Biomedical Data Science Grand Rounds with Jeremy Taylor, PhD

“Estimation and interpretation of time-varying effects of demographic, tumor, access to care and SES factors in cancer survival studies”

4/11/2024
12 pm - 1 pm
Location
In-person at DHMC, Auditorium H (or via Zoom)
Sponsored by
Geisel School of Medicine
Audience
Public
More information
Biomedical Data Science
603-646-5723

The Department of Biomedical Data Science at Geisel invites you to attend a Grand Rounds presentation by Jeremy Taylor, PhD, Professor of Biostatistics and Radiation Oncology, University of Michigan School of Public Health, on Thursday, April 11 from 12:00-1:00pm at DHMC, Auditorium H (or via Zoom).

 

Talk title: “Estimation and interpretation of time-varying effects of demographic, tumor, access to care and SES factors in cancer survival studies

 

Host: Tor Tosteson, ScD

 

Location: In-person at DHMC, Auditorium H or via Zoom (no registration required)

 

Please see link below for more details.

 

Zoom meeting ID: 503 779 5102

 

Zoom passcode: 6501974

 

URL: https://dartmouth.zoom.us/j/5037795102

 

Phone (if needed for audio only, or to join by phone only): 669-900-6833

 

 

Presentation Summary

There have been thousands of studies that used the Cox proportional hazards model to analyze data and assess the association between factors measured at the time of diagnosis and survival outcomes in patients diagnosed with cancer. The Cox model makes the assumption of proportional hazards, but a more general model may give a more accurate description of the data. In a more flexible model, the hazard ratio associated with any factor can be allowed to vary over time. In this talk I will describe how we overcame the computational and statistical challenges of fitting such a model when the datasets are extremely large. I will then use the model to examine data from the SEER and NCDB cancer registries. I will take a pan cancer approach and consider 14 different cancer types. For SEER I consider a basic set of standard factors (age, race, stage, year of diagnosis, gender) and two outcomes, either death due to cancer or death due to other causes. The analyses reveal some substantial time-varying associations, especially for stage and age. The NCDB registry only has death from any cause, but it has additional factors, including comorbidities, SES variables and access to care variables. We investigate whether the time-varying age effect can be explained by other factors. The answer is generally no. I will invite the audience to speculate on possible reasons for the time varying association with age.

 

Location
In-person at DHMC, Auditorium H (or via Zoom)
Sponsored by
Geisel School of Medicine
Audience
Public
More information
Biomedical Data Science
603-646-5723