Workshop 4: Causal Interpretations involving Discrete Mixture Modeling: Implications for Interventions for Depression and Suicide

The 4th 2013/2014 Workshop on Quantitative Methods in Education, Health and the Social Sciences (QMEHSS) will be on Friday February 21st from 11:00-12:30 and will be led by Dr. C. Hendricks Brown. The seminar will be held in the NORC conference room 344. NORC is located at 1155 E. 60th Street.

 

Causal Interpretations involving Discrete Mixture Modeling: Implications for Interventions for Depression and Suicide

C Hendricks Brown

Northwestern University, Feinberg School of Medicine

Different forms of mixture modeling can be used to examine heterogeneity of intervention impact.  Recent developments in this field have led to causally interpreted models, i.e., ones that can express differential impact on a subgroup, even if that particular subgroup cannot be completely identified. The first of these approaches involves mixtures of populations with different distributions of potential outcomes in randomized trials.  We highlight how we can construct “dual response” growth mixture models order to assess what proportion in a trial would benefit more from an intervention compared to control, what proportion would be harmed, and what proportion are likely to remain unchanged.  An example that combines four prevention trials is discussed.  A second discrete mixture approach involves different mediational or moderation relationships in a randomized trial that can be accounted for by mixtures.  We illustrate this on a synthesis of antidepressant trials. A third approach is to consider mixture analyses applied to repeated cross sectional associations across time, especially when policy changes, such as the FDA’s adoption of a black box warning for youth suicide exposed to antidepressants cause changes in the rates of medication usage in different subpopulations.  We discuss mixture methods that hypothesize that a portion of a population will be affected by such policies, while other parts of the population are likely to be unaffected.  This method is illustrated using data from repeated youth household surveys.  Implications, challenges, and limitations of these methods are discussed.