Introduction to epidemiology for clinicians

The website of the Department of Epidemiology, Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences((Department of Epidemiology, Okayama University Graduate School of Medicine, Dentistry and Pharmaceutical Sciences. Accessed March 12, 2021. http://www.unit-gp.jp/eisei/wp/?page_id=298)) introduced an article series titled "Epidemiology Series for Clinicians." Unfortunately, perhaps due to specification changes on J-STAGE, all the links were broken.

As an aside, Hiroyuki (Hiroyuki Nishimura)((Hiroyuki, Hiroyuki Nishimura (@hirox246) / Twitter. Twitter. Accessed March 12, 2021. https://twitter.com/hirox246)) recently introduced an interview video with So Takei((So Takei (@sosotakei) / Twitter. Twitter. Accessed March 12, 2021. https://twitter.com/sosotakei)) on the YouTube channel "Gairoku ch," remarking that "you can watch it for free and it's more useful than a mediocre self-help book, so it's a shame that people aren't finding the information"((Hiroyuki, hiroyuki. [Hiroyuki] Monkeys of the same stature play together 2. Drinking WALBERGEM. 2021/02/21 D22.; 2021. (from 27:30) Accessed March 12, 2021. https://youtu.be/qW7EN1F5SW8?t=1650 )).

I felt it would be a waste if this epidemiology series remained buried, so I have relinked the articles here. Please refer to the full text through each DOI link.

Epidemiology series for clinicians

Part 1: Epidemiology and statistics in clinical research (General epidemiology)

Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 1: Epidemiology and statistics in clinical research (General epidemiology). Journal of Japanese Association for Acute Medicine. 2009;20(5):288-293. doi:10.3893/jjaam.20.288

Awareness of the importance of clinical research is steadily growing in Japan, yet clinical studies conducted mainly at regional core hospitals remain few in number, and in terms of quality, almost none can withstand publication in English-language journals. This occurs because many clinicians place too much emphasis on statistical knowledge and lack a sufficient epidemiological perspective. Good examples of this include a tendency to focus heavily on hypothesis tests and p-values, insufficient consideration of bias, and failing to quantitatively estimate the effects of treatments or exposures. When conducting clinical research, statistical knowledge—used for understanding populations and samples, calculating sample sizes, comparing baseline characteristics, and adjusting for confounding factors through multivariate analysis—is certainly helpful to have, but many clinicians do not understand that executing clinical research requires a considerable degree of epidemiological knowledge. In particular, when clarifying research hypotheses, setting control groups, building analytical models, and examining study results and biases, high-quality clinical research cannot be conducted without an epidemiological perspective.

Part 1: Epidemiology and statistics in clinical research (General epidemiology)((Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 1: Epidemiology and statistics in clinical research (General epidemiology). Journal of Japanese Association for Acute Medicine. 2009;20(5):288-293. doi:10.3893/jjaam.20.288))

Part 2: Reviewing epidemiological terms and how to read papers (Specific topics in epidemiology 1)

Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 2: Reviewing epidemiological terms and how to read papers (Specific topics in epidemiology 1). Journal of Japanese Association for Acute Medicine. 2009;20(6):338-344. doi:10.3893/jjaam.20.338

How to read papers (critical appraisal) is a fundamental skill for conducting research, and in recent years it has also become a required competency for clinicians practicing EBM. Very many clinicians struggle with reading papers, and the authors believe that studying basic epidemiological knowledge and how to write papers may be the shortest path to acquiring the skill of reading them. As basic epidemiological knowledge, it is vital to understand the epidemiological measures and effect measures needed to quantitatively estimate the impact of an exposure (treatment) on an outcome, as well as the study designs needed to consider bias. Meanwhile, regarding how to read papers, once you understand the paragraph structure of a paper and can practice "paragraph reading"—reading paragraph by paragraph—your reading efficiency improves dramatically. And reading papers while referring to various reporting guidelines on how to write them is the best training.

