Visualizing the Data Consumption Continuum

As this is intended to be a blog about visualization it seemed only fitting that I devote this post to visualizing the concept I introduced last week that I called a Data Consumption Continuum.

It all started with …

A few years ago my good friend and colleague at the time at Northeast Georgia Health System, Zach Ivie, helped me solidify a visual very close to the following.

DataConsumptionContinuum_Image1

Despite the fact that this is a rather crude version of the work of art that Zach, a graphic guru, helped produce, I believe the message will come across loud and clear anyway. That message is that as you move through the Data Consumption Continuum there is a higher level of skill demanded of both the data consumer and the data producer. However, the benefit to the company of making those investments is clearly that there is a higher level of return on investment as well as a higher level of process improvement that can be achieved.

Not very hard concepts to grasp, but we found it very useful in trying to help data consumers understand that we weren’t trying to replace their roles. We needed them to understand we were trying to help them become even more valuable to the company. Having a visual that they could readily grasp was very effective in our efforts to establish trust that we saw them as teammates and provide the incentive to do the hard work that needed to be done to progress through the continuum. Obviously the same image helps build support through management as well. (In hindsight I would add “Time needed to produce” along the bottom as an annotation as well to help management understand each step in the continuum takes longer than the prior steps.)

Much appreciated input came along …

I presented the graphic when I spoke at the 2014 Qlik World Users Conference in Orlando.

DataConsumptionContinuum_Image2Among othersI met at the time, two gentlemen from Lee Memorial Hospital named Marcello Zottolo and Roger Chen stood out. Over the course of a few weeks following the conference we spent a lot of time chatting and growing our friendship. During which time these Process Improvement gurus asked why the image was like steps instead of being a straight line up the continuum. I indicated that I wanted to illustrate very clearly that you didn’t move up the continuum by accident. That there had to be clear and pronounced commitment to doing so. But I asked how they it. They were gracious enough to share an image with me that was something like the following, and you know I love visuals:

I liked the implications that their suggestion implied. Clearly we should never be standing still and that even the stages of the Data Consumption Continuum are in fact a process. That the hard work I indicated could be visualized as a ball rolling forward, and that there should be documentation supporting the standard work after advances are made. So I adjusted my thinking to include their great thinking and suggestions along with my basic premise that the different stages required a knowing decisions.

Combining the thoughts provided room for annotations to note exactly what I believe those commitments are: Ever increasing levels of trust. There are clearly static reports that are simply needed as “checklists” for workflows. However, there are also thousands more  developed simply so that business data consumers can avoid having to accept any responsibility. It’s easy to say “I took off the report that Dalton built for me. Clearly he did it wrong.” Which is also ok with me because I can say “Clearly they gave me bad requirements.” Thus the proliferation of 18 versions of the truth continues. Moving to Guided Analytics forces those data consumers to accept the responsibility for making selections filtering the data they are then responsible for the choices they make. Which only makes sense as they know their business better than IT and a lot is missed in writing up requirements documents. As you are probably well aware from experience getting others to accept responsibility is easier said than done.

Moving along the continuum from Guided to Self Service Analytics forces a department wide level of trust on IT’s part that if they construct governed data sources they can trust the business units to build the applications instead of IT building them. The next stage requires an enterprise wide level of trust that the investments of the amount of time, salary and technology required will pay off. (Clearly there are a lot more complexities than just “trust” but the symmetry in the visual was to hard to pass up.)

DataConsumptionContinuum_Image3

One of the great pleasures in my search for more effective ways to communicate and be the catalyst for change is the identification of others in different fields who are also working tirelessly to train up. Others who continually reiterate and show examples to help crack my think skull.

Never stop consuming knowledge …

One such field that I’m trying to hone my skills in is Storytelling. Cole Nussbaumer is one of those afore mentioned people that I love learning from on an ongoing basis. I’ve always been a self-professed data nerd and was always of the mind “For crying out loud the data speaks for itself.” Cole has helped me realize that’s not even close to reality in the field of Business Intelligence. Her blog conveniently named for people like me Storytelling With Data is a resource I’ve referred to many times. She backs up her pontificating through examples to show the flow through projects. For me seeing a visual as a starting point then seeing it progress forward using her techniques step by step has really helped me understand the value of storytelling and more importantly how to do it.

But how does this relate to this post? Well if you give me a second I’m getting to that part of the “story.” I’ve begun realizing that Actionable Intelligence requires not only the data and visuals but storytelling to explain things so that executives fully understand the ramifications. Thusly I present my current approach of Visualizing the Data Consumption Continuum.

DataConsumptionContinuum_Image4

Disclaimer: Clearly I’m no graphic artist. Hopefully my crude crayon type drawings aid in your ability to consume the concept of the Data Consumption Continuum. 😉

 

 

 

Posted in Training, User Adoption, Visualization | Leave a comment

The Data Consumption Continuum

Consum Mass QuantitiesWe humans love to consume.

Food.

Water.

Fuels.

And my personal favorite Chocolate.

You name it we want to consume it. Health warnings have little effect at deterring us from consuming mass quantities of the wrongs things. Yet sadly there is one thing that we need to survive and it seems that no matter how hard we try we simply can’t consume enough of it … DATA.

We need to consume data to survive in our own lives, as well as within our occupations. Yet try as we may we seem to be starving for it. I believe part of the problem is that unlike eating chocolate there is no “right” way to consume data. We humans are all over what I call the Data Consumption Continuum and can’t figure out how to accommodate one another’s ferocious appetite for this particular commodity.

Static Data

Reports are the most basic form of data consumption. Static reports have been around much longer than we have had computers and or a specialty field called Business Intelligence. They are a wonderful thing in situations where the subject matter doesn’t change. Whether it be workflow reports, ie lists that I mentioned in my last post we love to consume or whether it is truly static data like reports of what our Accounts Receivable was as of a particular date.

So there are good reasons to have the 3,209 static data reports your company has. But it simply doesn’t make sense to have business meetings end with the conclusion “Yep our company is losing money. Let’s all have a great week and enjoy the last days before we are unemployed.” Someone in the meeting is inevitably going to ask the “Who, what, when, where and why” questions. Thus report 3,210 ends up being born over the course of the next several hours, days, weeks or months.

Guided Analytics

Many including yourself may think of guided analytics as the polar opposite of static reporting. I’m not one of those people.

I believe guided analytics are a wonderful way to speed up the process of asking questions and getting answers. They allow you to “drill, drill, drill” until your heart is content. My posts have praised and demonstrated the many different ways that visualizations enable us as human beings to more rapidly perceive and consume massive quantities of this wonderful commodity we call data.

For some guided analytical applications are like a sudden speed reading course. The applications allow them to consume the data from 3,209 static reports in 1 sitting. For others it’s like handing their companies 0’s and 1’s to Michelangelo and suddenly a beautiful portrait of their company appears.

