Jean-Patrick Tsang, PhD & MBA (INSEAD)
Tel: (847)920-1000
Email: bayser@bayser.com

Igor Rudychev, PhD
Tel: (847) 679-8278
Email: igor@bayser.com

Boltzmann/Predictive Modeling

Problem #1:

Question: How differently do we proceed today to understand patient behavior?

Boltzmann1

Answer: Pretty much the same as before!  The sad truth is we still go for the average patient behavior to make decisions and drive forecasting models, unwittingly foregoing vast amounts of extremely valuable information.

What We are Foregoing?

Background: What we are looking at here is the drug utilization behavior of real people under real-life conditions (e.g. holidays, disruption in production, price increases, etc.), not that of a few hundred patients in a “lab” setting.  By zeroing on the average, we lose the opportunity to see:

  1. Full range of Real-life Behavior.
  2. Structure of Segments.
  3. Contextual Information upon which behavior is predicated.

 

Problem #2:  How do we analyze patient behavior?

Boltzmann2

Answer: Study different aspects of patient behavior in isolation.  While this is a good first step, we do not move on to piece the whole thing together.  For instance, we do not venture to understand how compliance impacts persistence.  This is probably because up to now we did not have the analytical tool to do so.

Approach – Summary

Key Features of the forecasting model we propose:

    Behaviors Modeled as Distributions
    • Writing Rx behavior of physicians.
    • Filling Rx behavior of patients.

  • Behaviors are context-sensitive
    • Patient profile
    • Physician profile
    • Managed Care and drug delivery environment
    • Course of Treatment and Algorithm

  • 3-Player Game
    • Patient plays with Physician for the enjoyment of Pharma
    • Patient-Physician game favors Patient
    • Pharma can alter the course of the game by changing the behavioral distributions of Patient and Physician.

ABOUT BAYSER | CONTACT | SITE MAP