Healthcare companies conducting campaigns to promote healthy and cost-effective behaviors often see a lot of variability in their results. Change in the message, time of the year, or population can blur the Average Treatment Effect (ATE). It is then difficult to decide to what extent the intervention was successful and to discover which aspects could be improved. To overcome these challenges, we developed Hierarchical Bayesian Models (HBM) to evaluate our campaigns.
With HBM we can constantly optimize interventions that span multiple lines of business and time periods by combining similar campaigns. This approach unlocks new insights by considering multiple sources of influence at multiple levels, handling highly skewed distributions, increasing statistical power, and quantifying financial cost and benefit with credible intervals expressed in dollars. Overall HBM can help companies learn more from their outreach efforts and make more informed decisions. The advantages and challenges of HBM are illustrated with CVS Health campaigns aimed at optimizing site of care utilization.