Predictive Model + Segmentation + Intervention

I was fielding a few questions today about the predictive model from Express Scripts.  The concept of predictive modeling is one that everyone is working on and holds great allure.  BUT, it is only a piece of the puzzle.  In the dialogue, I identified three key tenets for success.

1. Predictive Model – Can you predict who is likely to act in certain ways – be adherent, log-in to the member portal, use mail order, switch to generics, pick up the phone?
2. Segmentation Model – Once you can predict people, can you develop a segmentation model about how to get them to take action based on different attributes – gender, age, past behavior, preferences, income, other?
3. Intervention Strategy – If you can predict who is most likely to act and know what type of segment they fall into, do you have a cost-effective intervention strategy to get them to take action…right message at the right time using the right channel?  For adherence, this could be reminders, coaching, devices, or other tools.  (As many people say, a less sophisticated strategy executed perfectly is better than a complex strategy executed less than perfectly.)

And, then you need to study and refine these on an ongoing basis especially since topics like adherence may be affected by macro-economic trends (e.g., economy), patient beliefs (e.g., fatalism), and other attributes (e.g., plan design) on top of the attributes in your models. 

I do believe we’re early in the days of modeling and that the access to data and greater availability of informatics resources will increase the development and focus on these models.