The definition of the word continuum is “a continuous sequence in which adjacent elements are not perceptibly different from each other, although the extremes are quite distinct.”

For all of the wonderful things guided analytics offer I see them as a mere single step away from static reporting for one reason. They only allow you to answer the questions that you already had in mind when the applications are built. If you think you will want to see how surgical costs relate to length of stay you build your application to contain that data. It’s great that we have the technology now that enables us to consume that many 0’s and 1’s at the same time, in really slick visual ways. But is it really so magical that we are able to answer the questions we had when we asked for the application to be built?

Self-Service Analytics

Hopefully you now understand why I’ve used the phrase “data consumption continuum” and why I believe guided analytics are but a single step away from static reporting. If you don’t then you stopped reading already, so those that are still tracking with me are ready to take another step forward on the human consumption continuum to self-service analytics.

I’m honestly baffled by the concept that self-service analytics are a way to allow end users to quickly visualize data from some a single data source. Woo-hoo look at the pretty pictures I can create for you from your XLS file or a single SQL query. Really??? That’s a leap forward in technology? I’ve never known a version of Excel that didn’t offer charting right inside the tool itself. If all you want are pretty graphics from a single data set just use the tool your data is in which is likely Excel.

I believe self-service analytics enables us to answer the questions that we didn’t have when we built our guided analytics solutions. It enables us to consume all of the data we built into our application and then begin consuming more data. Data that perhaps wasn’t even available when our guided analytics applications were constructed.

The important thing in my mind is that self-service analytics must offer the ability to consume new data as well as our existing data sources. When I get to the end of the road and can’t get answers just looking at my cost and length of stay data I need the ability to now consume readmission data along with that data not instead of it. To consume patient vital information along with it. To consume patient satisfaction data along with it. Whatever the new data may be self-service analytics should be an additive process. A step forward on a continuum of our consumption of mass quantities of data. One that allows us to grab the data and move on rather than having to go through a long requirements and prioritization exercise with IT.

Predictive Analytics

For many data science is very hard to understand. It seems that they think Data Scientists go into a room with their magic potions and terabytes of data and emerge with all of the answers to the company’s problems. That’s simply not the case.

Data scientists simply apply age old statistical formulas to data. The same data that we display in static reports. The same data that we display in guided analytics applications. The same data that we consume in self-service analytics. But they do so in a slightly advanced and more scientific approach. You or I as mere mortals say “My spidey senses are tingling. I think there may be a relationship between our profit and the patient’s length of stay.” We ask for a report, we use guided analytics or perhaps self-service analytics and if we see even a minor trend we immediately jump to a conclusion there is a cause and effect, not just a relationship and we say “Aha I’m a genius! Quick change everything in our company I’ve found the problem.”

Data scientists say “give me the data for every variable we have and I will help you find the BEST correlations. The ones that statistically have the highest probabilities. The factors that actually lead to patients being readmitted for instance.”

That is a great thing. But it’s not like they simply run a magic statistical formula and come to the answers because that isn’t how statistics works. They methodically run formulas on different combinations of the variables. What about A, B, C, D and E? Nothing there so let me try A, B, D, E and F. Nothing there. Let me try this. Let me try that. Let me try the other thing. And they churn and churn and churn very methodically until months later they provide an answer. An answer that in the past could have been achieved with more people and more time. So my assertion is that like the steps forward from static reporting to guided analytics and then to self-service analytics predictive analytics is a step forward simply in that it enables us to do things faster.

Each phase I’ve discussed allows us to take a step forward. A step that speeds things up. A step that allows us to consume more data. But the incremental steps are obvious to follow from one to the other. Yet if you look at how far apart static data is from self-service analytics you see that those extremes really are quite different. Static data reports identify the historical data we already had in our systems. While the result of predictive analytics is even more data that we can use and when combined with our current data can identify alternative actions we can take. Prompting us to take action rather than simply reporting to us. In other words a Data Consumption Continuum.

I have two reasons for this post:

  • To help you realize that wherever you may happen to be along the continuum in your ability to either produce or consume data you should consider the fact that the others in your company who may be at a different stage aren’t “dinosaurs” and they aren’t just “resistant to change”  and aren’t really that different from you. If you change your focus and understand the key concepts that differentiate the stages you will be better equipped to communicate with them so that you can both move forward as a team.
  • To help you realize that instead of looking at each phase as something completely different that requires it’s own tools you should consider thinking of implementing a data consumption “platform” rather than implementing new tools for each stage you progress.  A platform that enables you to surface your valuable data one time and reuse it over and over along the various stages.

Like what you are reading be part of the conversation and share your thoughts. Also consider showing me some love and following @QlikDork on Twitter.

 

Posted in Random thoughts, Self Service | 2 Comments

Crawl. Walk. Run. Fly?

CrawlingIn 30 years as a parent I can tell you I’ve had many memorable moments with my 2 amazing daughters. As humans, most of our memorable moments around babies involve movement. Let’s face it as cute as they are it gets old just watching them lay on their backs and smile. So when they first get the desire to start moving and can make themselves roll over we get excited. When they can finally control the movement of their hands and their legs and can crawl to us our hearts jump for joy. When they can take those first Frankensteinish steps to our waiting, open arms we get overwhelmed. Once they’ve mastered balance and movement and begin running we get to play our first games of chase and tag with them.

Crawling. Walking. Running. Memorable moments indeed. Those movements enable us accomplish so much. How much more could we accomplish if we ever learned how to fly. Don’t look at this post like I’m crazy … we’ve all jumped off the porch and flapped our arms wildly. Some of you jumped from heights higher than a porch attempting to fly and broke some bones in the process. You know who you are.

Yes there is a point to all of this. As the parent of 0’s and 1’s (data) for the past 30 years I’ve seen it crawl, walk and run and I’m beginning to see it fly. No I’m not talking about the Cloud, I’m talking about Self Service Data Visualization. I suppose I should back up a bit.

Those who work in IT, specifically those who work in data administration type roles work with data at least 8 hours a day. For most of us that is more time than we spend with our own kids. We have principles, best practices we follow to protect the data. In many ways we are more highly trained to protect the data than we are to protect our own children. So it should be of no surprise to you in business roles when you deal with IT folks that are a little over protective of their data children.

Crawl.

