The role of good questions

Fox MP, Edwards JK, Platt R, Balzer LB. The Critical Importance of Asking Good Questions: The Role of Epidemiology Doctoral Training Programs. American Journal of Epidemiology 2020;189(4):261–4.

This paper differs from the usual fare. As I recently left my relatively well-paying position to pursue an epidemiological doctorate, this paper jumped out at me.

Common comments I received when I announced my intention to go back to school included “but you don’t need a doctorate!” or “you’ve already learned everything [at work]”. Ironically these comments primarily came from PhD-level scientists. First, that is categorically untrue. Second, I would try explain what I dont know is hard to quantify – I don’t know how to conduct research, design studies, or ask good questions. So you can see why I wanted to read this paper.

Asking good questions

Good questions have a clear problem and a focus on relevance for the population. The authors frame epidemiological questions around an exposure-outcome pair, for which there are three main types: descriptive, predictive, and causal.

An important takeaway from this section: methods should follow from the question, not vice versa. Under the same treatment-outcome paradigm, the authors describe the underlying three possible questions:

1. Descriptive: Did people exposed have different survival rates than people who were not exposed?
2. Predictive: Can we predict the probability of survival as a function of baseline factors including the exposure?
3. Would patients survive longer, on average, if they received the exposure vs. not received?

For descriptive, we would not be concerned with controlling confounding. For causal questions, specififying or using the “causal roadmap” can help with study design and analytic approaches.

What to emphasize in doctoral training programs

I find this section interesting. Here, the authors assert that trainers often assume students entering doctoral programs have had training in asking good questions. Even with a Master’s and 6 years of work, my experience is the complete opposite. I feel as though I have a pretty good grasp on analytic methods themselves, but none on evolving or framing a study question. I have so many questions – how specific or general should I be? How do students have such specific populations or disease areas in which they want to study?

Rethinking programs

The article is definitely slanted towards how to ask good causal questions. In this section, the paper proposes students be taught to emulate target trials to estimate causal effects, which causes students to have to think about an intervention as well as potential biases.

It is also suggested that programs look into the breadth of their taught statistical methods: programs should continue to update their curriculum for modern approaches, but not at the expense of formulating good study questions.

A good point is that as students are “oriented to a questions-first approach, they require more training in such methods to be able to answer the relevant question”. However, I would posit that literature exists applying novel methods, but does not exist for asking good questions. Also, formulating good and interesting questions and drive novel methods development.

Finally, they propose an “apprenticeship” model for training epidemiolologists. First, you specify the problem. Then, you select the correct tool for each task under the supervision of an experience colleague. This differs from honing just a specific tool you use over and over again, and instead collect different tools for different problems.

Honestly, the above paragraph appears to be at odds with the idea of a dissertation, where we are supposed to focus in on a specific problem and/or method. Instead of highly specialized dissertations, what if we focused on broader case studies and study designs?