[Physical Therapist] Before diving into complex analysis, I want to solidify the foundation of my research design

― Encountering a book that bridges the basics of epidemiology and causal inference ―

🔶 "Complex research" is becoming more common at academic conferences

Recently, when listening to conference presentations, I feel like I have more opportunities to see complex statistical methods and research designs I haven't heard of before.

Research methods are advancing day by day.

The emergence of new analytical methods and research designs, which allow us to approach phenomena that were previously invisible, is, of course, something to be welcomed.

Clinical research, public health research, and the utilization of medical data all need to evolve with the times.
In that sense, I think the attitude of challenging oneself with new methods is very important.

However, on the other hand, there are times when I want to stop and think for a moment.

🔶 How reliable is that conclusion?

What is important in research is not the use of complex analysis itself.

🔷 To what extent can the conclusions obtained from that research be trusted?

🔷 Is it reproducible when another researcher tries to verify it in the same way?

🔷 Does that research design really answer the question?

This is what is important.

Sometimes, while listening to a presentation, I think,

"Did they use a complex method because it couldn't be clarified with a simpler analysis?"

I sometimes find myself being suspicious.

Of course, complex methods themselves are not bad.
Complex questions sometimes require complex methods.

However, the reliability of research is not determined solely by the sophistication of the analytical method.

🔶 The strength of research is determined by the foundation of the design

Rather, it is important how carefully the foundation of the research is designed.

For example, how were the target samples obtained?
Is there any selection bias involved?
How was the data measured?
To what extent were confounding factors identified in advance, and how were they reflected in the research plan?

If these fundamental parts remain ambiguous, no matter how sophisticated the statistical methods used, there is a limit to the reliability of the conclusions.

It is difficult to completely remove distortions that occurred at the research design or data collection stage through advanced analysis later on.

An 'Ultra-C' (that's old... very Showa era) type of analysis method cannot save all the weaknesses of a research plan.

🔶 Novelty is not just about 'complexity'

In research, novelty is required.
That is a matter of course.

However, in the pursuit of novelty, one might feel compelled to choose unnecessarily difficult methods.

But novelty does not lie solely in using 'unheard-of analytical methods'.

Verify simple questions with a careful design and in an honest manner.

Valuing the selection of subjects, the method of data collection, the consideration of confounding factors, and the transparency of the analysis plan.
That accumulation is also a great value of research.

Science is an endeavor that grows stronger by being falsified.

↓↓ I have written an article about falsifiability

That is precisely why it is important that others can verify, reproduce, and critically examine the work.

🔶 Now is the time to return to the basics of epidemiology

I believe that those involved in clinical research and medical data research need to learn not only statistical methods but also the research design itself.

Is it descriptive?
Is it predictive?
Is it causal?

First, organize the type of question you are asking.
Then, consider which research design is appropriate. Furthermore, examine where confounding, selection bias, and information bias might enter.

↓↓ I have written in detail about adjusting for bias

If you skip this order, no matter how advanced the analysis you perform, the overall outlook of the research will be poor.

Modern causal inference is by no means something that exists beyond the basics. Rather, I feel it is a tool for clarifying the fundamental concepts of epidemiology and thinking more carefully about research design.

🔶 A book I want such people to pick up

There is a book I would like people with this kind of awareness of the issues to pick up.

It is "Introduction to Epidemiological Design: A Causal Approach to Health Sciences."

This book is an introductory text for learning the basics of epidemiology and modern causal inference, centered on research design.
Regarding "the basics of epidemiological thinking essential to medicine",

Just recently, a colleague asked me, "Jiro, what exactly is the 'epidemiology' you majored in?" Even working at a research center, the term "epidemiology" is not yet widely understood.

I believe this book serves as a bridge for clinical researchers, medical professionals, and students to move from the basics of epidemiology to causal inference.

🔶 What you can learn from this book

"Introduction to Epidemiological Design" is a practical introductory book for getting closer to causality

"Introduction to Epidemiological Design: A Causal Approach to Health Sciences" is not just a book that explains epidemiological terms and research designs.