Part 2: Reviewing epidemiological terms and how to read papers (Specific topics in epidemiology 1)((Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 2: Reviewing epidemiological terms and how to read papers (Specific topics in epidemiology 1). Journal of Japanese Association for Acute Medicine. 2009;20(6):338-344. doi:10.3893/jjaam.20.338))

Part 3: Formulating research hypotheses and Directed Acyclic Graphs (Specific topics in epidemiology 2)

Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 3: Formulating research hypotheses and Directed Acyclic Graphs (Specific topics in epidemiology 2). Journal of Japanese Association for Acute Medicine. 2009;20(7):397-403. doi:10.3893/jjaam.20.397

It is no exaggeration to say that the quality of the "research hypothesis" determines the quality of clinical research. However, little literature describes how to formulate and refine a research hypothesis, and many clinicians conduct clinical research based on inadequate hypotheses. When formulating a research hypothesis, it is most important to properly define the participants, exposure (intervention), control, and outcome, and to thoroughly examine and discuss them based on a counterfactual model. Building on that, it is important to develop research and analysis plans and examine confounding factors that require adjustment based on a Directed Acyclic Graph (DAG), which has come into use in epidemiology in recent years. Once you understand a few simple rules, a DAG is an extremely useful tool that helps both when conducting clinical research and when reading papers.

Part 3: Formulating research hypotheses and Directed Acyclic Graphs (Specific topics in epidemiology 2)((Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 3: Formulating research hypotheses and Directed Acyclic Graphs (Specific topics in epidemiology 2). Journal of Japanese Association for Acute Medicine. 2009;20(7):397-403. doi:10.3893/jjaam.20.397))

Part 4: Thinking about bias and interpreting results (Specific topics in epidemiology 3)

Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 4: Thinking about bias and interpreting results (Specific topics in epidemiology 3). Journal of Japanese Association for Acute Medicine. 2009;20(9):794-800. doi:10.3893/jjaam.20.794

Association is not causation. This is a famous phrase, and bias is defined as occurring when association and causation differ. The counterfactual model and directed acyclic graph (DAG) introduced in Part 3 of this series are extremely useful tools for understanding and organizing bias. It is useful to organize confounding bias as arising from a common cause in a DAG, selection bias as arising from adjusting for a common effect in a DAG, and information bias through an understanding of "non-differential misclassification." Meanwhile, when interpreting results, it is important to use a 2×2 table to examine in which direction and to what extent a bias shifts the "estimate" away from the "true value"—that is, whether the bias causes overestimation (away from the null) or underestimation (toward the null).

Part 4: Thinking about bias and interpreting results (Specific topics in epidemiology 3)((Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 4: Thinking about bias and interpreting results (Specific topics in epidemiology 3). Journal of Japanese Association for Acute Medicine. 2009;20(9):794-800. doi:10.3893/jjaam.20.794))

Part 5: The role of statistics in clinical research (Specific topics in epidemiology 4)

Hirokazu Komatsu, Etsuji Suzuki, Hiroyuki Doi. Part 5: The role of statistics in clinical research (Specific topics in epidemiology 4). Journal of Japanese Association for Acute Medicine. 2009;20(10):851-859. doi:10.3893/jjaam.20.851

Regarding the statistical knowledge required for clinical research, the fundamentals are first being able to sort out which statistical test to use in which situation when comparing baseline characteristics, and understanding logistic regression analysis as a multivariate analysis used to adjust for confounding factors. In addition, because many clinical studies use survival time as an outcome, it is also necessary to acquire knowledge of survival analysis. When comparing baseline characteristics, you determine which test to use by understanding whether each variable is a "continuous variable" or a "categorical variable," and whether you are comparing "2 groups" or "3 or more groups." Logistic regression analysis is often used when the outcome is binary, and can estimate the effect of an exposure as an odds ratio. Meanwhile, the Cox proportional hazards model used in survival analysis can estimate the effect of an exposure as a hazard ratio. To acquire knowledge of clinical research amid busy clinical practice, studying epidemiology thoroughly first and then building up the necessary statistical knowledge is likely the most efficient approach.

