#TableauIsWhy I see the underlying data that matters most…
The dashboard
I had been using Tableau for a few years as a consultant at Teknion Data Solutions, when I built what I felt was one of the best dashboards I had ever created up to that point. It was a Patient Readmission dashboard that allowed doctors and patient care specialists at a hospital to find readmissions of patients based on all kinds of dynamic parameters and filters such as the window of time, the diagnosis codes, the procedures, and the attending physicians.
It’s one of the most technically intricate dashboards I have ever built and it required using quite a few advanced features of Tableau: data blending, layered table calculations, complex action filters, and context filters. Additionally I had to work closely with the data modeling and ETL teams to make sure the source data was in exactly the right shape and contained the fields necessary to work seamlessly in Tableau.
When I was done, I had a dashboard of which I was very proud. It was something that, short of a custom application, only Tableau could do – with its unique abilities and features.
The moment of clarity
And then, I actually used the dashboard…
… and clicked on the mark that indicated a patient visit…
…and I saw the details of the patient:
That was a moment of clarity. The data wasn’t just numbers and text and structures. The marks weren’t just bars and circles and lines.
The mark I had clicked on represented the real stay in a real hospital by a real person who had a lot of very real serious issues in their life. The line in the Gantt chart was a part of their life – how she had visited the hospital multiple times to be treated and hopefully made well.
And I realized that data visualization and dashboards aren’t just a way to see data – they are ways to change things for the better. Hospitals could use this dashboard to understand and improve patient care. Stays could be shortened, procedures could be analyzed to see which were more effective, and readmission rates could be cut. Real things could be made better.
The underlying data that matters most
From a technical perspective, I always tell others the first question to ask is “What does one record of my data represent?” But that’s also a good question to ask from a human perspective as well.
Is a record of data a human life? Does the mark on a graph represent an individual’s job performance? Does the bar chart of sales impact commissions for people trying to make a living?
The responsibility to get things right is immense for those of use engaged in data visualization and visual analytics. We must not think of the data as merely numbers, dates, and text. The data and the visualization of the data impacts real lives. We must dig deeper to understand the underlying data that matters most.
And the opportunity to make a positive difference is great…
“Human Trafficking”
Nelson Davis
Each mark is a population in slavery
“The Adoption Gap”
Steven Carter
Each mark is a country and the number of households needed to adopt all orphans in that country
“Endangered Safari”
RJ Andrews
Each mark is a species with indication to its endangered status
Josh,
My most important work to date in Tableau has been a deep dive into medical data. I spent over a year going to “medical school via data”. Due to HIPA restrictions and confidentiality, I have not been able to publish anything from this work. I wrote about the experience briefly here (https://3danim8.wordpress.com/2014/04/05/how-i-have-been-having-fun-with-tableau-and-medical-data/).
Unfortunately I have not been able to show and demonstrate the absolute mastery that Tableau gives us for uncovering insights in medical data. Suffice it to say that me and my partner were able to combine Alteryx, Tableau and Statistical modeling to investigate the efficacy of medical treatment technologies given to patients. We investigated three primary diseases including pneumonia, sepsis and heart failure with key measures of readmission rate, mortality and length of stay using several years worth of actual patient data. We determined with absolute certainty which treatment technologies worked and which didn’t do so well. We uncovered performance differences between doctors and hospitals. We investigated comorbidities and uncovered some amazing things that even surprised the chief medical officer.
This work remains the most significant achievement of my career and I wish the analytical platform we developed could be unleashed on other diseases and applied at other hospital systems. I share your enthusiasm for Tableau and medical data. I expect that the day will come when all hospitals will realize the value of this type of analysis.
Ken
Hi Joshua,
Regarding this:
“I had to work closely with the data modeling and ETL teams to make sure the source data was in exactly the right shape and contained the fields necessary to work seamlessly in Tableau”
What is the right shape and some example fields for a dashboard like this?
Getting data for events like this into the right shape is challenging! That’s because there are two basic “shapes” of data — and each works well for different kinds of questions. One shape contains all the various dates as columns — i.e. one row stands for one patient visit and has a column for Arrival Date, Admitted Date, Surgery Date, Discharge Date, etc… That works well if you need to be able to calculate differences between the dates or need to be able to easily compare the dates. That’s fairly easy because it’s all in one row. But it’s hard to visualize all the dates together, because which do you use for the date axis? The second shape of data has a single date column and then a row for each event (e.g. Row 1 – Event Type: Admit, Date: March 21; Row 2 – Event Type: Surgery, Date: March 22). That kind of shape works better for visualizing (like the Gantt chart) because you have a single date for a single axis. However, calculating or comparing events is more difficult because you have to be able to work across rows within certain partitions (so, think table calcs). Hope that helps!
Thanks! It does. I had not seen the answer before