|
Abstract:
|
Housed within a patient portal, e-Assist is a decision support program that assists patients with a physician recommendation for completing colorectal cancer screening (CRCS=1 if done, 0 otherwise). The patients are randomized to e-Assist or usual care within physicians. However, the physician-specific effects of e-Assist on CRCS vary randomly over physicians of varied composition, and compliance to treatment assignment is imperfect. Because a control patient is unable to access e-Assist and forced to receive usual care, the patient could be "complier" if she would have taken e-Assist or "never taker" otherwise under the alternative assignment to e-Assist; a patient assigned to e-Assist is an observed complier if the patient took the assignment or never-taker otherwise. Assuming compliance missing at random, we estimate a joint random coefficients model for CRCS and compliance by an accelerated combination of the EM algorithm and Newton Raphson via adaptive Gauss Hermite quadrature. Random coefficients are physician-specific never-taker, control-patient CRCS rates and complier average causal effects whose means, variances and correlations describe the population of compliers.
|
