix of “_M” and each physician ID consists of only numbers. (a) Physician-patient links – a red circle represents a physician and a patient is being represented by a green diamond; (b) Corresponding physician collaboration network; and (c) Detected communities within the physician collaboration network–physicians belonging to the same community are being represented by the same shape with the same colour. Three communities have been identified in this representative physician collaboration network.