I just love checking off items in lists. I think we all do. There is a certain sense of accomplishment each time we check an item off. Which is why I think so much of Business Intelligence surrounds simply generating lists and static reports. Don’t believe me? The Checklist Manifesto: How to ChecklistManifestoGet Things Right by Atul Gawande, is a best selling book that shows the power that simple checklists have in the complexity of our lives and how they can help us avoid failures. Kind of like allowing our “data babies” to crawl. We control how much of it goes out and to whom it is allowed to “crawl.” Don’t get me wrong there is a huge part of the workforce in many companies who rely on those lists to do their job processes. IT management also loves lists and static reports. They can easily be used to justify more head counts. If each report takes 1 week to build and you want 52 reports … boom 1 person year totally justified. Wait you want thousands of reports … “woo hoo” an entire data services team springs into being. You get to see your data. IT data workers still control the data. IT management grows its organization. It’s a win-win-win.

Walk.

Sooner or later though executives start asking for more. They want KPI’s, Dashboards and Scorecards. For some these are the Holy Grail in a matter of speaking in the field. But when you really think about all of these, they are simply the next incremental step in growing up. All the check marks, arrows and circles simply tell them the answers to the questions that they previously asked. Unlike crawling though, I think this is kind of like taking the first actual steps because at least the pretty pictures allow them to consume the data faster than they could if they had to read 317 separate static reports. IT is still central in the process, they get more staff, they get more resources and now there is direct interface with executives. It’s an ever bigger win-win-win. Our data babies are growing up indeed.

Run.

More than once in my posts I’ve shared the concept that Analytics allows the end user to answer tDataConsumptionhe question that they had in their head when they started, but it also allows them to ask the next question. Analytics allow business users to not only see issues but drill into them. Find the roots of successes or problems. So you probably aren’t surprised that I consider Analytics to be like running in terms of consuming data and I consider it a thing of beauty. Kind of like watching my granddaughter sprint across the soccer field and score a goal. It’s also a little scarier for those data parents in IT, because Analytics can’t be done without a lot of input from the actual business users and without a lot of learning about the business processes by the IT staff constructing the analytics applications. Our data babies sure have grown, and very rapidly indeed. Where has the time gone?

Fly?

If you consider how much we can accomplish simply by running just imagine what we as humans could accomplish by flying. Similarly Self-Service Data Visualization has that same great potential compared to just crawling with lists/static reports, walking with dashboards or running with analytics applications. Unfortunately for those of us in IT this goes against everything we’ve been trained to do. Everything we’ve spent our entire career perfecting … controlling the data. Self Service Data Visualization means trusting others to do the right thing with the right data. Trust me when I say “that is as hard for data parents to swallow as it was for me giving my daughter’s hands in marriage.” But can you really blame IT for fearing that? For 30 years I’ve seen business users combine data from cocktail napkins, flat files, spread sheets and personal hunches then deliver numbers in a meeting that directly conflict with those before them. So have executives. Which is exactly why so many companies are stuck watching their data crawl, walk or run … but never get to see it fly.

Flying

So what’s the answer? In my humble opinion it involves a marriage between IT and the business community. We’ve all seen that in human marriages opposites attract. So why do we allow them to repel and work against each other in offices? IT has the staff to properly govern and protect the data to ensure a single source of truth. That has to be respected. The business community on the other hand has the knowledge about their processes that in most cases IT completely lacks. That also has to be respected.

Companies can continue to allow rebels do self-service from untrusted sources and continue to plummet to the ground as a result. Companies can continue to allow IT to completely control all access and enforce that all data requests have to be resolved by them and continue to plummet to the ground as a result. Or they can arrange a marriage between the two. One in which IT is trusted to provide a single source of truth data libraries where the business users can then serve themselves. One in which we see our “data babies” leave the nest and fly.

But hey what do I know, I think I’m the parent of a couple of trillion 0’s and 1’s.

 

Posted in Self Service | Tagged , , , | Leave a comment

Light Dawns on Marblehead

I recently had the opportunity to demonstrate Qlik Sense for the Director of a large health system’s Revenue Cycle team. The consultant who recommended us spoke with me ahead of time and we got along great. “This guy has been around forever and has seen everything. He isn’t interested in the same old boring ways of seeing the current numbers” he said.

Hehe. That certainly makes at least two of us. Are you pretty tired of seeing the same old account balance visualizations? Yeah. So that makes 1,293,394 of us that have seen the same things for years and want something new.

Then our discussion led to something really interesting. “He is really looking for some way to visualize the entire flow through the Revenue Cycle process, not just the current status.” This demo just got really interesting. “How can we show claims and dollars as it flows from state to state through the process?” Suddenly “light dawn’s on Marblehead” and I’ve got it baby …  A Sankey Diagram.

I’d been waiting for over 6 months for a perfect case to utilize a Sankey Diagram and voila here it was, teed up for me perfectly and here is what I came up:

Claims#Sankey

This chart represents # of Claims. The AR section in blue is the number of claims for patients where all of the data was available on first try and could be billed. The Pre-AR section in purple represents the patients where something was askew in their charts and could not be billed.

“Big deal!” You say.

“Even dreaded pie charts can show that context comparison.” You say.

“Did I forget that visualizations are supposed to add value to the raw data?” You ask.

Stay with me about 2 minutes longer and you’ll see the method to my madness of using a Sankey Diagram where you see the entire flow. The proportions are about even on the left, but notice in the up right corner that the proportions aren’t. Also notice one of the wonderful features of a Sankey Diagram, if you hover over an area it turns dark so you can see the path that area has traveled as well as the number it represents. It’s 100 and it brings to the surface the fact that a smaller percentage of claims end up being paid within 30 days when we have to do work to the patients data in order to file the claim. The proportions are almost even for those paid within 60 days and the proportion of the dreaded money that doesn’t get paid for 90 days or longer is higher for those claims that needed to be worked.

This my friends is where the Sankey Diagram stands alone in the value it can add. You can easily scroll your eyes through the “flow” and track anything that seems suspicious. And it gets better.

Like any other chart the “# of Claims” doesn’t have to be on the only measurement you visualize. Would you like to see the Dollar ($) values? Why not, you have nothing better to do and I’m sure your curious … here it is:

Claims$Sankey

Right away something else jumps out … the proportion of claims was pretty close for AR and Pre-AR but the dollar values, are very far apart. How many bar/pie charts/tables would we have had to look at to uncover what was blatantly obvious thanks to our new friend the Sankey Diagram. Almost immediately we see that we are working a lot of cases to clean up data so we can bill but the dollar volumes for the cases are smaller than the dollar volumes of the “clean data.” What’s causing that? Why do claims that have to be worked before being billed take so much longer to pay once they are billed? True analytics will answer the question you had when you started but will also cause you to ask more questions.

I’m lucky in that I had a consultant, tell me about a customer who isn’t just satisfied with the status quo to challenge me to use a visualization I had been dying to try for 6 months. Consider how you could shake things up completely by using a Sankey Diagram or other sexy new visualization.

Forget the example data that the author chose to show (sales/dollars) and focus instead on what the essence of the visualization does for the end user. Why did someone spend weeks/months of their lives designing this new way of looking at 0’s and 1’s?  What will it let your end users see in a way that they probably couldn’t get by viewing 10 raw numbers.

