1. Time-varying effect modeling to study developmental and dynamic processes STEPHANIE T. LANZA, PHD DECEMBER 14, 2015

2. Outline 1.Background 2.Study I: Nicotine addiction o Recovery is dynamic 3.Study II: E-cigarette use o A developmental perspective 4.Future directions

3. Outline 1.Background 2.Study I: Nicotine addiction o Recovery is dynamic 3.Study II: E-cigarette use o A developmental perspective 4.Future directions

4. Interdisciplinary background Training ◦Mathematical Science, Psychology, Applied Statistics, Human Development Current position ◦Professor, Biobehavioral Health ◦Scientific Director, The Methodology Center methodology.psu.edu Develop and disseminate new methods for social, behavioral, and health sciences research focusing on vital public health issues, including substance abuse and HIV

5. My orientation Human behavior as it relates to health is dynamic (and multifaceted) Orientation relevant for understanding ◦Behavioral change – across development, time ◦Changes in process – across development, time ◦Differential intervention effects – across development, time

6. What is the time-varying effect model? We know that negative affect and craving to smoke are tightly linked among addicted smokers. What does recovery look like? TVEM can address questions such as: Is negative affect differentially associated with craving at various points in the smoking cessation process?

7. Suppose an intervention was conducted to reduce alcohol abuse 2-arm RCT Post-baseline measurement ◦One time point ◦Two time points ◦Multiple time points (moving window) A thought exercise

8. Effect of intervention: One time

9. Effect of intervention: Two times

10. Effect of intervention: Multiple times

11. Longitudinal data typically collected to capture temporal changes in a process Natural to expect that both the outcome and the relationships between the covariates and the outcome may change over time TVEM is designed to evaluate whether and how the effects of covariates change over time TVEM: Direct extension of regression

12. Regression coefficients express associations between variables Traditional regression predicting outcome Y TVEM allows coefficients to be dynamic TVEM: Direct extension of regression

13. In TVEM, estimate regression coefficients (i.e., associations) as flexible function of continuous time Use figure to interpret a “coefficient function” Coefficient functions are estimated

14. Let’s think about smoking cessation Regression coefficient expresses this association: TVEM allows coefficients to be dynamic: Let’s review the state of the science…

15. Data requirements (intensive longitudinal data)

16. Only one or a few waves, but many ages sampled Example: The National Longitudinal Study of Adolescent to Adult Health (Add Health) ◦Nationally representative sample ◦4 waves of data collected from 1996-2008 ◦N~12,000 (core sample) ◦34,562 person-times (spans ages 12-32) Data requirements (cross-sectional and panel studies)

17. Add Health: Coverage across age

18. Development of TVEM 1990’s ◦Functional regression analysis introduced in statistical literature (Hastie & Tibshirani, 1993; Hoover et al., 1998) 2010 ◦SAS software released

19. Development of TVEM 2012 ◦Tutorial paper: Shiyko, Lanza, Tan, Li, & Shiffman (2012) Prevention Science ◦Pre-conference workshop: Society for Research on Nicotine and Tobacco ◦NCI R01: smoking cessation dynamics using TVEM 2013 ◦Pre-conference workshop: Society for Prevention Research ◦NIH funds supplemental issue of Nicotine and Tobacco Research ◦Application paper: Liu, Li, Lanza, Vasilenko, & Piper (2013) Drug and Alcohol Dependence

20. Development of TVEM 2014 ◦Supplemental issue published (Lanza, Piper, & Shiffman, Eds.) ◦Other researchers publishing with TVEM 2015 ◦Summer Institute on Innovative Methods ◦Pre-conference workshop for Society for Ambulatory Assessments (SAA) ◦SAS software expanded – random effects, weights

21. Outline 1.Background 2.Study I: Nicotine addiction o Recovery is dynamic 3.Study II: E-cigarette use o A developmental perspective 4.Future directions

22. NCI R01 Advancing Tobacco Research by Integrating Systems Science and Mixture Models Goal: Apply innovative methods to existing data from an RCT to advance scientific knowledge that will inform the next generation of smoking interventions

23. Nicotine addiction Tobacco use: Leading global cause of preventable death Each year, kills millions and costs hundreds of billions of dollars Institute of Medicine, 2011 95% cessation attempts end in relapse; withdrawal symptoms primary reason Improved understanding of withdrawal symptoms and how treatments alleviate them could: ◦Lead to new treatments ◦Inform tailored treatments (to people, to time)

24. Wisconsin Smokers’ Health Study 1504 daily smokers enrolled in smoking cessation RCT Placebo group ◦Counseling only Treatment groups ◦Counseling + single med (lozenge, patch, Bupropion) ◦Counseling + two meds Real-time assessment of dynamic phenomena (withdrawal symptoms, mood, behavior)

25. Study design EMA: 4 assessments per day ◦Waking ◦2 random ◦Bedtime Assessed withdrawal symptoms (craving, affect, hunger), smoking, motivation, etc.

