This webinar is part of the Advanced Methods Webinar Series
Survival analysis or time-to-event analyses have been popularized for making predictions about future events based on some exposure in the past. The methods are familiar to epidemiologists but also to actuarial scientists / insurance companies to estimate risk.
These types of analyses are often affected by the quality of data available – in particular, the granularity (months, weeks, days) of time used to capture events (including outcomes, time-varying covariates/exposures). The ability to represent cohort versus secular trends and truncated events is also affected by the time scale used. In both cases, the representation of censored observations can have important ramifications on the study findings.
This presentation will give a quick overview of Cox proportional hazard models, covering the assumption of independence of censoring, as well as ways to address the lack of independence using inverse probability censor weighting. It will go into detail about what censoring entails, and how left-, right-, and interval-censoring relate to the observation window. Methods of addressing left and right censored events will be discussed by exploring the benefits of changing the time scale of investigation.
Some examples and cautionary tales from the literature will be highlighted. Sample code from SAS, R and (possibly) Stata will be provided.
Dr. Andy Kin On Wong is an epidemiologist at the Dalla Lana School of Public Health, teaching the Advanced Methods in Epidemiology course – survival analysis module. He is also an imaging and data scientist primarily appointed at UHN, under the Joint Department of Medical Imaging.
His research is focused on developing methods for quantifying musculoskeletal image features from MRI and CT scans using Python. He practices a combination of more traditional algorithms and deep learning approaches (neural networks). He currently applies epidemiological methods to study how musculoskeletal tissue interactions could impact sex and gender-calibrated differences in pain perception using a combination of survival analysis and structural equation modeling.