|
Abstract:
|
Cancer progression outcomes are often identified via interval diagnostic examinations. For example, men with low grade prostate cancer may opt for active surveillance (AS) with periodic biopsies to monitor disease status and treatment referral only if disease progresses. Recommendations for AS implementation are based on observations AS cohorts, but studies vary in terms of biopsy frequency and patient compliance. Our goal is to study the underlying risk of disease progression from low to high grade disease and compare it across cohorts. The empirical data cannot be used directly, because variations in biopsy frequency and dropout affect observed frequency of progression. To enable comparisons, we used latent continuous time Markov chain models. The models treat biopsies as discrete observations of an underlying continuous-time process which we estimate while considering dropout as a competing risk. Our results show that in some cases the underlying risk of progression is fairly similar across cohorts despite apparent differences in empirical upgrading distributions.
|
