ABSTRACT
The main motivating data set comes from the Interstitial Cystitis Data Base (ICDB) (Propert et al., 2000). Interstitial cystitis (IC) is a painful condition due to inflammation of tissues of the bladder wall. The cause remains unknown and there are no effective diagnostic tools. It is usually diagnosed by ruling out other conditions such as sexually transmitted disease, bladder cancer, and bladder infections. Some studies (Kirkemo et al., 1997) suggest that IC can be characterized by at least one of the following symptoms: pain in the pelvic or bladder area, urgency (pressure to urinate), and frequency of urination. It has been shown that the different symptom patterns are caused by different underlying disease mechanisms. Thus, to get a better understanding of this chronic disease, it is important to determine whether symptoms tend to co-fluctuate together (suggesting a single underlying aetiology) or vary independently (suggesting multiple mechanisms at work). The ICDB kept track of pain, urgency, and urinary frequency, for a prevalent cohort of patients with IC over a couple of months. For each patient, the pain score, urinary urgency, and urinary frequency were observed as three major longitudinal outcomes. Also, we have the demographic characteristics and clinical biomarkers of patients which were also recorded over time. The pain score and urinary urgency take ordinal/continuous values, while the urinary frequency is an integer that counts the number of urinary episodes during a given period. It is thus natural to assume the urinary frequency to follow Poisson distribution. The goal of the analysis is to use regression models to investigate the association between demographic/clinical biomarkers and IC-symptom variables. Also, it is of interest to study the change in correlation over time between the continuous and Poisson outcomes which will provide implications on the disease mechanism. There are missing values in the longitudinal outcomes because some patients drop out of the study. This motivates the need for developing statistical methods to deal with missing data.
