Think Like A Data Analyst — Act Like An Archaeologist

Anyone who has read this blog will probably find clues suggesting that I am a big Dan and Chip Heath fan (and by fan I mean “groupie”). I am also a big fan of Nathan Yau at Flowing Data.  He has a consistently good blog with great content and useful ideas about how to visualize data. I have meant to build on one of his blog postings for a while, and here it is. (I hope he also feels that ‘Imitation is the sincerest of flattery.”)

Yau called his posting “Think like a statistician–without the math.”  I am going to use his outline to make a slightly different point: think-like-an-archaeologist-even-though-you’re-an-analyst. At least as important as understanding database structure or business intelligence software or sophisticated statistical methods are several other skills not often taught in college statistics class (or in Analytics departments at large corporations). We could start with the maxim to Dig. Dig. Dig.–but there is more.

1. “See the Big Picture”: Archaeologists sift through massive amounts of material not to find some pieces of broken pottery but to understand civilizations. This curiosity about the import of fragments of information is arguably the most prized attribute for an effective health care data analyst. Identifying a large increase in musculoskeletal costs is interesting. That’s reporting. Analytics goes much further … like this: 1) Go back to the data and discover this increase was driven by a large number of back surgeries, 2) then link that finding to the fact that there is a large percentage of overweight/obese members, 3) then notice that a very low percentage of members were participating in the available weight management program, 4) then identify that only a few members had engaged with the health coaches around “elective” surgeries. This “path” through the data is analytics — and analytics is often a journey of discovery.
2. “Look Outside the Data”: We are talking about applying context and perspective. For an archaeologist, this is exactly the point–to place what they find alongside what they already know about the other historical events. In healthcare, sometimes the benchmark data is even more valuable the client’s own data. Knowing that an Employer’s average hospital days per 1,000 is 350 is not nearly as informative as knowing that the Employer’s hospital days per 1,000 than is 32% above their industry peers.
3. “No Agendas”: It is important to have a hunch when you start looking for meaning across multiple health care data sets. It’s OK to start with a strong belief that diabetes or heart disease are key drivers of trend (because they are). However, it is important to let the data speak for itself. Unexpected and valuable findings will emerge as you set aside the assumptions and look for patterns, correlations, and outliers. (For an interesting discussion of archaeologists as detectives uncovering new patterns and rewriting history, listen to this podcast at RadioLab about Oxford teams unpacking ancient garbage.)
4. “Ask Why”: Letting the data speak for itself is important; however, that does not mean that having a sufficient amount of cynicism and mistrust for the data is bad. Listening to your “gut” doesn’t sound like a valuable skill for a data analyst; however, there have been many, many times when I’ve questioned the accuracy of a particular health care measure and, as a result, discovered a major data error or wrong calculation. It’s easy to be captured by a particular find, and misunderstand its significance (or lack thereof). It’s just as easy to forget that the data isn’t necessarily “innocent” and must always be contextualized, tested, and cross-checked.

Health care data analysts may never be mistaken for Indiana Jones (although if we could carry to work those really cool whips it would help). The information is like treasure, though, that leads us to rich veins  or patterns in a population and helps us fend off rising healthcare costs and poor health.

Report from Integerated Care Forum (DMAA Conference)

I am speaking to a group in Reno next week about ways to measure the efficacy of Wellness programming, and I recently had the opportunity to attend the Care Continuum Alliance (formerly Disease Management Association of America (DMAA)) in Washington DC.  Basically, I am all about in Care Management thinking this month.

For those who did not attend the DMAA conference, these are the four themes I heard:

Niche Branding. In the “olden days,” vendors supplying care management used a scattershot method for distributing wellness programs: shoot a lot of stuff out there and hope some of it hits the target. Now the large vendors are looking to package and brand themselves in the marketplace.  Some are focusing on members with very high risk (e.g., APS Healthcare).  Others are selling themselves based on their ability to “touch” a broad population segment (e.g., Health Dialog and Alere).  What interests me is how these organizations will combine/modify current offerings to take advantage of emerging technology and the changing health care system. 

