Statistical modeling for medication adherence

Patients not adhering to their medication regimens has been shown in multiple studies to yield sub-optimal health outcomes as well as decreased sales opportunities for pharmaceutical companies. Various CRM efforts over the years have tried to address this. Now a new tool may be avaialable.

There was a fascinating article in the NY Times Health Blog recently about Fair Isaac (FICO)creating a logistic regression style "medication adherence score" much like their credit score, that predicts likelihood to adhere based on demographic, financial, and transactional variables. For example: Are sixty-something
middle class Midwestern grandparents with low net worth particularly non-adherent?

If a score comes up as such, would a pharmaceutical company target them with a an extra adherence mailer, kit, or a premium adherence pill bottle? Or call center support? Since these are expensive resources, targeting using a validated model score seems attractive.

Of course, these resources could be offered to anyone, but could be promoted selectively.

I think likely the modeling score was validated on historical data. It will be fascinating to see how accuracy is tracked moving forward. Also, whether pharmaceutical firms or communications agencies will adopt it.

Perspective in measurement and in life

In measuring time series data, one realizes that significant outliers or spikes will take on significance only over a certain duration and a certain time scale of granularity. A daily spike in website visits may seem signficant within last month's daily visitor plot. However, it may seem less so with a weekly plot. Or it may lose significance further compared to other larger spikes on a duration plot of an entire year.

This spring weekend has brought a new meaning to judging the relative importance of events, and in juding magnitudes of scale of importance. This humble overworked professional has tried to keep up with assignments over the weekend. He's also watched his Amazon rankings of his book on a daily basis, and wished they would nudge higher. These issues pale in comparison to the pleasure of riding bikes and playing baseball with kids as spring draws near. On the negative side, both hassles and joys are tempered by stuggles friends have with flooding nearby.

Furthermore, all of these are tiny in perspective when considering the natural disasters in Japan that happened within a day but whose duration of impact will be measured in years and tens of thousands of lives.

Benoit Mandelbrot RIP, an ode to modeling complexity

An intellectual giant has passed, and his life and research can be inspiring to all of us, whether a young theoretical mathematician (as I once was in college), or a data-driven marketer (where I have ended up so far). The news came to me rapidly . via Twitter from Analytics Bridge as well as several techy blogs.

Benoit Mandelbrot (see the NY Times Obituary) was famous for developing fractals, that are mathematical models of the complexity of nature, especially growth patterns. These models are of infinite complexity, and usually beautiful to render, like the image shown on this blog post. Read one of his biographies about how Mandelbrot got started thinking about fractals as he investigated a question on "how long is the coast of Britain?" and realized "it depends on how closely you look."

This brings me to the point of online digital metrics, which can be simplified with shallow views like visits, page views, or friend counts. Or, one can embrace the complexity and do full path analyses, understand behaviors, see the patterns in the viral spreading of social networks. The insights of going in deep can lead to brainstorming that turns around a business. Even to beautiful graphics like Mandelbrot fractals.