I last posted here just before I started a new job at Fred Hutchinson Cancer Center. Moving to Seattle and getting oriented at work kept me busy for several months, and I had neither energy nor ideas for posting any personal writing. Then, for a couple of years, I was occupied with new hobbies, becoming acquainted with the Seattle area, and some health issues. Finally I’ve found myself with energy and interest to write again.
When I was nearing graduation with an MS in statistics in 2017, my first job interview was at Fred Hutch. Seattle and Fred Hutch both were quite appealing — I would almost say it would’ve been a “dream job” at the time — but I wasn’t offered the position. That rejection may have been for the better because I ended up having valuable experiences at other institutions around the country. In any case, it is interesting that I arrived at Fred Hutch (though not with the group who interviewed me previously) six years later.
For the past three years, I’ve worked in a group that coordinates and provides statistical support for clinical research in early cancer detection. A large part of the work is for the NCI-funded Early Detection Research Network (EDRN). My part has been to carry out statistical analysis and to assist with data collection and data management. For EDRN, most of the statistical analysis consists of evaluating biomarkers intended to be used as cancer screening tests or as diagnostic tools to complement existing screening tests — ROC curves and AUCs ad nauseum, among other things. The group at Fred Hutch helps design and implement EDRN studies to collect biospecimens, medical imaging, and clinical data from medical centers across the US. The biospecimen and image repositories then serve as resources for investigators working to develop or validate biomarkers and algorithms for detecting cancer.
Before coming to Fred Hutch, I had little experience with cancer research. It took a year or two to become mentally oriented to the specific problems of cancer detection. Now I’ve worked on enough projects to have a decent understanding of the clinical and statistical objectives and constraints in this area, which has led me to have some thoughts on particular projects and methodology. I’ve also continued to be interested in statisticians’ training and mindsets as I’ve observed how I and other statisticians have successes and failures in our contributions to research teams. I plan to write on these and other topics in the coming months.
Three particular topics I want to work on are the following:
- In 2022–23, I wrote a paper on the potential for a professional doctorate degree in statistics, an idea that came to me while I was a PhD student in biostatistics and thinking about the differences between training of biostatisticians and medical doctors. I submitted the paper to a couple of journals but didn’t get around to revising it because my attention turned to my new job. I think the paper could be heavily revised and re-submitted now, and I may have new ideas about how to present and further explore the idea.
- In the same vein, I would like to develop and write about my views on what should be the ideals among statisticians and what makes a person effective as a statistician.
- I’ve been attracted to Bayesian analysis from the beginning of my career in statistics, but I’ve hardly used it at all. Recently, I’ve been thinking about Bayesian methods for two reasons. First, we are using a Bayesian model in a clinical trial where I have an opportunity to see advantages and disadvantages, especially when the analysis is interpreted by clinicians. Second, I’ve encountered a question in cancer screening methodology — whether a screening test should incorporate risk factors or not — that seems loosely related to the use of frequentist analysis based on p-values and p<0.05 versus Bayesian analysis in routine research.