Engagement. In a recent McKinsey Quarterly article, authors Sundiatu Dixon-Fyle and Thomas Kowallik noted that 80% of heart disease could by avoided by observing the wellness Holy Triad: regular exercise, healthy eating, no smoking. But we know by looking at benchmark (and our neighbors) that most people don’t do these three things. Why not? Here’s what Dixon-Fyle and Kowallik say:

Research shows that this disconnect occurs because people often circumvent the steps of rational decision making when faced with choices. They sometimes make decisions without considering all the facts, by using irrelevant facts, or by making faulty calculations based on the facts they have. …
Understanding how these factors prevent change is a key starting point for payors, but one they often have only limited experience with. In addition to gathering information about their members’ health status and clinical risk factors, payors need to apply consumer-insight techniques from the field of marketing to understand and segment consumers by their motivations and barriers to change. Using these insights, payors can then design appropriate consumer-engagement and behavioral-change interventions, and target them to appropriate members.”

Several key note speakers and many of the concurrent sessions addressed the engagement problem. Happily, there are some interesting, innovative approaches emerging in the marketplace that take combine social media technology and social behavior theory–specifically workplace exercise competitions supported through online communities and measurement technologies (like the Virgin HealthMiles Pay for Prevention program) that allow members to see and measure their own activity and receive reward for achieving specific goals. (For an interesting look at analyzing personal metrics see “Know Thyself,” from Wired magazine.)

Device Proliferation. That brings me to the Exhibit Hall where I was astonished at (assaulted by) the show of new devices for measuring and enabling healthy behaviors. Many of the vendor booths demonstrated new devices aimed at facilitating data collection for those engaged in care management programs. Most of these devices were (very) cool including glucometers embedded in cell phones and the pedometers with USB ports to facilitate health data collection.

Focus on Analytics: One of my mantras is “You can’t manage what you don’t measure.” Measuring and reporting personal data seems to motivate the behaviors for which everyone is looking. So, the idea that the data is important for making “change” was everywhere. Data also showed its star power in some recently released studies underscoring the importance of wellness programming. For years academics, consulting firms, and research organizations have generated studies attempting to prove once and for all that care management programs not only pay for themselves, but deliver a positive ROI. (See a particularly good response to the ROI question by SH Woolf. ) Speakers spoke confidently about their studies “proving” that the question was resolved.

Stay tuned.  There are going to be many more changes to Population Health Management in years to come.

Data Delivery Is Not an “All-You-Can-Eat” Buffet

I had the brilliant idea recently to take my family to the “Old Country Buffet” for their All You Can Eat dinner buffet.  I remember going when I was younger and it was my Grandpa’s favorite place for a special meal.  My first plate was satisfying (or as satisfying as could be expected).  Plate two was not so good, and by the time we got to the ice cream sundae bar…. <wince>  This was a classic quantity over quality problem: it was all too overwhelming to find the good food among the canned and rewarmed, and I was taken in by the magnificent display of big carbos. Frankly, I didn’t choose my calories or my portions very wisely.

Some health plans treat their customers like their data delivery is an All You Can Eat buffet.  They send long reports filled with traditional cost and utilization measures sliced five different ways. They offer unlimited self-reporting to their clients–and the clients (or their advisors) dutifully fill their plates with reams of the-data-equivalent-of big carbos. While most employers may come away feeling full, they’re not necessarily satisfied.

I spent part of a day recently with the CFO of a well-known health plan.  We were talking about the importance of telling compelling stories with their health and productivity data to communicate value. I shared with him my ideas for fixing their externally facing Employer Group reports as well as my ideas for improving the production process. I was surprised at how quickly he engaged. He totally got it.

Here’s why he knew what I was talking about: He told me that when he took over as the CFO of this health plan, his staff handed him a 90 page quarterly report — built manually in PowerPoint — reviewing information they thought was relevant to his role.  It was a massive data dump. Over the next few months he tried unsuccessfully to make modifications to the report to make it more useful to him.  Eventually, he scrapped the old report completely and built a new report including ONLY the pieces of information that were actionable in his day to day decision-making.

