Design and analysis of arm-in-cage experiments: inference for three-state progressive disease models with common periodic observation times.

We develop statistical methods for designing and analyzing arm-in-cage experiments used to test the efficacy of insect repellents and other topical treatments. In these experiments, a controlled amount of the treatment is applied to a volunteer's forearm, which then is exposed to the insects by being placed into a special cage. Arms are not kept in the cages continuously, but rather placed there periodically for a brief period of time, during which it is noted whether an insect lands (but does not bite) or (lands and) bites. Efficacy of a repellent can be described using a progressive three-state model in which the first two states represent varying degrees of protection (no landing and landing without biting) and the third state occurs once protection is completely lost (biting). Because subjects within a treatment group follow the same cage visit schedule, transition times between states are interval censored into one of several fixed intervals. We develop an approach that uses a mixture of nonparametric and parametric techniques for estimating the parameters of interest when sojourn times are dependent. Design considerations for arm-in-cage experiments are addressed and the proposed methods are illustrated on data from a recent arm-in-cage experiment as well as simulated data.