Visualizing Hospital Networks
Wed Nov 14, 2018 · 72684 words

Visualizing a Network

My goal is to build an interactive network model that jiggles and is generally fun. I ended up exporting the final product into Shiny, and it looks like this:

Below, I’ll show you how I created this project, starting with a similarity matrix of hospital websites and a list of hospital attributes. To see how I created the similarity matrix, check out Part One. If you’re interested in doing more network visualization, I highly recommend this guide. It’s a great page, and has tons of options for visualizing a network.

Setting Up the Workspace
First, I need a few packages. I use iGraph and visNetwork. Note that, if you don’t want your graph to be extremely interactive, you can generate some pretty interesting graphs with just iGraph.

library(igraph)
library(visNetwork)

Next, I need to import my data. This data was generated using the American Hospital Directory.

#Links
highmat <- readRDS("~/Box Sync/Dissertation (1)/Text Analysis Paper/Data/CMS/high_similarity.rds")
#attributes
cleanedcms <- readRDS("~/Box Sync/Dissertation (1)/Text Analysis Paper/Data/CMS/cleanedcms.RDS")

Preparing the Data for visNetwork
For the final visualization, I’ll be using visNetwork, but I need to get my matrix into a data format that visNetwork will recognize. For that, I’m using iGraph.

#This creates the iGraph object
inet <- graph_from_adjacency_matrix(highmat, mode=c("undirected"), weighted=TRUE, diag=FALSE)
inet <- simplify(inet, remove.multiple = T, remove.loops = T)

#This takes the object and generates a DataFrame with a list of IDs and connections
Dedges <- as_data_frame(inet, what ="edges")

This gets my “edges” (the connections between hospitals) all set up and ready to go. However, I also want to set up my “nodes”–the actual websites. I have a comprehensive list of attributes that I got from the AHD, but I need to do a bit of tweaking in order to get my attributes of interest to show up on the final graph in an intuitive way. The way this works is pretty fun–you simply add new “variables” to a node dataframe that specify characteristics. visNetwork knows that “shadow” is a parameter for the graph.

#Create a new dataframe from my Cleaned CMS data
vis.nodes <- cleanedcms 
 #identifies that a hospital's CMS is its ID
vis.nodes$id <- vis.nodes$CMS
#I want my nodes to have shadows (so they look exciting)
vis.nodes$shadow <- TRUE  
#text on click--the hospital name
vis.nodes$title <- vis.nodes$hosp_name 
#scale dot size to match the number of beds
vis.nodes$size <- (cleanedcms$Beds)/25 
#This creates a list of colors (as I am too lazy to pick my own)
colrs <- rainbow(87, alpha=1) 
#assigns node color by hospital network
vis.nodes$color.background <- (colrs)[as.factor(vis.nodes$network)] 
#node border should be black
vis.nodes$color.border <- "black"
#I don't want each node to have a label
vis.nodes$label <- NA

Creating the visNetwork Graph
Finally, I graph my network using visNetwork. I highly recommend checking out the visNetwork website for more information about creatig your own dynamic network. A few key points that I figured out:
-You can speed up the generation of the network, and increase its motion, by turning off stabilization.
-If you want to change the shape of the network, use “visPhysics” not “visLayout”. visPhysics actually provides tools to change the layout, while visLayout seemed a bit clunky and didn’t work for me.
-Related: You can change how close your dots are by changing the “gravitational constant” in physics.

par(mar=c(0,1,0,1))
visNetwork(vis.nodes, Dedges) %>% 
  visOptions(highlightNearest = T, 
             selectedBy = list(variable="control2", main="Select by control type"),
             width="500px", height="500px",
             collapse = list(enabled = TRUE, clusterOptions = list(shape = "square")), 
             autoResize = T) %>% 
  visPhysics(stabilization=FALSE, solver="forceAtlas2Based", forceAtlas2Based = list(gravitationalConstant=-5))

Figure 1: Hospital Website Similarity

The above works, but isn’t very versatile. I can open it in a webpage, save it as HTML, or embed it in RMarkdown, but I wanted to be able to share it liberally. So I converted it into a Shiny application. Using Shiny allowed me to turn it into a web application, available from anywhere, and I can embed it in most sites using “iFrame”. The process of turning the project into an app goes a bit beyond the purview of this writeup, but it wasn’t as hard as it sounds!


back · meet allison roberts · data projects · publications/presentations · healthcare improvement · teaching · main