I have one question regarding my career. When I studied at school, I learned classical stuff, such as linear/logistic/Poisson regression, sampling, and so on. No ML, no causal inference, no clinical trial. I also learned Bayesian stats, but my level was very low. After graduation, I have been working as a biostatistician. You could imagine what I have been doing. I would like to hear your advice as to what I should learn and focus on. Most probably in the future I will continue to provide statistical consultation to faculty, fellows, and residents of say, some School of Medicine. I am not sure, but it seems that right now I am very interested in clinical trial.
It’s a very tough question, because I question everything about the usual biostatistics curriculum.
Everything I know about that biomedical researchers should learn about biostatistics, other than clinical trial design and predictive modeling, is in BBR. Almost everything I know about predictive modeling and model validation is in RMS. Lots of miscellaneous topics are covered in Statistical Thinking.
I think we should teach Bayesian methods before teaching any frequentist methods, and a champion for that is Richard McElreath through his book Statistical Rethinking. It’s a rocky road because of the majority of researchers clinging to the classical methods they were first introduced to, even though they never understood them (as witnessed by how often they misinterpret p-values and power calculations and multiplicity).
It’s time to use Bayesian methods in all consultations, and to go to the trouble of working with collaborators to understand why. Getting backup through this site, stats.stackexchange.com, and the Stan discourse group will help.
Related question, let’s say a clinician is ready to take the leap and undergo formal university level training in biostatistics with an emphasis on clinical trials and medicine, would you recommend any particular program or university?
I feel now that researchers begin to snub classical stuff, and they lean toward more advanced stuff, such as causal inference, ML, and Deep Learning. So I am wondering if my toolkit (I think I have been basically a regression-type of guy) is outdated.
I noticed that some researchers actually did not know why they wanted to perform causal inference (matching, weighting, IPTW, and stratification). In my eyes, they just followed suite.
Right now I am learning by myself clinical trial (say, Bayesian Adaptive Design) and causal inference like crazy. I am tired of being judged as “you are just crunching numbers”, which occurred during one of my interviews. BTW, after this interview, I went to check one of this particular interviewer’s publications, which is from 2023 to find out how good he is. Oh, my. He still dichotomized continuous variables. I came to feel that he must have not read your blog articles where you was strongly opposed to this approach. I see. That is his understanding of “crunching numbers”: run analysis without careful thinking and do whatever is convenient.
But anyway, I want to take my biostatistician skills to the next level.
Regardless of anyone’s professional advice, sounds like some arrogant people are projecting their own imposter syndrome onto you. Don’t let that get to you
No one know’s everything, if you don’t believe me just follow the discussions here. Some of the brightest statisticians I know write here and they are confident enough to admit when they are not familiar with a specific methodology.
How can you do any Causal or ML without understanding stats and regressions?
What are your current resources? Bayesian Adaptive Design is quite advanced.
Also - what exactly would you like to do?
The current state of statistical practice is quite depressing to me, and as implied earlier adds to the difficulty of finding good educational programs or teachers. I do not accept it when an experienced statistician makes the mistakes I made 40 years ago. (I want to always be making new mistakes!). But I’m sure there are excellent, modern biostat programs in some places; we just need to compile them. It’s easier to look for the best role models, e.g. Stephen Senn, Niki Best, Richard McElreath, etc. For the non-statistical part there is an excellent clinical investigation training program at U. Cal San Francisco.
For learning BDA, I just googled “Bayesian Adaptive Design” and used whatever I got. I will not take any courses. In the future, I would like to become an experienced biostatistician.
