Multi-state modelling for examining adherence to cervical cancer based on research with Institute for Clinical and Evaluative Sciences
Abstract
Cervical cancer screening has decreased cervical cancer incidence rates by allowing for the early detection of abnormalities. In July 2015, 36% of the eligible women in Ontario were overdue for screening. There is a need for the implementation of appropriate statistical methods to understand factors associated with screening adherence so that target populations requiring encouragement to participate in screening can be identified. We studied individual and physician-level characteristics associated with screening adherence through the implementation of a time non-homogeneous multi-state model on a province-wide longitudinal cohort (N = 1,156,720). The data was extracted from Ontario Health Insurance Plan Registry (OHIP) with index dates from 1991 to 2013. Characteristics were incorporated into the model as time-varying covariates. We were interested not only in how these covariates were associated with the rate of becoming adherent but also the disparity between immigrants and Canadian residents. Individuals with fewer comorbidities, higher income, prior visits to a primary care physician, and residential stability had higher rates of becoming adherent. The rate of adherence was 6.9% lower among immigrants compared with non-immigrants. After including prior screening history into our regression model, we found that women who had smaller proportions of prior time spent non-adherent had higher rates of becoming adherent in the future. The implementation of a multi-state model allows us to monitor trends longitudinally and to provide insights so we can mobilize healthcare resources efficiently to minimize the time spent non-adherent in the appropriate populations.
