Creating automated visualizations can be difficult when working data that has not been standardized. In a large hospital system, standardization requires multi-level communication across many departments as well as strict adherence to those standards so that processes can be implemented. Alex Ratliff talks us through how he created a dashboard around ever changing standardization issues.
Tell me about your background and role at Data Tapestry.
I’m currently a data scientist at Data Tapestry. My background is in math and analytics. During the first three years of my career, I worked at a company that contracted with the department of defense. While I was there, I worked on making dashboards to monitor data flow to make sure the data was being processed correctly. There were different sites that make data transfers. Our job was to make sure each job had the proper amount of bandwidth. I mostly used R and SQL to manage that.
Tell me about the dashboard project you worked on.
When I started working on the project, there were a few Tableau dashboards in place. The idea was to have a centralized location where they could view performance metrics as well as benchmarks for specific hospital facilities. The metrics were self-reported by the hospital. Previously, they had been using spreadsheets to track this information. So, I needed to aggregate several data sources, ensure standardized metrics across all the facilities, and eventually automated the process.
Can you describe the relationship you had with the stakeholders?
The team was very interested in having a data driven solution. However, I had to manage expectations around how long some of the work would take. Much of that work was coordinating many individuals to commit to a standardized way of data collection. Once those standards were implemented, I was able to automate data refresh feeds so that the dashboard was as close to real-time as possible.
How did you handle data quality issues?
One of the first things I encountered was that the data had been ingested incorrectly. Some rows were being recorded in the wrong field. I had to constantly verify the data with the stakeholders and then monitor the data refresh to make sure the quality was consistent.
What kind of techniques did you use to aggregate all of the data?
I combined several different data sets and spreadsheets. The metrics for one facility might be spread across the different data sets. So, in addition to standardizing metric names, we also had to create a facility crosswalk to ensure we were identifying the correct facility across various data sources
Were there any interesting challenges that you encountered during the project?
I had to automate the data refresh script so that the stakeholders could see the dashboard with the latest data. This involved some code conversion, error handling, and a lot of code cleanup.