Click here to check out the Sankey Diagram as well as dozens of other great visualizations you can utilize in Qlik Sense.

Posted in Visualization | Tagged , , , | 2 Comments

Visualizing End User Adoption

This past week I had my annual review. This time of year always makes me envious of those that produce widgets. I would love to be able to show my boss a list of all of the widget producers and say “See boss I’m in the top 2% of all of the widget producers in the company and the top 5% of widget producers around the world please compensate me accordingly.”

Since you are reading this post the odds are high that like me you produce Business Intelligence Applications and aren’t producing widgets either. So how do we evaluate our work? How should management evaluate us?

One way to evaluate our work might be to simply count the number of applications that we build. Of course I could barely contain a laugh just writing that. Obviously that is wrought with problems so let’s not even consider this option.

In a strictly financial sense many types of business can measure the return on investment (ROI.) But perhaps the application we spent 9 months building is intended to help resolve bottlenecks in the company that will lead to improved patient satisfaction. The resolutions that surface may cost the company more money. Does that mean we failed? Certainly not. So we can’t measure ourselves by dollars spent and dollars saved either.

If you follow industry pundits, tweets and other social media you might be familiar with the focus of many in the industry to focus on “user adoption.” Evaluating to what degree users actually utilize our applications is probably a good way to measure ourselves. It could be argued that it isn’t a perfect measure of our efforts, however, it does seem to be a pretty good measure of our effectiveness. Because whether we like it or not, our jobs involve more than just slapping an application together. End user adoption, or the lack there off, measures our ability to brand, market and support our application. It is also a pretty good representation of how trustworthy the data in our application is. One of the most important things that end user adoption will measure is our ability to effectively visualize the data in ways that encourage usage.

Taking advantage of Qlikview logfiles

One of the nice features of Qlikview is that it retains a log file in the background on the server that retains information about every single end user session that is invoked. Since the introduction for this post was so long I will spare you the pain of reading the raw data of a session log file and skip right to ways to effectively visualize end user adoption using the data that those logfiles contain. Please refer to other posts and discussions directly in the Qlik Community about where to find and how to access these log files.

The session log files contain information that would let us look at things like “how many users used the application” “how many times were sessions invoked” and “how many minutes were used.” Thus the first chart I present contains all 3 of those measures.

Evaluation_Method_Users

The first point I want to make is that I’ve masked the real document names. I did this for two reasons. First you don’t need to know what my real document names are. The more important reason is that I don’t want know what the real document names are. At least for the duration of the time I’m trying to figure out how to effectively measure “end user adoption.” That seems rather odd so let me explain.

Overcoming bias when choosing how to measure

I believe that we all have biases. I haven’t developed all of my companies applications and frankly I have some favorites of those that I have developed and some that I was forced against my will to develop. If I knew what the application names were I could be inclined to choose and recommend the metrics that make “my” applications look the best.

If you refer back to the chart you will see that Application 69 has the greatest number of users by a large margin. If I knew that Application 69 was written by me I could immediately come to the conclusion that our end user adoption should be based on the number of users that use the application. If I also wrote Application 85 I would probably really push for that policy. “Show me the money.”

But wait someone else on my team seems to have an objection because it appears that Application 85 has a lot of distinct users but only has a tiny amount of Sessions and very tiny amount of minutes. Hard for me to argue with that, and I put my outstretched hand back into my pocket.

A discussion ensues for several minutes and perhaps we re-sort the chart by number of sessions. Then by Number of Minutes.

Evaluation_Method_Minutes

The author of Applications 33, 49 and 56 now suggests that we evaluate end user adoption by the number of minutes used. I’d like to vote for that since I was the author of Application 69 but I also authored applications at the bottom of the chart for number of minutes. I’m kind of in a no win situation on this.

Can you understand my point for masking the document names so that we don’t really know which application was developed by whom? If we are choosing a method of evaluation we need to hide the real document names so that nobody pushes for a choice just because it is better for them.

Perhaps of equal importance can you appreciate the beauty of having all 3 columns displayed with numbers as well as bar charts? Obvious patterns jump off the page that help you avoid jumping to quick conclusions just based on 1 value or the other. If we are going to devise the method of coming up with an evaluation method we need the visualization to be really crisp, and this method provides that.

You might be screaming “You rotten Qlik Dork … just tell me which of the measures is the right one to use!” To which my reply is a resounding “None of them and yet all of them.”

You see nobody said we had to use a single value to do the evaluation of end user adoption and there is so much more that we can do with Qlikview to present a more complete picture. The chart below slices and dices the data a few other ways that presents a different picture.

Evaluation_Method_2

The first column interprets the average number of minutes per session. I might argue that value really represents user adoption of data analytics applications. Regardless if the application was built for a team of 5 or 50 to consume it reflects how long users stay engaged with the application. If we believe that is the goal then perhaps this is the perfect measure. Woo hoo. I think I wrote application 53.

Oh wait a second the other developer raises their hand to complain yet again, and points out that average is a really poor statistical indicator and that Median is a better measurement because it isn’t so swayed by outliers. In theory I agree, but as the author of application 53 it appears this statistics mumbo jumbo is costing me a big fat raise because while the average number of minutes per session is the highest, the median number of minutes per session is a measly 5. Phooey on heat maps I say, because if it weren’t color coded nobody would have spotted the 5.

Whether we used average number of minutes, or median number of minutes both point out something very interesting. If you look at the very bottom and see the numbers for Application 69 it appears that any of the single measurements like # of Users/Sessions/Minutes alone didn’t show a complete picture. Lots of total users and minutes, just not many minutes per session. Quantity for sure but not necessarily much analytical quality.

The third column illustrates a completely different measurement, the number of sessions per user. In other words how frequently are users engaging with our application. Like the raw data displayed in chart 1, displaying all 3 of these combined measurements helps paint a broader picture: Is our application engaging users for a very long time? Are they engaging once every 6 months, or are they coming back every other day and working?

Box plots to the rescue?

If we produce a box plot and make a few minor tweeks we can see that in fact Application 53 does in fact have a very high max value but the very low median of 5.

BoxPlot_Evaluation

But the beauty of what a box plot visualizes for us can best be seen as I scroll to the right a bit. Notice for applications 87 and 8 both have pretty high medians, which we would see in the heat map chart, but more importantly you can see that even their lowest values are near 10 minutes per session. Meaning when these applications are used they are used for a good amount of time and the time is pretty consistent in a predictable range. Perhaps we could measure the end user adoption based on the predictability and consistency with which users engage?

BoxPlot2

Of course any kind of visualization of end user adoption would be incomplete if we didn’t look at the values over time so that we could see if things were getting better, stabilizing or getting worse.