26. Time-varying effects of smoking interventions Goal 1: Study the underlying dynamics of craving during cessation attempt Goal 2: Estimate effect of treatment on decoupling craving from its key drivers (e.g., negative affect) From Lanza et al. (2014) Nicotine and Tobacco Research

27. Measures Outcome: Craving during first two weeks of quit attempt ◦Intensively assessed via EMA Predictors: ◦Baseline nicotine dependence (not time-varying, but effect can be!) ◦Negative affect (time-varying) Moderator: Treatment group

28. Specify model What varies with time? ◦Mean craving (intercept function) ◦Negative affect ◦Effect of negative affect (slope function) ◦Effect of baseline dependence (slope function)

29. Effect on craving: Negative affect Lanza et al. (2014) Nicotine and Tobacco Research Placebo Treatment

30. Effect on craving: Baseline dependence Lanza et al. (2014) Nicotine and Tobacco Research Placebo Treatment

31. Implications for smoking cessation Think differently about treatment effects With time, treatment changes the relationship between baseline dependence and craving Treatment diffuses role of negative affect – a key driver of craving – early in quit attempt

32. Broader implications Effects of static “baseline” variables can change over time Effect of treatment in standard RCT may be time-varying ◦Model intervention processes we posit Could inform tailoring of treatment to individuals and to time (adaptive intervention designs)

33. Outline 1.Background 2.Study I: Nicotine addiction o Recovery is dynamic 3.Study II: E-cigarette use o A developmental perspective 4.Future directions

34. NIDA R01 Age-varying effects in the epidemiology of drug abuse Goal: Apply TVEM to existing national data to advance scientific knowledge about the etiology of substance use, co-use, comorbidity with mental health problems, and health disparities

35. E-cigarette use among adolescents Developed as “reduced harm product” thus often considered safe alternative to traditional cigarettes (Cobb et al., 2010) Inhalation-activated devices; heat produced which turns solution (nicotine, other additives) into vapor ◦Eliminates combustion/smoke, but long-term effects of use inconclusive (Chapman & Wu, 2014; Cobb et al., 2010; Pepper & Brewer, 2014, Williams & Talbot, 2011) Adolescent use rising rapidly ◦Lack of FDA regulations ◦Gateway to traditional cigarettes?

36. National Youth Tobacco Study Cross-sectional data from 2014 CDC to assess “tobacco-related beliefs, attitudes, behaviors, and exposure to pro- and anti- tobacco influences” 22,007 US middle- and high-school students ◦ages 11-19 (mean 14.5) ◦49% female ◦48% White, 17% Black, 29% Hispanic

37. Etiology of traditional and e-cigarette use Goal 1: Estimate disparities in rates of use across adolescence for sex and race/ethnicity population subgroups Goal 2: Estimate prevalence of use of both products as a continuous function of age throughout adolescence Lanza et al. (in preparation)

38. Measures Current traditional cigarette smoking ◦1 for use in past 30 days, 0 otherwise (6.4% yes) Current e-cigarette smoking ◦1 for use in past 30 days, 0 otherwise (9.2% yes) Age (to nearest year) Sex, Race/ethnicity (moderators)

39. Specify model (logistic TVEM) What varies with age? ◦Probability of cig use ◦Probability of e-cig use ◦Effects of sex, race/ethnicity ◦Effect of cig on e-cig (age-varying odds ratio)

40. E-cigarette and traditional cigarette use: Sex differences (ages 11-19) E-cigarettes Traditional Cigarettes Female Male Female Male

41. E-cigarette and traditional cigarette use: Race/ethnicity differences (ages 11-19) E-cigarettes Traditional Cigarettes Black White Hispanic Black White Hispanic

42. Use of both products (odds ratio, ages 11-19) Among those age 12, adolescents using e-cigarettes are >40 times as likely to use traditional cigarettes compared to those not using e-cigarettes

