Random-pairing bootstrap and mutual informational correlation: A novel approach for assessing contamination in health behavioral intervention trials

Crosstalk causes contamination to behavioral intervention trials, presenting a big threat to intervention research in public health. Essentially, crosstalk can be considered as a process of information synchronization between participants in the intervention and the control groups, leading to behavior change among the control participants that is intended for the intervention participants.  Despite the scientific and practical significance, two challenges prevent researchers from assessing the effect of crosstalk in behavioral intervention studies. First, it is very hard if not impossible to assess who have talked with whom; and second no method readily available to quantify the effect of crosstalk. To overcome these challenges, in this study we devised the random-pairing bootstrap (RPB) method and adapted the mutual information (MI) technique from the information sciences to quantify crosstalk. The RPB method helps identified those who might have talked based on the correlation of the target variables evaluated with MI technique that is capable of quantifying both linear and nonlinear effect from crosstalk.  After a systematic test of the RPB-MI methodology with simulated data, data from a randomized controlled trial (n=1360) funded by the National Institute of Health were analyzed.   Results from the simulation data analyses indicated that RBP-MI method can effectively detect the known crosstalk effect.  Results from real data analysis showed larger within-group crosstalk than cross-croup talk.  These findings suggest the validity of the RBP-MI method in quantifying crosstalk. Further tests with diverse data are needed to establish the utility of this methodology in health behavior intervention research.