Evaluation_Trend

The wonderful thing is that while I focused on each method individually the great thing about visualizing data in Qlikview is that we can keep the entire picture together so that we get a true overview. A scorecard of sorts for each application.

WholePicture

The truth about visualizing end user adoption

In the end the truth about measuring end user adoption is simple – every application is unique. Sorry to break this to you 7 pages in but you can’t compare an application that was built to surface a small set of data in a scorecard fashion to an application that is really meant to be used as an ad hoc interface to find a cohort of patients. Perhaps 80 supervisory people per month use the scorecard for 1-2 minutes each time. While only 5 people use the other application for 2 hours at a time multiple times per month.

Don’t fret though? Just because I can’t compare Application 69 to Application 53 doesn’t mean I can’t apply what I know about Application 69 and the intended audience and take steps to interact with the users and figure out how to improve the appropriate set of numbers. I may never get more than 10 users for Application 22, but if I can address issues that users have perhaps I can get them to engage 3 times per week instead of 3 times per year. I can add value to Application 72 and instead of end users engaging for 2 minutes per session I can increase their engagement to 10 minutes per session.

Posted in User Adoption, Visualization | Tagged , , | 3 Comments

Visualizing Length of Stay

What do the numbers 3.53, 17.6 and 4 all have in common? 

They are completely useless when displayed by themselves because they have no context.  

Length of Stay is a vastly important metric in health care and here is the most common way to display it. AVGLOS 

Perhaps you can make it prettier using a gauge, an LED, some giant sized font or some really out of this world java extension but will that really change the fact that it’s basically a meaningless number without context? 

So often in the health care field we are so starved for data we can’t wait to slap the values on the screen and then start slicing it and dicing it before really thinking through the more basic question “What value does the number actually have?” Prettier isn’t better … it’s just prettier.

Average LOS is a real number that truly represents our average LOS. But does average length of stay truly represent how well we are doing? Is it fair to compare our average length of stay to anyone else? Is it even fair to compare the average length of stay within our organization between time periods? What about comparing the average length of stay between specialties?

LOSBySpecialty

I submit that any comparison of the average length of stay is like comparing the size of pizzas to the size of chocolates. One is much bigger than the other but who cares … we expect it to be. Just like we would expect the average length of stay for obstetrics patients to be less than the average length of stay for cardiology patients.

But LOS is important and the purpose of analytics is to measure where we are and help us find areas that need improving so comparisons are only natural. So how can we go about visualizing the length of stay in a meaningful way that doesn’t involve comparing pizza sizes to chocolate sizes?

Visualizing Length of Staying using CMS for the context

The answer lies in the fact that the CMS already publishes a guide for what they consider the proper pizza and chocolate sizes to be. I mean they publish a guideline of the expected length of stay numbers by MSDRG. [Click here to go to the CMS site so you can download the 2015 CMS MSDRG metrics] I’m sure you are not at all surprised to find that the expected length of stay for the obstetric MSDRG codes are much less than those for the cardiology MSDRG codes. We can then compare our length of stay numbers to the expected guidelines on case by case basis.

Once we download those guidelines and load the data we can simply compare our pizza sizes (LOS) to the expected values and say “Our pizza is that much higher or lower than the expectations.” Then we can visualize the results in terms of the difference in days. We are either under, right at or over on each case and we can display that average.

AVGLOSDifference

Now for another basic analytics question. Is average even the right thing to display? Our average includes outliers doesn’t it is that really what we want? When you pull the CMS expectations you will notice they provide you with 2 different numbers. One is the Geometric Average Length of Stay (GMLOS) which already has exclusions applied and one the Arithmetic Average Length of Stay which is pure average. The GMLOS is typically the more valuable of the two numbers and if we compare our numbers to that but then include our outliers we aren’t really doing ourselves justice.

The better plan for us would be to also devise a way to eliminate our outliers. Rats my data scientist is off today, so I can’t go that route. Instead lets simply take our median difference.

LOSDifference

Kind of subtle unless I drive your eyes to it but focus on the Obstetrics specialty and notice that in the raw average length of stay numbers it is by far the lowest and we might choose to ignore it, but when you look at the median difference between our length of stay and the CMS expected GMLOS you will see it is actually the only one over the guidelines.

Folks how we go about visualizing length of stay data, or any of our data, really matters. We have to think carefully about what we are presenting for our directed analytics applications or we will have our analysts spending time evaluating how to improve the wrong things all together.

Now let me challenge you again with another question … is the difference in days from our length of stay to the geometric mean length of stay numbers from the CMS even the right number to be evaluating? If our pizza is a full inch wider than an 18” pizza and our chocolate is only .5” larger than a typical 1” piece of chocolate which are we actually furthest away from the goal on?

The chocolate! It’s 50% larger than the expectation while the pizza is a meager 5.5% larger than the expectation. Quite frankly I’m happy about that as I love chocolate so you can’t convince me that 50% larger is a bad thing, but when it comes to measuring ourselves against the CMS expectations 50% is much worse than 5.5% isn’t it?

Here is the beauty of using the % comparison … it fits perfectly with a gauge. You know those beautiful looking, eye catching, chart types that 99.9% of the time shouldn’t be used. This is perfect opportunity for us to use them. The center of the gauge represents the fact that we would be perfectly in synch with the CMS guidelines while anything under means that we are doing better than the guidelines (green) while anything to the right means that we are over the guidelines (red.)

MedianDiffPercent

Now we are really cooking. At the beginning we had a bland simple number 3.53 days which told us absolutely nothing. Now we have a drop dead gorgeous way of visualizing length of stay and using the CMS guidelines as the context for our comparisons. 3.53 days means nothing because we don’t know if we sold 100 pizzas and 2 pieces of chocolate or if we sold 100 pieces of chocolate and 2 pizzas. But a % difference lets us compare how we are doing and is adjusted on an item by item basis.

Drilling into the numbers

But wait there is more.

Once we have narrowed down the service line, the physician group how do we drill even deeper to know where the problem is? Might I suggest using what statisticians use … a box plot. A box plot is a wonderful way to visualize complex data that enables you to see the ranges. You can quickly see the median (black line) of your data, the values from 25 – 75% of your data (blue) as well as the range of the outliers in your data what is referred to as the whiskers.

BoxPlotPercentage

Assume that the encoded values represent the true MSDRG’s for Cardiology visits. Notice that while the median for 0909 is higher than the median for code 0268. That’s 1 piece of information for sure. But is it equally important to see that the range of values is much more tightly grouped and that it’s median value lies pretty much right in the middle? Is important see that a full 70% of our values are just barely above 0, while the top 25% make up a very scattered range of numbers?