This move to pare down and rethink is an example of the approach Jason Cohen, founder of Smart Bear Software, takes on fixing a business:

“Your assumptions are mistaken; are you testing them? Your gut knows your logic isn’t airtight; are you listening? Your current ideas lead you to better ideas; are you adopting them?  Success doesn’t come to those with the best ideas, but those most willing to seek better ideas.”

Most Health Plans need to clean up their reporting. They need to create something leaner and more nourishing. They need to re-examine the information they are sending and ask: is this useful? is it meaningful? can my client use this information to make better decisions? and can they make better decisions today?  Can we do this better?

We are going to try Chipotle this weekend. I don’t have quite as much choice, but  in the trade I get recognizable ingredients, custom built to meet my needs, with a cost and time efficient delivery.

Concrete Numbers: Dragging Data into the Everyday

At the risk of looking like a Dan and Chip Heath groupie, I liked their posting about data this month in Fast Company. To illustrate the importance of attaching data to specific experiences, they work on a statistic from the film Super Size Me comparing the ratio of McDonald’s advertising budget relative to that of “eat-healthy” campaigns.  Ok, good enough, they say, but watch us make a “statistic” meaningful:

So suppose your 5-year-old daughter watches three hours of cartoons every Saturday morning and sees two McDonald’s commercials per hour. Every Saturday, then, Ronald McDonald engages her six times.
How long will it be before your daughter sees a fruits-and-veggies commercial? She’d wait about 14 months to see the first one, and she’d have a driver’s license before she saw 10 of them (the same number of McDonald’s ads she’d see in two Saturdays).

Wow.  Now I really get it and I understand better what is at stake (and what conversations I need to have with my kids begging for french fries).

This thinking is powerful when applied to healthcare data. Most of the time we are talking about very large numbers and rates of increase that don’t sound overwhelming–8% rise? not too scary–when taken in the abstract.  But what if we put that into specific context?:

In 2008, ABC Company has $20 million in fixed costs, of which half is health care. Health care costs rise 8% annually until 2013, reaching nearly $15 million. Holding other expenses flat over this time, ABC must cover an additional $5 million. Say it sells a widget with a contribution margin of $50 per unit. ABC must sell an additional 100,000 widgets in 2013 compared to 2008 just to keep the same profit.

Here’s another example:

Joe, one of the most experienced shop foremen, has a heart attack while watching the Redskins lose to the Lions.  The ambulance ride and his 5 day stay costs $20,000 in medical claims.  He takes 4 weeks of disability and then takes two weeks to transition back to full time.  All in all, the $20,000 in medical costs ends up being dwarfed by the additional costs (estimated at $80,000) including sick-pay, disability pay, lost productivity, and overtime pay to cover replacement workers.

Chip and Dan Heath provide the takeaway:

A good statistic is one that aids a decision or shapes an opinion. For a stat to do either of those, it must be dragged within the everyday. That’s your job — to do the dragging. In our world of billions and trillions, that can be a lot of manual labor. But it’s worth it: A number people can grasp is a number that can make a difference.

In our capacity as people who examine health care data for non-healthcare data-experts, it’s our job (a critical job) to create specific applications for our clients and/or members. The numbers we look at matter, and they need to be presented to our clients in ways that can make a difference.

Exceptional Presentation: “Let My Dataset Change You Mindset!”

I’ve written many times in this blog about how data, presented effectively, can change the way we see problems and think about solutions.  I found an exceptional example of this in a lecture given by Hans Rosling, a professor of global health at Sweden’s Karolinska Institute, to the US State Department this summer.

In case you are not familiar with Rosling, this is an excerpt from his profile on Ted.com:

What sets Rosling apart isn’t just his apt observations of broad social and economic trends, but the stunning way he presents them. Guaranteed: You’ve never seen data presented like this.By any logic, a presentation that tracks global health and poverty trends should be, in a word: boring. But in Rosling’s hands, data sings. Trends come to life. And the big picture — usually hazy at best — snaps into sharp focus.

(Click here for link to presentation.)

Rosling’s presentation shows us overall global trends in health and income across the last 200 years.  His innovative data-bubble software–called  Gapminder–exposes common myths about health care and the developing world. It is an amazing combination of statistics, humor, and subject mastery. If you are like me, you won’t be able to watch it only once.

