Statistics Seminar - Richard J. Cook - Cox regression with dynamic markers measured at covariate-dependent visit times
Title: Cox regression with dynamic markers measured at covariate-dependent visit times
Speaker: Richard J. Cook (University of Waterloo)
Abstract: Cox regression is routinely used to assess the association between internal time-dependent covariates (markers) and failure times. The markers are typically only measured at clinic visits when careful examination can take place, or when blood samples are drawn for testing, so they are strictly under intermittent observation. Examples include measurement of blood glucose control in diabetics , blood pressure in hypertension trials, or markers of bone destruction in patients with skeletal metastases. We consider the asymptotic bias of estimators of the marker effects on the hazard for failure when the marker values are carried forward until the next clinic visit. We also consider the setting where the marker values, often reflecting the severity of disease activity, affect the intensity for clinic visits. A joint model is then proposed for the marker, visit, and failure process to mitigate this bias. Simulations are reported on to investigate the finite sample performance of naive and valid estimators and a clinical illustration is provide.
Date/Time: Tuesday October 5, 2021, 3:30 - 4:30
Location: VirtualJoin Zoom Meeting
Meeting ID: 971 9900 3250