Sold on the idea of visualizing length of stay numbers in terms of a range, but aren’t loving the box plot format? Yeah me neither. The following depicts a box plot but with options changed so that it simply looks like a typical bar chart which is much easier for the average analyst who isn’t familiar with box plots to consume.

RangeOfValues

Notice in the chart above that code I340 shows a very interesting and telling range. 50% of the values are very tight together and are just slightly below the CMS guidelines in terms of # of days, while the remaining 50% range to nearly 50 days over the CMS guideline.

The value 3.53 has some meaning in that it truly is the average length of stay. But in the big picture the purpose of data visualization and anlytics is to provide a much deeper context than any single number can possibly portray. Our job as architects/developers/designers is to provide as much meaning as possible in as many ways as it takes for the analysts to consume the true picture and adjust processes to help the organization improve. Visualizing length of stay just happens to be a really complex metric to do that for.

 

Posted in Visualization | Tagged , , | 2 Comments

Visualizing the Story

“It was the best of times,

It was the worst of times,

It was the age of wisdom,

It was the age of foolishness …”

These are the unforgettable opening lines to Charles Dickens’ classic “A Tale of Two Cities.” One of the first pieces of classic literature I ever read. I had just entered the 10’th grade and our teacher loved this book so much that we literally spent half of the entire year just reading and reviewing this book.

I began thinking about this book after viewing this piece of artwork from Stefanie Posavec.Prints3 The complete piece of art represents a chapter by chapter, paragraph by paragraph, sentence by sentence and word by word depiction of part one, of one of her favorite books. She quite literally visualized the story.

If I had a fraction of Stefanie’s creativity, the graphic art capabilities or the time to truly do it justice I would love to visualize “A Tale of Two Cities.” I’m more of a people person than a words person so I would probably begin with a wonderful Chord diagram that mapped all of the characters to various character traits. How many traits would Charles Darnay, Doctor Manette and Madame Defarge have in common?

SankeyDiagram

No wait the main point of the story is the incredible transformation in characters like Sydney Carton. What great paths a Sankey might divulge if used to map Sydney’s transformation from a lazy alcoholic into a selfless martyr. Where do his changes intersect with others in their development?

Sorry I digress … the whole point of sharing those opening lines wasto describe the period in which we find ourselves, not theirs.

We have petabytes of data available and yet business users can’t access a fraction of it.

We have examples of truly great work, ridiculous amounts of computing and graphic horsepower at our fingertips and yet we can’t build data visualizations that business users want to use despite being starved for data.

I have a confession to make

I recently completed a webinar in which quite frankly a fear raised its ugly head and I backed down from it. I was demonstrating what I had worked on with Qlikview and some of the super cool new functionality of Qlik Sense. Not a problem. I’m very comfortable talking about my data and my work.

My fear came through in my choice to not present one of the coolest new features of Qlik Sense which is “story telling.” I wasn’t really sure how to frame the graphics I was showing as a story that would be compelling. So I chose to simply avoid the issue.

Im pretty confident statistically that there are lots of others who are very much like me.  We can import data from cockail napkins, we have incredible tools at our disposal but what we don’t have is the background in story telling.

Quite honestly I’ve avoided learning the art of story telling because I’ve felt like presenting my “view” of the story is against the rules. Aren’t we just supposed to share the underlying data without any of our own prejudice?

The data should just speak for itself shouldn’t it?

It’s the readers responsibility to know what they want the data for isn’t it?

Everything I’ve read in the past few weeks since my presentation has told me … NO.

Everything I’ve read or listened to lately indicates that great info graphics and great news stories have one thing in common … the author presented and guided the story from their point of view. They don’t alter facts to hide other views. They simply provide a lead or direction to the story to ensure that they have at least presented a path for the reader. If the reader chooses to dive really deep they can.

Local Hospital Workers

Battle 5 Headed Monster

Is my take on how our colleagues in data journalism might approach the very situation that many of us in the healthcare field find ourselves. Nearly every day we battle:

  • Meaningful Use and a myriad of other governmental regulations
  • Converting more and more processes from paper to electronics
  • Lessening payment percentages from the government and insurance companies
  • Increasing number of patients with super high deductible insurance plans
  • An aging population that has multiple complications
  • A rapidly growing antibiotic resistant population

Ooops! That’s 6 heads. Unlike data journalists I don’t have an editor to check my facts. Don’t let that cause you to miss this really important point … the data visualizations we build should have attention grabbing headlines to them.

For many many years data journalists have set the example of how to draw readers in and yet we struggle and sulk at the end of each day because with all of the resources available to us we don’t have the kind of adoption rates for our applications that they get with a pen and paper and brute force in gathering their data sources.

What I’m learning tells me that the next big step is to back up those headlines with key points that keep the audiences attention … like the 6 that I mentioned. Those points should then be able to be drilled into until they exhaust the data you have available. The path that your users choose to take is up to them. That’s where your directed story changes course and becomes truly interactive and user guided.

The main point I’m trying to make is that regardless of the medium we should all be visualizing the story. The vast amount of detail which we can provide in our data visualizations should be the only thing that separates us from data journalists who are bound by a limited amount of space.

My own metamorphosis may not be quite like that of Sydney Carton.

My work may never draw the kind of audience that Stefanie Posavec has.

But my friends my eyes are being opened widely to the great synergy between data visualization and data journalism. To anyone who tries to stand in my way as I continue to learn this great craft of story telling I say … “Off with their heads.”

Posted in User Adoption, Visualization | Tagged , , , , , | Leave a comment

Visualizing the Busy-ness

BoomNot that anyone would be surprised to discover a direct correlation between supply and demand but in my last post on “Visualizing Knowledge” the scatter plot proved to be a very advantageous chart type in that it showed there was a very distinct correlation between the two. On the busiest days (Monday) it took longer on average for a patient to transition from walking in the door to being placed in a bed.

The scatter plot was a good visualization choice to help us quickly and effectively see the relationships by day of week. The goal of this post is to try and drive deeper because while “day of the week” is a common unit of measure in reality that is a rather large unit of measure that is comprised of 24 individual hours.

More specifically we want to find a way of visualizing the busy-ness hour by hour. Just for fun let’s challenge ourselves to provide what Albert Cairo refers to as the “boom” effect in his book “the functional art.” In simplest terms we want the graphic to show up as a visual pyrotechnic. I want our visualization to explode off the page into the readers mind.

The natural starting point I suppose would be a pivot table. It’s a simple way of visualizing a measurement like “# of people who walk in the door” across multiple dimensions like “Day of the Week” and “Hour of the Day.” Easy peasy right?

PivotTable

Easy peasy to create perhaps, but is it really so easy to read? A pivot table may be the perfect choice for multi-demension analysis when both dimensions have few values. But in this case we have 168 unique cells and it is all but impossible to spot any immediate patterns. Fortunately I specifically indicated that we needed to provide a “boom” factor in our visualization or we might have been tempted to stop and simply say “look Mr. E. D. Director you asked to see 168 values and I showed them to you.”

