Ready to Quit: Predictive Analysis on Member Readiness of Enrolling Smoking Cessation Programs in Kaiser Permanente Southern California
predictive modeling, smoking cessation
Background/Aims: We describe a predictive analysis to help regional operations engage smoking members with high likelihood of enrollment into smoking cessation services.
Methods: A combination of investigation methods was used: 1) data mining and basic free text parsing and recognition; 2) predictive modeling to discover predictors of member enrollment in smoking cessation services and score the future population to identify smoking members with high likelihood of enrollment; and 3) randomized control trial to (a) identify the best protocols to outreach these members, and (b) refine the model and establish a learning cycle to guide operations.
Results: We have identified and processed a wide range of predictors including smoking habit, social/economic factors, enrollment history, significant medical conditions, family characteristics, medical utilization history, online interaction and other health/social behaviors. The model performed very well in validation process, with c-stat of 0.76, which provides a good fit and predicts power. The findings of significant predictors were in agreement with many literatures. These predictors include past enrollment, sex and disease burden. We also found that race/ethnicity and online activity, especially patient/physician secure email communication on smoking matter, strongly predict likelihood of enrollment.
Discussion: In the Kaiser Permanente Southern California region, approximately 200,000 smokers can be identified through social history and ICD-9 codes in Kaiser Permanente HealthConnect. The top 20% of the modeled population represents 50% of potential enrollees in smoking cessation programs, and this population will be targeted in the outreach effort in a randomized control trial designed as follows. A population of 10,000 will be randomly selected to not participate in phase one, but will be eligible to participate after the first phase. Of this 10,000, 5,000 will receive no intervention and 5,000 will receive only existing auto-dialer outreach, both of which are within the normal standard of care. The remaining 30,000 members (10,000 per month) will be randomized by medical center into three treatment groups of equal size: auto-dialer plus live phone call, auto-dialer plus letter, and auto-dialer plus email.
Meng D, Garrido T. Ready to Quit: Predictive Analysis on Member Readiness of Enrolling Smoking Cessation Programs in Kaiser Permanente Southern California. J Patient Cent Res Rev 2015;2:98. http://dx.doi.org/10.17294/2330-0698.1100