This presentation is a perfect example of how data, when mined and combined with innovative design, generates learning and knowledge that changes minds and makes a difference.

*Mr. Rosling’s presentation is part of the TED-talks, (TED: Technology, Entertainment, Design).  TED is an invitation-only event where the world’s leading thinkers and doers gather to find inspiration.

Update from Health 2.0 – San Francisco

By Dan West

First a quick definition: what is Health 2.0? Wikipedia tells us that “Health 2.0 is the use of a specific set of Web tools (blogs, podcasts, tagging, search, wikis, etc) by actors in health care including doctors, patients, and scientists, using principles of open source and generation of content by users, and the power of networks in order to personalize health care, collaborate, and promote health education.”

With all the recent media attention on health plan-related incompetencies, it was refreshing to hear from a few health insurance companies taking advantage of modern technology to better serve their customers. Health 2.0 can be an important method by which to educate members and encourage the proper and effective use of their health benefits. For example, the State of Minnesota has offered tiered networks for around 8 years now with layered incentives based on quality. Through Health 2.0 methods, Minnesota is able to better steer its customers to its most effective providers.

Mohan Nair of Regence BlueCross BlueShield emphasized the need to balance medical management with consumer engagement. Thirty percent of its employees use the web, and by leveraging Health 2.0 methods it is more able to “get the consumer to recognize that it’s his money we’re saving.” Chris Ohman, senior VP of health operations at Kaiser Permanente, reiterated this. Kaiser has identified five chronic conditions that contribute roughly 70% of medical costs. Managing those diseases, while partially the responsibility of Kaiser, rests mainly on the customer. By getting lab and test results on the web, it can better encourage its customers to manage their conditions more effectively.

Health 2.0 isn’t just a subset of Search (personalized results), however. Ingrid Lindberg of Cigna pointed out that her company is on Facebook, Twitter and Second Life, all with the goal of facilitating interaction with Cigna. Regence BCBS offers unregulated forums and blogs for customers to connect and air grievances. Rather than deleting negative posts, which would destroy the online community it’s building, it views them as valuable feedback. Negative feedback is just a sign a company isn’t serving its customers needs after all.

Health Plans and Data Mining — Look at Mint.com

Health Plans have been agonizingly slow to take advantage of the vast amounts of data available to them. Admittedly, they have some valid excuses. The barriers put in place to protect patient privacy and a lack of consistent data standards makes health data mining a formidable task. However, when I read recently that Mint.com is selling its two year old business to Intuit for $170 million, I couldn’t help but feel a little frustrated–and not just frustrated that I am not still 28 and selling a company for $170M.  After only two years, toddler Mint is aggregating data from its 1.5 million users and actively mining it for valuable information on consumer spending patterns.  This is a initial graph illustrating spending and saving trends. Click here for graph.

Looking octogenarian and flat-footed by comparison to Mint, Health Plans continue to struggle mightily to productively mine their data.  They don’t seem to be able to focus themselves long enough to develop the vision or lasting initiatives that will allow them to make use the vast amounts of data they have to work with.

Here are some reasons why Health Plans have such as hard time with this:

Lack of motivation: Health Plans have benefited  from controlling a market missing many of the typical economic (supply-demand) factors. Increasing demand for healthcare services combined with the insulation from true costs have resulted in an industry “padded” from economic realities. It doesn’t appear that the current efforts at healthcare reform will remove much of the padding any time soon. On the other hand, more patients are becoming proactive and accountable healthcare consumers through more powerful and useful tools, and Health Plans are facing new competitors. I can imagine a time when  Health Plans will be left to compete on their ability to deliver value.

Slow to learn: Health Plans have been slow to adopt strategies that would allow them to mine their own data and improve competitiveness with strong analytics. Some examples: provider contracting still occurs with limited data; predictive modeling to identify and stratify patients for better care management are typically weak and under-invested; and customer-facing reporting is often disjointed and lacks organized, consultative insights.

Data systems with clashing cultures — often mirroring organizational dynamics: Most Plans are struggling to get their data systems on the same page–literally. Data systems within health plans were typically designed in isolation of each other. The backbone of the current systems — legacy claims processing — was very transactionally focused–think Baby Boomers. More recent data innovations, including care management systems that emphasize capturing more free-form clinical information, are another generation–Gen Y. While it is possible to have productive relationships between the two, it takes finesse and patience.

