Table 3.

Elicitation of distributions for the sensitivity parameter ∆. Our subject-matter expert was asked to fill out this form to elicit the values for the sensitivity parameters. Her responses are shown in bold.

Consider two women who have the same history of smoking and weight change and are receiving the same intervention (exercise or wellness). Patient A is observed at time t − 1 but drops out of the study and is not seen at time t, while patient B remains in the study and is observed at time t.
If the probability that patient B smoked during week t is p, what is your best guess (median) for the probability that patient A (who dropped out) smoked during week t? Also provide a lower bound and upper bound on reasonable values.
Treatment	Prob. observed patient B smokes (p)	Best guess	Lower bound	Upper bound
		for the probability that the unobserved Patient A smokes
Wellness	25 %	50 %	40 %	60 %
Wellness	50 %	70 %	60 %	80 %
Wellness	75 %	95 %	90 %	100 %
Exercise	25 %	50 %	40 %	60 %
Exercise	50 %	70 %	60 %	80 %
Exercise	75 %	95 %	90 %	100 %
If the observed patient B has an expected percentage weight change from baseline of w at week t, what is your best guess (median) for the expected percentage weight change from baseline at week t for patient A who dropped out? Also provide a lower bound and upper bound on reasonable values.
For reference, the average weight change was 2.4% and the standard deviation was 2.6%. Also, negative values are allowed if it is believed that the patient will have lost weight since baseline.
Treatment	Weight change for observed patient B (w)	Best guess	Lower bound	Upper bound
		for the expected weight change of the unobserved Patient A
Wellness	2.5 %	2 %	0 %	5 %
Exercise	2.5 %	1.5 %	0 %	4 %