Abstract:
|
When the unit of observation is nested within a patient it is not uncommon that the outcome is not observed for each unit, instead only a combined outcome is observed at the patient level. We consider the case of exchangeable units with a binary outcome that is partially observed, i.e. a patient-level response is observed only if all the units have a response. We develop a regression model to evaluate the effect of unit-specific covariates on the binary outcome, and compare its performance with several ad-hoc procedures. The proposed method is applied to data investigating the effect of characteristics of spinal discs on discogenic pain, assuming that a patient is pain-free only if all of the discs are healthy.
|