43. Implications for policy and prevention Identification of key ages of risk can inform targeted, age- appropriate intervention Traditional and e-cigarette use go hand in hand, particularly in very early adolescence Early use of e-cigarettes significantly more likely among Hispanic youth, suggesting greater risk for future nicotine dependence

44. Outline 1.Background 2.Study I: Nicotine addiction o Recovery is dynamic 3.Study II: E-cigarette use o A developmental perspective 4.Future directions

45. New NIDA P50 Center of Excellence (PI: Linda Collins) Center for Complex Data to Knowledge (CD2K) in Drug Abuse and HIV Behavioral Science My scientific Project: Multidimensional and Dynamic Moderation in Drug Abuse and HIV Studies

46. New information contained in contemporary data sources Intensive longitudinal data (ILD) Electronic medical records (EMR) Genetic data Big data = big opportunity

47. New approaches to health behavior change Adaptive interventions, Mobile interventions, Precision medicine ◦Stress, mood, behavior Dynamical systems: Engineering meets health

48. Innovative applications of TVEM Dynamical (time-varying) effects of interventions Developmental associations Associations across historical time Complex link between age-of-onset and later outcomes

49. John Dziak Runze Li Rebecca Evans-Polce Michael Russell Sara Vasilenko KEY TVEM COLLABORATORS

50. R01-DA039854, P50-DA010075, P50-DA039838 National Cancer Institute R01-CA168676 CURRENT FUNDING

51. Recent literature using TVEM: Dynamic effects Lanza, S. T., Vasilenko, S. A., Liu, X., Li, R., & Piper, M. E. (2014). Advancing understanding of the dynamics of smoking cessation using the time-varying effect model. Nicotine and Tobacco Research, 16, S127-S134. doi: 10.1093/ntr/ntt128 Liu, X., Li, R., Lanza, S. T., Vasilenko, S. A., & Piper, M. E. (2013). Understanding the role of cessation fatigue in the smoking cessation process. Drug and Alcohol Dependence. 133, 548-555. doi: 10.1016/j.drugalcdep.2013.07.025 Mason, M., Mennis, J., Way, T., Lanza, S., Russell, M., & Zaharakis, N. (in press). Time-varying effects of a text-based smoking cessation intervention for urban adolescents. Drug and Alcohol Dependence. Shiyko, M. P., Lanza, S. T., Tan, X., Li, R., & Shiffman, S. (2012). Using the time-varying effect model (TVEM) to examine dynamic associations between negative affect and self-confidence on smoking urges: Differences between successful quitters and relapsers. Prevention Science, 13, 288-299. doi: 10.1007/s11121-011-0264-z Tan, X., Shiyko, M. P., Li, R., Li, Y., & Dierker, L. (2012). A time-varying effect model for intensive longitudinal data. Psychological Methods, 17, 61-77. doi:10.1037/a0025814. Vasilenko, S. A., Piper, M. E., Lanza, S. T. Liu, X., Yang, J., & Li, R. (2014). Time-varying processes involved in smoking lapse in a randomized trial of smoking cessation therapies. Nicotine and Tobacco Research, 16, S135-S143. doi: ntt185

52. Recent literature using TVEM: Developmental, historical time, age of onset Evans-Polce, R. J., Vasilenko, S. A., & Lanza, S.T. (2014). Changes in gender and racial/ethnic disparities in rates of cigarette use, regular heavy episodic drinking, and marijuana use: Ages 14 to 32. Addictive Behaviors. Advance online publication. doi: 10.1016/j.addbeh.2014.10.029. Lanza, S. T., & Vasilenko, S. A. (2015). New methods shed light on age of onset as a risk factor for nicotine dependence. Addictive Behaviors, 50, 161-164. doi:10.1016/j.addbeh.2015.06.024 Lanza, S. T., Vasilenko, S. A., Dziak, J. J., & Butera, N. M. (in press). Trends among US high school seniors in recent marijuana use and associations with other substances: 1976 to 2013. Journal of Adolescent Health. Schuler, M. S., Vasilenko, S. A., & Lanza, S. T. (in press). Age-varying associations between substance use behaviors and depressive symptoms during adolescence and young adulthood. Drug and Alcohol Dependence. Pub’d Vasilenko, S.A. & Lanza, S. T. (2014). Predictors of multiple sexual partners from adolescence through young adulthood: An application of the time-varying effect model. Journal of Adolescent Health, 56, 491-497. doi: 10.1016/j.jadohealth.2013.12.025

53. Stephanie T. Lanza SLanza@psu.edu methodology.psu.edu THANK YOU!