In his afore mentioned book “the functional art” Alberto Cairo spends a great deal of time explaining the science behind how we visualize anything as humans. In one section he uses an image of what could be a pivot table and says “The brain is much better at quickly detecting shade variations than shape variations.” The point he was making is that its nearly impossible for humans to see that many numbers side by side and on top of each other and make them out. In the following try and find all of the 6’s :

4 3 6 9 1 6 5 7 8 2 4
9 8 4 6 3 2 1 9 5 3 1
7 2 8 1 4 5 9 6 7 3 1
2 4 1 5 6 8 1 4 2 5 3

He then shows an alternative image with the exact same sequence of numbers but uses shading and suddenly what we as humans can do is made abundantly clear:
4 3 6 9 1 6 5 7 8 2 4
9 8 4 6 3 2 1 9 5 3 1
7 2 8 1 4 5 9 6 7 3 1
2 4 1 5 6 8 1 4 2 5 3

Cairo immediately goes on in his book to describe the Gestalt theory that human brains don’t see patches of color and shapes as individual entities, but as aggregates. How can we take advantage of that? We need that kind of impact for Mr. E. D. Director. We want him to be able to immediately visualize the busy time periods but avoid all of the busy-ness of 168 cells with numbers.

You probably guessed that we don’t want to use a 168 slice pie chart.

Nor do we want to use a line chart with 7 different lines each with 24 points.

What we want is affectionately known as a heat map.

HeatMap

There can be no mistaking the busiest hours across all 168 unique cells.

There can be no mistaking the obvious patterns either.

The purpose of the heat map is to allow you to visualize your data in a way that makes the data explode off the screen and into your users heads.

In the previous post we used the scatter plot to determine the pattern between the busiest days of the week and the time it took to get a patient from the door into a bed. So let’s keep driving into that analysis.  The neat thing about Qlik Sense is that it allows the end user to simply drag/drop a different measurement onto our visualization and voila they now see a heat map of the median minutes to bed for patients.

HeatMap2

If the problem were simply one of a lack of beds we should see a pretty close relationship between the two. But that isn’t what we see at all is it?

How come there are obvious patterns of when the busy time for arrivals can be absorbed?

Why is their a pattern across all week days where time to bed slows down around 6 PM and remains the slowest from 7-10 PM?

Those questions are what drive true data analytics.

Are you helping your users not only answer the questions they came into the application to ask, but also providing them those answers in a way that leads them to ask even bigger questions?

Questions that involve processes and not just data.

Questions that involve staffing models and not just data.

If we can provide them the answers to those questions in any manner we did our job.

If we can provide them the answers to those questions in a way that removes the “busy-ness” of all of the data from the important data we are trying to visualize and do it all in a way that provides a “boom” effect we can certainly take 2 seconds to pat ourselves on the back before digging in to the next problem.

Posted in Visualization | 5 Comments

Visualizing Knowledge

BedI love it when a plan comes together.

For the past several weeks I’ve been reading furiously and working like a dog doing my day job, while also trying to come up to speed on Qlik Sense as quickly as I can. Today everything seemed to come together so well I felt like I just had to share.

My first project in Qlik Sense involved trying to replicate as much of the functionality of our Emergency Department Dashboard as I could. If you’ve ever been to an Emergency Department you probably dread going not just because you are in a crisis but there is often a fear that you’ll “be waiting forever to be seen.” So one of the major metrics used to evaluate Emergency Departments is the time it takes to get the patient into a bed. An arrival to bed time of 15 minutes means it took 15 minutes from the time the patient arrived in the ED until they were taken to a bed. That 15 minutes would include the time to register you when you walked in the door, find out what you believe is wrong with you, determine the severity of your condition and find a location to put you.

Naturally it only made sense for me to begin visualizing this “Door to Bed” metric with Qlik Sense and then work up from there.

This week one of the really fascinating articles I read was an interview with David McCandless about his new book “Knowledge is Beautiful.” His previous work was titled “Information is Beautiful” so naturally the interviewer was asking him questions about the difference between “information” and “knowledge.” He responded with “… what I’ve discovered through my work is that data is granular, information is clusters and bites of something more structured; and knowledge is something deeper and richer, more interconnected and complex.” In the article David shared additionally that in the first book he had shared singular visualizations, but in his new book he couldn’t stop at one graphic because he wanted to answer all the key questions and address all the aspects of a given subject so that it would knowledge.

On the dashboard for this new Qlik Sense application I already displayed the simple “Median Minutes from Door to Bed.” While a critical piece of the puzzle it’s only a very small part of the bigger picture and would be considered “information.” Like David I wanted to go deeper.

I chose to begin my exploration with a scatter plot that would show the Median Minutes to Bed across the Number of Arrivals and the Day of the Week and this is what I came up with.

TotalArrivals

Looked really neat but as I began selecting various Months I realized what I was displaying was very misleading. The information was totally correct but I happen to know that our busiest day of the week by far is Monday. Yet because I had chosen a month that happened to have more Wednesday’s, Thursday’s and Friday’s the total number of arrivals for those days was greater. It’s only logical that if the ED is really busy the time to get you into a bed would be greater right? But total number of arrivals over the course of a month doesn’t really equate to busy when the number of those days of the week can vary. Rats!

Here is one of the beauties that I’ve already discovered with Qlik Sense … it’s so easy to drag and drop or choose a different metric that it was nothing for me to choose “Average # of Arrivals” instead of “Total # of Arrivals” so that the chart could depict the “busy” aspect of the knowledge I was trying to visualize and as I was hoping a clear pattern emerged. Something that wouldn’t necessarily jump off the page if I had chosen a pivot table to show the raw data. There is a clear correlation between how busy the ED is and how long it takes to bed patients.

AverageArrivals When I removed the date filter another interesting thing jumped out at me. Or I suppose in this case it would be more fitting to say something disappeared … Friday.

OverallAverageArrivals

Turns out that with no date filters at all the average number of arrivals per day and the median minutes to be bedded are identical for Thursday and Friday. Not that I like having the chart miss a day because certainly users would be flustered. “Oh yeah I can change the definition of the measurement in 5 seconds to include a single decimal and voila there is a slight difference.”

Here is where everything kind of came together. David’s comment was that he couldn’t stop at a single visualization so I didn’t want to stop at a single visualization. In one of the key note sessions at the Qlik World Conference Stephen McDaniel said one of his favorite ways to visualize data was using bar charts that were synched up. I loved the concept but more often than not I’m dealing with dimensions that have 100 or so options and synching bar charts isn’t easy. Earlier this week an article was posted on Twitter again with that type of visualization and I retweeted that I couldn’t wait to find a great fit to use that. Boom here it was in front of me … there are only 7 days in a week. That can all fit in a simple chart without scrolling so I had my use case.

