A Visual Analytics Design for Connecting Healthcare Team Communication to Patient Outcomes

TL;DR

This work frames healthcare team communication as time-evolving bipartite networks derived from EHR access logs and introduces EHRFlow, a visual analytics system to connect network structure and dynamics to patient outcomes. It combines a flexible network-measure space with an MLP-based optimization to derive an interpretable communication-effectiveness metric and uses network perturbation to identify influential HCPs and notes. For efficiency, it analyzes latency and frequency along time-respecting paths and visualizes information flow at global and local scales. Through case studies on cancer patient data and expert feedback, the approach demonstrates how targeted communication patterns and note dissemination can potentially improve patient survival and offers a generalizable framework for analyzing teamwork in complex, time-sensitive environments.

Abstract

Communication among healthcare professionals (HCPs) is crucial for the quality of patient treatment. Surrounding each patient's treatment, communication among HCPs can be examined as temporal networks, constructed from Electronic Health Record (EHR) access logs. This paper introduces a visual analytics system designed to study the effectiveness and efficiency of temporal communication networks mediated by the EHR system. We present a method that associates network measures with patient survival outcomes and devises effectiveness metrics based on these associations. To analyze communication efficiency, we extract the latencies and frequencies of EHR accesses. Our visual analytics system is designed to assist in inspecting and understanding the composed communication effectiveness metrics and to enable the exploration of communication efficiency by encoding latencies and frequencies in an information flow diagram. We demonstrate and evaluate our system through multiple case studies and an expert review.

Figures (7)

• Figure 1: Illustrations of time-respecting paths: (a) presents an example of a directed temporal graph. The colored nodes are the analysis targets. (b) follows the red highlighted edges, retrieving nodes that can be influenced by node $D$. (c) follows the green highlighted edges, retrieving nodes that can influence node $B$.
• Figure 2: (a) The star pattern in undirected and unipartite networks. (b) The two-star pattern in directed and bipartite networks. The blue nodes represent one type of nodes and the red nodes represent the other type of nodes. Nodes $i$ and $j$ constitute the two stars in the respective types of nodes.
• Figure 3: (a) and (b) represent two patients with different survival outcomes. (c) is the metric swarmplot displaying the communication effectiveness score of every patient in patient group A. (a1) and (b1) are the stacked boxplots of the aggregated distance measure for the two patients, with each patient's distance measure depicted using a vertical line atop the aggregated distance stacked boxplot distributions of all patients with different outcomes. (a2) and (b2) illustrate the measure evolution of the aggregated distance for the networks of the two patients, brushed with the chosen observation time window. (a3) is the global information flow diagram, and (a4) is the local information flow diagram, presenting the disseminated set reachable subnetwork of the ego HCP (i.e., the green cross) who treated the survived patient highlighted in (a).
• Figure 4: A comparative study of disconnected communications between two representative patient networks selected in \ref{['fig:ui']}(a) and \ref{['fig:ui']}(b). (a1) and (a2) accentuate the reviewed and disseminated reachable subnetworks of the survived patient, while (b1) and (b2) highlight the reviewed and disseminated reachable subnetworks of the deceased patient. Disconnected individuals from the ego HCP are circled in brown.
• Figure 5: The development of reachable subnetworks over time for the survived patient's network chosen in \ref{['fig:ui']}(a). The top and bottom rows depict the expanding reviewed and disseminated sets after each critical treatment stage, indicating that the selected HCP actively reviewed and wrote new EHR documents promptly.