UW Biiostatistics & Medical Informatics UW Biostatistics & Medical Informatics UW Madison UW Biostatistics & Medical Informatics Site Map
UW School of Medicine and Public Health UW Madison


 

 


Other Seminar Series

Seminars

General Departmental Seminar Series


Selection Bias in a Partially Retrospective
Molecular Genetic Neuro-Oncology Study

Rebecca Betensky
Department of Biostatistics, Harvard University

Tuesday, Dec 5, 2000, 12:00-1:00 pm

K6/115-117 Clinical Sciences Center, 600 Highland Avenue

ABSTRACT

Oligodendrogliomas are a common variant of malignant brain tumors, and are unique for their relative sensitivity to chemotherapy and better prognosis. For these reasons, the identification of an objective oligodendroglial marker has been a long sought-after goal in the field of neuro-oncology. To this end, 75 patients who received chemotherapy at the London Regional Cancer Centre between 1984 and 1999 were studied. For 50 of the patients, chemotherapy was planned from diagnosis, whereas for the remaining 25 patients, chemotherapy was not planned at the outset, but was used to treat a tumor that had recurred following initial radiation therapy. Because the group of 25 patients included neither those patients whose tumors never recurred nor those patients whose tumors recurred but were not treated with chemotherapy, issues of selection bias were of concern. We propose a test for selection bias based on a minimally selected $p$-value, analyzed via a refined Bonferroni correction derived by Worsley (1982). The test relies heavily on the presence of the randomly selected subsample of the study. We find there to be significant evidence of selection bias in the neuro-oncology study. Further, we propose estimators for the overall probability of selection and for the probability of selection given baseline predictors, and find that much of the selection bias can be explained by one baseline genetic feature. Lastly, we show that it is essential to adjust comparisons of treatment strategy for the selection bias; naive comparisons of response and of survival are significant, whereas properly adjusted comparisons are not. We assess the performance of the test and estimators in a simulation study.


Back to General Departmental Seminar Series

 

Internal Use | Site Map | Search |
Overview | People | Training | Research | Seminars | Employment | Links |
Biostatistics Program | Clinical Trials Program | Medical Informatics Program | Biomedical Computing |

Copyright © 2006 The Board of Regents of the
University of Wisconsin System

 

UW Madison UW School of Medicine and Public Health