Lack of standardization: Mint was able to accomplish its data mining feat because the financial services industry adopted and implemented data standards many years ago. When Mint sweeps data into its systems, the processes for integration is manageable. While the healthcare industry has taken strides in this direction, it has a long, politically sticky, way to go.

I know it isn’t fair, exactly, to compare a new bright start-up to an entrenched Health Plan, but its time for Plans to step up and make use of their competitive advantage: data.

What do copy editors and data miners have in common?

I learned long ago not to be too disappointed in myself when the editor returned my writing with more edited sentences that unedited sentences. After scoping the project, performing the necessary background research, getting through an initial draft, and agonizing through several rounds of re-writes, I am burned out. Fresh eyes and a different perspective almost always result in improved writing. In fact, good editing, or finish work, often requires different tools or skills. A very good editor inevitably breathes new energy into the writing process by solidifying tenuous connections and building out nascent ideas.

Most large scale data projects need the equivalent of a good editor.  In my work on health IT projects for health plans, implementation team burn-out happens virtually 100% of the time. When the data finally ends up in a data warehouse or data mart, everyone is tired and happy to take a long vacation from the data. They’ve worked around the clock to accomplish a massive amount of work and feel done. The frequent result is a very expensive, nicely organized data warehouse that is not used effectively.

That’s where experienced data miners (aka editor) can make all the difference. Just as there is a difference between being a writer and an editor, the skills required to transform health data into meaningful information is different than the skills required to build the data warehouse. Experienced health data “miners” look at the vast amounts of data and crystallize the “story” told by the data.  They cut away all the superfluous data and draw out what is important to the end user. Just as a good copy editor often clarifies a point by shortening the text and improving word choice, good data analysts communicate key messages from the data using skillfully designed charts, graphs, and consultative information.  Data visualization goes a long to effectively communicating key messages.

Getting value from a newly established health data warehouse defies the 80/20 rule.  With new data systems, 100% of the value comes from the last percentage of time spent culling out and communicating what is important in the data.

If you are a writer, find a good copy editor.  If you are in the middle of a health data warehouse implementation, get a good data expert to write the data story.

Wanted: Data Geek/Visual Designer

I was motivated to get back to blogging by a recent post from Michael E. Driscoll.  (See his great mashup of scatter plots with Andy Warhol’s Marilyn Monroe.) Driscoll takes a quote from a McKinsey Quarterly interview with Hal Varian, Google’s Chief Economist:

“The sexy job in the next ten years will be statisticians… The ability to take data—to be able to understand it, to process it, to extract value from it, to visualize it, to communicate it—that’s going to be a hugely important skill.”

I’ve never really thought of myself as someone with…well…sexy skills.  (I immediately sent this blog to my wife, who also has never really thought of me as having sexy skills.)  I remember in college walking from my advanced statistics class in the math building to my pottery class in the art building.  I remember hoping that my earthy pottery friends wouldn’t see me coming out of the math building and vice-versa. But the need to live two separate lives is over. Now, according to Google’s Chief Economist (and who am I to argue with Google’s Chief Economist?), I can simultaneously and proudly hold the pulse of both art and math.

Data visualization has emerged as a vital area of development. As the amount of data available to us grows exponentially (think petabyte hard drives selling at Costco), our ability to process and utilize the information flowing from all this data depends on its interpretation and presentation.  Data is really only useful if we can expose patterns and correlations in the data and then represent those patterns and relationships in an accessible manner. The visual representations of the patterns and relationships found in data serves to enhance our ability to understand the “story” contained in the data. Surely, there is a meaningful relationship between the explosion of data and the resurgent interest in powerful Design as marked by the number of articles focused on design in magazines like FastCompany and Wired. (In fact, Wired’s latest issue 17.07 is a particularly striking look at how data and design work together in all sorts of venues/markets.  In addition, see these sites pitching Data Visualization for some examples of beautiful renderings of information: Smashing Magazine; Web Designer Depot)

As I say in every blog, the value of data is only as good as its ability to support decision-making. Companies that succeed will do so through sophisticated and compelling analytics. Sophisticated analytics rely on more and more complex algorithms as well as the integration of multiple data sources. Data design and representation–the way charts, graphs, and tables are designed–plays an integral role in allowing readers to analyze and interpret the data in support of decision-making.  The equation looks like this: the data determines its visual design; the visual design supports a competent and complete analysis; competent and complete analyses lead to better business decision-making.

