Reconciling disease progression risks across multiple clinical cohorts

• Activity Number: 138 - Statistical Methods for Electronic Healthcare Data
• Type: Invited
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Jane Lange and Lurdes Inoue and Ruth Etzioni*
• Companies: Fred Hutchinson Cancer Research Center and University of Washington and Fred Hutchinson Cancer Research Center

Abstract

Cancer outcomes are often observed at discrete times via doctor-patient encounters or specialized diagnostic evaluations. In the setting of low-risk prostate cancer, patients may opt for a regimen called active surveillance in which biomarker measurements and biopsies are performed at given frequencies to determine whether their disease has transitioned to a state that warrants active treatment.. It is unclear whether observed differences in progression risks across cohorts are due to difference in implementation or in underlying risk. We present two distinct approaches to addressing this question, one based on joint longitudinal-failure time modeling and one based on multistate and hidden Markov models. In analysis of data from four North American active surveillance studies, both methods reveal that relative differences across cohorts in underlying risk are quite different than those based only on the empirically observed discrete outcomes. We conclude that integrating electronic health care from multiple clinical cohorts with diverse surveillance schema may require modeling that permits assessment of underlying risks as they evolve over time.