Predictive Modeling in Healthcare

Ok, so these are my musings over predictive modeling, not a knock on it. (That is the disclaimer). So, as I was sitting around, pretending to watch TV and ignoring the dog who seemed to want to go out and play, I thought about Predictive modeling, in a healthcare setting. Specifically, in a Healthcare Provider setting. I actually asked this question on several groups on LinkedIn and asked folks if they have had success with it. The only person who responded with a success story was Mr. Alex Zverev (you can view his profile on LinkedIn here: http://www.linkedin.com/pub/alex-zverev/1/a01/b03), and the scenarios in which he has had successes here (http://www.linkedin.com/groupAnswers?viewQuestionAndAnswers=&discussionID=90342387&gid=93115&commentID=65616962&goback=%2Egmp_93115%2Eamf_93115_10544349&trk=NUS_DISC_Q-ncuc_mr#commentID_65616962). So here are some of my thoughts.


My primary interest is in the Return on Investment of using predictive modeling. I see quite a few software providers out there touting to have predictive modeling capabilities, but haven't heard of a lot of success stories, especially in a healthcare provider setting. Even lesser information is available on the ROI of implementing a predictive modeling solution.

To me, an ultimate predictive modeling solution would be something that can predict the stock market, which has infinite number of variables to consider. But if it were that simple, everyone would be doing it. On the other hand, in healthcare, people are touting Clinical Decision Support capabilities using predictive modeling. "Which patient of yours is most likely to develop cancer?", for example. In my humble opinion, that again is quite a stretch, because of the number of variables that need to be taken into account, not to mention "objective research" that is available to create the model in the first place. 
For example, it would be easy to say that a smoker of Asian descent between the ages of 18-40 may develop cancer quicker than others. But what if he is a smoker with healthy eating habits and hits the gym 4 days a week? What if that person only smokes three cigs a day? What if he has no genetic predisposition to cancer? To me, this a cool exercise to conduct and eventually, as you gather more and more data and "evidence" really starts supporting your research in cancer, your model becomes much more reliable and this will start generating a measurable ROI, by reducing the cost of treating a patient through early screening and through preventive medicine. 


My thought is that if you are going to do predictive modeling, start with an area with a limited number of variables. Your "bang for the buck" would be realized sooner and it would be greater in that scenario. For example, the scenarios that Alex describes (Capacity Planning and Measured Display Times for Display Stations) have a better chance of an "immediate" ROI than, let's say, a cancer predicting algorithm. Now, if you are reading this, and have had successes with using Predictive modeling in different settings other than the ones described above, please let me know. I'd like to hear your stories.

Golf is like Business Intelligence!!

I was at the golf course this past weekend. As I was having a "heart to heart" conversation with the course (ok, it was more like me pleading to the course), I realized how much the game of golf is like investing in BI.

Let's take a look. Golf is unlike other games we play. For example, if you want to play soccer, you'll go get a ball, get 22 of your friends together , find an open space and take four sticks and stick them at the opposite ends for goal posts and you are ready to go. Not with GOLF!

Even before you set foot on the course, you have to buy a set of clubs. ($2000 for a decent set). Balls - $15.00 a dozen, unless you want the soft core max distance "I can fly like a bird" kind.
Tees - $1.50
Gloves - $16.00
Shoes - $150.00
Attire ; $100.00
Lessons (unless of course you want to invest in more balls and be the laughing stock of your buddies) - $500
Round of Golf - $50.00
Your game? - Sketchy at best for the first few rounds.

Now let's look at your BI initiative:
Hardware - Depending on how many you buy, could go upto $150K.
ETL Software (Server licences) - $750K
ETL Software (Developer licenses) - $150K
Database Server license - $100K ( Again, depending on how many you buy, this could go upto $500K)
BI Server License - $60K
BI Developer Licenses - $2K

Now, you have the equipment. Next step? Hire a Golf Pro (Usually known as Big Six consulting firm) to do a gap analysis and come up with a roadmap for implementation - $200K

You can't go golfing without your buddies now, can you? (Well, if you are like me and obsessed with the game of golf, you might just do that.) But I am not talking about us corner case scenarios then, am I? Let's call your buddies your implementation team. Total cost over a period of 8 months of implementation? $1.9 M

So, like the game of Golf, even before you set foot on the course, you have spent about $2M. Then, after the round of golf (implementation period), about $4M. Get the picture? ROI??

What can you do to avoid costly mistakes?
Step1: The first thing to do is to go through the "Measure Everything That Really Impacts Customers (METRIC)" process . Once you have the repository of metrics and have decided how you want to see your data and how often, then you can go to step 2.
Step 2: Prioritize your measures according to business need. Apply the 80/20 rule. Basically, 20% of your metrics should give you 80% of the value.
Step 3: Evaluate technology solutions that fit your need
Step 4: Purchase
Step 5: Implement
Step 6: Test and Deploy

Maximum value can be achieved with a little careful planning upfront. Now, you can go play a good round (and if you are in Jacksonville, call me for that round) of golf knowing that your BI initiative is better than 60% (according to Gartner) of the initiatives out there!