The great appeal of this book is that it carefully details the know-how on "how to think in order to get as close to causality as possible," rather than just accepting observed associations as they are.

In epidemiology and clinical research, various indicators of association appear, such as risk, odds ratios, and incidence rates. This book also carefully organizes these indicators.

It is not so detailed that you get lost along the way, nor is it too superficial. If you can thoroughly understand this volume, I think it will become a very significant foundation for reading epidemiological research.

On the other hand, a fair number of mathematical formulas appear.

For those reading it for the first time, there may be moments where it feels a bit tough.
I myself cannot understand every formula fluently from the start.

However, I often read through it while having AI easily explain the formulas I don't understand.
Thinking about it that way, I believe the hurdle for mathematical formulas has been lowered considerably.

Until now, there may have been parts I just skipped over because "a formula appeared."

However, in reality, having formulas and simple calculation examples gives reality to abstract explanations. I felt it was a bit of a waste to just run away saying, "It looks difficult because there are many formulas."

What left a particular impression on me was the way of thinking about bias.

In research, various biases such as confounding, selection bias, and information bias become issues. However, the important thing is not that "you should adjust for everything."

What bias should be addressed?
What bias does not have a major impact on estimation?
What should be adjusted for, and what should not?

I wish this way of distinguishing things was shared more widely.

When looking at clinical research, I sometimes see explanations that say, 'We just adjusted for it using multivariate analysis.'

However, from the perspective of causal inference, it is not the case that the more you adjust, the better.
In some cases, adjusting can actually distort the estimation.

This book explains those basic concepts of causal inference in a way that connects them to research design.

While it can be read as an introductory book on epidemiology, I think it is more than just an introduction; it is a bridge to moving on to specialized books on causal inference.

For those who want to improve their ability to read research.
For those planning clinical research.
For those who want to relearn the concepts of research design before statistical analysis.
And for those who want to properly understand the difference between 'association' and 'causation'.

For such people, 'Introduction to Epidemiological Design' will be a very practical and valuable book.

This book is one that carefully shows you that path.

🔶Before moving on to complex analysis

Learning new methods is important.
Challenging yourself with complex analysis is, of course, also necessary.

However, before that, it is just as important to check the foundation of your research.

Is this question descriptive? Is it predictive? Is it causal inference?

Is this way of gathering subjects appropriate?
Is this way of collecting data appropriate?
Is this way of organizing confounding factors appropriate?

And, can this research design really answer what you want to know?

I believe that returning to these questions supports the reliability of the research.

Before complex analysis, solidify the foundation of your research design.

As a book to build the basic strength for that, 'Introduction to Epidemiological Design: A Causal Approach to Health Sciences' should be a valuable book for many clinical researchers and medical professionals.

🔶For those who find self-study difficult

mJOHNSNOW, an online school specializing in medical research and public health, to which I also belong.

In our reading group, we have been covering "What If," but starting in August, we have decided to move on to this book.

There is one month left until the reading group begins, so I would like to read it thoroughly in advance and participate with a solid understanding of the content.

↓↓ What is "What If"? You can download it here.

I feel the limits of self-study. However, I am not yet prepared to commit to going to graduate school. Above all, I want to verify a bit more whether this is truly a field I am interested in.

For such people, I think an online school is a very realistic option. Before making a big decision suddenly, you can first experience it as an entry point for learning. I think being able to use it in that way is also a major appeal.

↓↓ I am writing about new ways of learning that go beyond just graduate school.

🔸 Conclusion

Learning complex statistical methods and new research designs is undoubtedly important for future clinical research.

However, as a prerequisite, I believe it is necessary to have a basic epidemiological perspective on how to formulate research questions, how to select subjects, how to measure data, and which biases to be aware of.

Before proceeding to complex analysis, first solidify the foundation of your research design.
Think about associations as close to causality as possible.
For those who want to build the basic stamina for that, why not study epidemiology, starting with this book or others?