If knowledge is richer more interconnected then what else could I show along with the Median Minutes and Average # of Arrivals to help my audience? I chose to show the % of patients that chose to leave the ED without being seen. Our LWBS numbers are something that we also track and measure closely.

SynchedBarCharts

Would it be odd of me to say that my heart fluttered when the chart above came up.  Do you see it. It’s beautiful. It shows that while the Average # of Arrivals and Median Minutes are identical (when rounded) for Thursday and Friday there is a clear difference in the % of people that left without being seen. But why?

I honestly have no idea as of writing this post. For my learning though that question is secondary. I had just tied pieces of information together in way that displayed knowledge and more than that … in a way that immediately would lead the analyst viewing it to want to ask more questions. To dig deeper. To interact. Something that I’ve been reading about for a week in “The Funcational Art” by Alberto Cairo.

Like I said … I love it when a plan comes together.

Posted in Visualization | Tagged , , , | Leave a comment

Visualizing Lab Results

RealFishBones

Have you heard the joke about the lab technician that walks into the room to stick you with a 15” needle and draw your blood? Of course you haven’t that’s just not very funny stuff.

More than likely you are not as afraid of needles as I am, but I doubt anyone likes being told “I need you to roll up your sleeve.” Seriously do they really have to say out load “This is going to sting a little.” I pretty much guessed ahead of time that a long sharp object inserted into my arm would sting a little.

On a serious note though despite the anxiety I feel I am 100% aware of the very valuable insight that lab results provide physicians about my health so I reluctantly tolerate the “sting” and try not to cry until I’ve left. I tell myself I’m brave because I know that there are scaredy cats out there that don’t even see a physician due to their fear. When you are in the hospital it’s not as easy to hide from them though. The lab techs are generally at your bedside very early in the morning, more than likely waking you up, in order to remove a gallon of your blood, or however much those vials hold.

Your blood is then rushed to the lab where a technician runs the myriad of tests that the physician(s) have requested and then the results are generated. Until someone does something with the results of the lab tests the values are simply like all of the 0’s and 1’s that reside on our disks … useless.

In this post I’m going to discuss how I handled visualizing those lab results for the physicians rounding report I’ve been working on. I began my work on lab results the same way I would anything … “let me see what kind of data and what volume of data I’m dealing with.” LOTS and LOTS. I’m not kidding. It’s almost like every single drop of blood holds 1 MB of data or something. The following is for just the last 3 days of lab results for 1 patient.

AllResults

You didn’t like having to scroll all the way down here to finish reading did you? It gets worse … I want you to scroll back up and figure out what the most recent lab result value is for HCT. Painful isn’t it. Certainly we are not going to deliver that as our lab results visualization.

Let’s try a little harder

We can do so much better than just “displaying the data.” At minimum we can use the nifty little QlikView function PEEK when we load the data and rank the values by which is most recent, which is second and which is third.

Then we can at least provide a display of just the ‘Most Recent” data to save the physicians from trying to figure it out. We might even offer some color coding for values that the lab has indicated are too high or are too low. By doing that we take the 150 values above and pair it down to a much prettier 42 values.

MostRecentResultsColored

Gotta tell you it did take some time to convert from a table box to a chart so that I could do the color coding but that’s our job. Unfortunately I completely forgot about that whole “context” thing I mentioned in my post on visualizing blood pressure  and while making it easier to read and prettier I took away their ability to know how this most recent result compares with the 2 previous results.

Putting some polish on our chart

What the heck … we’ve got a few more minutes before lunch let’s really go out on a limb and change our straight table into a pivot table. Then the physician will be able to see how the results compare. If you refer to the pivot table below you’ll see how important that is for the value “BC” as the physician will clearly be able to see not only that it’s out of bounds but will also be able to see that the trend is getting worse not better.

All3ResultsSorted

Don’t pat yourself on the back just yet my friend … there are still 42 values that a physician would have to sort through. Not to mention the fact that I used green to indicate a low value and red to indicate a high value. Hello color blindness issues that I forgot all about. But you shouldn’t … click here for a really cool color blind simulator that allows you to upload your image and see how those with various forms of color blindness will see it.

Not to mention the fact that green carries a positive connotation and perhaps the peril with a particular value (thus wanting the negative connotation of red) comes with a value below the expected range. As they say “no good deed goes unpunished.”

I spoke with our Chief Medical Informatics Officer and expressed my displeasure with the whole process. First I haven’t even liked thinking about having blood being drawn and I just couldn’t find any really good way to visualize the lab results.

No matter how I sorted it you had to dig.

No matter what format I changed the charts into tradeoffs ended up costing me valuable information.

It just seemed that if I was the physician and was rounding I would want something better when I was standing next to a patient’s bed. I would want something that let me immediately look for the values I cared about based on the patient’s particular conditions. If I needed to see a particular result or few results I would want a way to immediately see exactly what I wanted without having to scan the other 41 values.

Going Old School

He agreed with me and after some thought walked to his white board. He started with the 5 words he knows really motivate me “you probably can’t do this” and continuedas he started drawing “in the old days when we had paper charting we used Fishbone diagrams…” He went on to describe how they used to use drawings that looked like fish bones and different shapes contained certain sets of values. The “key” values from a basic metabolic panel for instance looks very much like a small fish. The values from a complete metabolic panel look like a bigger fish. While the results from a Complete Blood Count (CBC) form a different shape.

Imagine that … before we had tons of computing power, had millions of rows of lab results to display in grid form or had access to 18,393 books on data visuzalition they were visualizing lab results in an awesome visual way. Simple. Clean. Immediately recognizable. Below I have just 2 of the several forms of fish bone diagrams so you can see the technique. The numbers on the left are the most recent, while the ones inside parenthesis are the previous value. In the top diagram you’ll notice that if either of the values is out of bounds (high or low) I’ve simply enabled the color of the border around the text box. More subtle than color as it blends into the diagram pretty well if you are looking just for Glucose for instance. But if you are looking for their Chloride level then it immediately hits you that the numbers are out of bounds. I’m not as thrilled that the same technique doesn’t work as well for the other diagram shape. Good thing you can comment and give me suggestions.

Visualizing Lab Results

Using the PEEK function I was able to isolate the most recent three lab results. Combining that with SET ANALYSIS I was able to choose which to add to my old school visualization. Without much effort you can do the same for as many previous results as your physicians might like. Guess I better start spending more time with these old timers that wrote things down on paper.

Alright I gotta run … I’ve got a whole bunch more 0’s and 1’s to dump on my rounding report, I mean visualize in a way that will lead to our physicians ability to quickly consume them.

Posted in Visualization | Tagged , , , | 3 Comments