Studies of recurrent events often collect data on the number of occurrences between screening examinations. Although exact event times are not available, surrogate measures of the latency times prior to detection at a screening examination can sometimes be obtained. For example, in studies of recurrent benign tumors, tumor size or grade can be measured. This article proposes a Bayesian approach for incorporating surrogate information in the analysis of interval count data. A non-homogeneous Poisson process model characterizes event occurrence, and a flexible ordinal regression model characterizes progression in the categorical surrogates with increasing time since onset. Generalizations are proposed for accommodating covariate effects and unexplained heterogeneity in the rates of onset and progression. A Markov chain Monte Carlo algorithm is outlined, and the methods are applied to uterine tumor data.

Keywords: Interval-censored; Multiple event times; Panel counts; Poisson process; Progression; Recurrent events; Screening data; Tumor multiplicity.

The manuscript is available in PostScript and PDF formats.