Seminar Series: Karen Kopciuk, PhD
Development and external validation of multivariable risk prediction models for cancer stage at diagnosis among males and females in Canada
Karen Kopciuk, PhD
Adjunct Associate Professor
Department of Mathematics and Statistics
Faculty of Science
Departments of Oncology and Community Health Sciences
Cummings School of Medicine
University of Calgary
Research Scientist
Cancer Epidemiology and Prevention Research
Alberta Health Services
Cancer Care Alberta
Short Biography:
Karen Kopciuk is a Research Scientist in the Department of Cancer Epidemiology and Prevention Research, Cancer Care Alberta, Alberta Health Services and is an Adjunct Professor at the University of Calgary (UC) in the Departments of Mathematics and Statistics, Oncology and Community Health Sciences (CHS). She is an active in a number of roles at the University of Calgary (UCalgary Biostatistics Centre executive, Rocky Mountain Data Science Network), with the Statistical Society of Canada (Awards, Women in Statistics) and is a board member of the Canadian Statistical Sciences Institute. Her research interests in statistical methods include time-to-event models and variable selection methods with applications in genetics and cancer and collaborative research in cancer screening, particularly with Indigenous partners.
Abstract:
Multivariable risk prediction models are being increasingly developed for diagnostic or prognostic purposes. However, before they are adopted by clinicians, these prediction models need to be validated on data not used in their development. This study used AB Tomorrow Project (ATP) data to develop sex-specific prediction models for cancer stage at diagnosis using partial proportional odds models. These prediction models were externally validated on the British Columbia Generations Project (BCGP) cohort using discrimination and calibration measures for ordinal responses.
Among the ATP participants in the development models, older age at diagnosis was associated with an earlier stage at diagnosis, while full- or part-time employment, prostate-specific antigen testing, and former/current smoking were associated with a later stage at diagnosis among males. Among females, mammogram and sigmoidoscopy or colonoscopy were associated with an earlier stage at diagnosis, while older age at diagnosis, number of pregnancies, and hysterectomy were associated with a later stage at diagnosis. Using the BCGP data to validate these results, the discrimination measure (ordinal c-statistic) for males and females was 0.58 and 0.53, respectively. Calibration slopes and intercepts for males and females per dichotomization of stage did not differ significantly from 1 and 0, indicating the model not over or under fit to derivation data and does not over or under predict risk, respectively. The estimated calibration index (ECI) measures the flexible calibration plots’ deviation from the diagonal and is desired to be close to zero. The corresponding ECIs for females and males were 0.32 and 1.11, respectively, indicating that the risk predictions for females were more moderately calibrated than those for males.
Although we identified important factors associated with cancer stage at diagnosis with the ATP data, our prediction models showed poor discrimination with the BCGP data. Models were calibrated in the mean with the model for females better calibrated than for males. Prediction models that identify risk factors associated with cancer stage at diagnosis can help identify individuals at higher risk of developing late-stage cancer and thus greatly improve cancer survival. Further research is needed to develop more robust prediction models.
Keywords:
multi-state models, survival data analysis, genetic risk, statistical genetics, early cancer detection, cancer screening
Date: Friday, January 20th
Time: 1:30 pm - 2:30 pm
Location: PHFM 3015 (Western Centre for Public Health and Family Medicine)