If I were back in college, I would have a newfound level confidence.  I would proudly brandish my calculator in pottery class, and definitely show-off my newest bisque-fired pottery vase to my statistics classmates.  After all, it’s the combination that makes me sexy.

Doing My Taxes and Fixing Health Care IT

Even with a good accountant, most of us consider tax-time brutal. It’s not only that paying our taxes hurts, but organizing the papers, filling out the confusing forms, and thinking through the financial events of the previous year—most of those events are misery too. These are organizational tasks that present huge psychological barriers for many of us.

Over the past 8 months, I have been consulting with one of the premier health care organizations in the country. We are working on a key initiative to extract and integrate data from the various systems across their organization with the objective of creating performance reports for their clients. It is a mission-critical project as they compete to attract and retain their clients (i.e., employers.)

My primary obstacle to moving the project forward has been getting them to confront the barrier of initiating the work required to integrate their data – even at the conceptual level. This conceptual work precedes the data discovery process that will inevitably uncover gaps in data quality, gaps in data availability, and inadequacies in the database structure. Although addressing these data issues requires time and attention, this company has the IT staff to do the heavy lifting with the data. As with my taxes, the hardest step is initiating the organizational process.

My experience has been that integrating these disparate data sources is complicated but achievable. Here is an example:

A health plan wants to be able to demonstrate the performance of their Wellness Programs. The Benefits/HR Manager will want to know the following:

• How many people are participating in the programs?
• In what programs are they participating?
• Does the program participation line up with the company’s employee population health risks?
• Is the numbers of members participating growing over time and how do these numbers compare to the employee participation for similar employers with similar demographics?
• Are population risk factors declining over time?
• Does the data to support a correlation between participation and reduced health risk and higher employee productivity?
• How much do the Wellness programs cost?

To answer these questions, the health plan will need to draw on data from a number of different sources: program participation records, historical population risk data, change in population risk over time, health care claims data, program costs, and employee productivity measures (estimated or actual.)

This is the point where many health plans freeze or procrastinate. What follows are my recommendations to get past the obstacles. (They might work for tax preparation as well.)

1. Define the Output: It is much easier to “visualize” how to integrate the data when you know what questions you want answered. The Wellness program questions listed provide the framework for the data integration requirements.
2. Don’t Be Constrained by Legacy Technology. Too many organizations are shackled to their existing technology. New sophisticated technology for data analysis is emerging so constantly that an easier (and often cheaper) solution may be available.
3. Know the Data: There is no substitute for understanding the data. In tax preparation, reading bank statements and investment reports is challenging, but doable. Reading health data is the same. From our Wellness example, most of the data sources (e.g., Wellness Program Participation) are very straightforward. Here’s what you should ask for:
   • Database schema. These are the database blueprints that show how the data is organized
   • File layouts. These are the documents that describe what data is supposed to be in each database
   • Meta data. This is data about the data, or in other words, information that describes data volume, data quality, and data idiosyncrasies.
4. Don’t Overshoot the Data Integration Efforts: Too often health plans are persuaded that in order to answer questions such as the Wellness questions above, they need to build an Enterprise Data Warehouse that allows for all data sources to be linked to all other data sources at the patient level. This is a terrific goal, but economical interim data integration steps exist to answer effectively these types of Wellness questions. If Account Structure fields are consistent across data sets, then data can be efficiently aggregated at the Account/Group level for performance reporting.
5. Have a Deadline: Data integration projects will not happen without the mental motivation that comes with a hard deadline.

The annoyance of tax season seems to have little lasting benefit aside from forcing me to scrutinize my financial situation. A dedicated effort at organizing data to proof benefits programming, alternately, is a valuable effort and can become an important tool in the toolbox for managing the health and productivity of your workforce.