Part 5: The role of statistics in clinical research (Epidemiology specifics 4)((Komatsu H, Suzuki E, Doi H. Part 5: The role of statistics in clinical research (Epidemiology specifics 4). Journal of Japanese Association for Acute Medicine. 2009;20(10):851-859. doi:10.3893/jjaam.20.851))

Causal inference in medicine (Review articles)

Causal inference in medicine: Part I

Suzuki E, Komatsu H, Yorifuji T, Yamamoto E, Doi H, Tsuda T. Causal inference in medicine: Part I. Japanese Journal of Hygiene. 2009;64(4):786-795. doi:10.1265/jjh.64.786

Finding general laws of cause and effect—that is, causation—and formulating them into theory is a central task across the natural sciences, not just in hygiene, epidemiology, or medicine (1). The natural sciences build on physics, chemistry, and biology, and medicine is one of their applied fields. Medicine can therefore be described as a discipline that treats causal relationships with human beings as its subject of observation. In clinical medicine, Evidence-Based Medicine (EBM) rests on "evidence" that presupposes a causal relationship between treatment (cause) and treatment effect (effect); if causal relationships did not exist in the first place, the very existence of EBM could be called into question. Similarly, how to infer causal effects is a central problem in hygiene and public health (2), and scientific knowledge of causal theory is necessary when making policy decisions (3). Causal inference is thus an important task across various fields of medicine, and epidemiological theory is what makes it possible. Epidemiology is the discipline that collects and analyzes human data, providing the concrete methodology and logic for causal inference in etiology or disease prevention (4). Indeed, the primary purpose of epidemiological research is to elucidate causation or causal relationships themselves; except for classical epidemiological studies aimed at surveillance, all epidemiological studies aim at causal inference (2-4). However, researchers have pointed out problems with causal inference in many epidemiological studies. Among these problems, this paper takes up the failure to clearly define research hypotheses from the standpoint of causal theory (5), and examines this point using the counterfactual model. Applying this model should clarify discussions in practice and research in hygiene and public health. Note that the discussion here, including the counterfactual model, addresses the problem of causal relationships across the natural sciences wherever outcomes are treated as events, and is not limited to medicine and epidemiology.

Causal inference in medicine: Part I. Japanese Journal of Hygiene.((Suzuki E, Komatsu H, Yorifuji T, Yamamoto E, Doi H, Tsuda T. Causal inference in medicine: Part I. Japanese Journal of Hygiene. 2009;64(4):786-795. doi:10.1265/jjh.64.786))

Causal inference in medicine: Part II

Suzuki E, Komatsu H, Yorifuji T, Yamamoto E, Doi H, Tsuda T. Causal inference in medicine: Part II. Japanese Journal of Hygiene. 2009;64(4):796-805. doi:10.1265/jjh.64.796

In medicine as a whole, not only in hygiene, identifying, validly investigating, and inferring causal relationships (causality) regarding health outcomes such as disease is a central task (1). Valid causal inference requires defining research hypotheses based on a clear causal theory (2). In doing so, it is important to properly understand the counterfactual model in causal inference and define the research hypothesis in a way that aligns with the study's objective (2). Testing a clearly defined research hypothesis then requires using an appropriate study design to minimize systematic and random error. In most cases, exposure to a specific disease risk factor is known to be associated with exposure to multiple other disease risk factors. Consequently, when inferring the causal effect of a specific risk factor on a specific disease, one must consider the possibility of confounding bias. Because randomly assigning exposure to disease risk factors is impossible in most cases, how to minimize confounding bias has been discussed as a primary concern. However, the more complex the situation under consideration, the harder it becomes to present research hypotheses—including confounders and covariates—clearly and accessibly. It also becomes difficult to properly select the covariates to adjust for when building a model. Adjusting for too few or too many covariates makes it impossible to capture a valid causal effect, making this an important problem in causal inference (3). Such problems stem from failing to clearly define research hypotheses through thinking grounded in causal inference, and from failing to present those hypotheses clearly. To present research hypotheses clearly and accessibly, it is useful to employ a model distinct from the counterfactual model: the graphical model. Focusing on directed acyclic graphs (DAGs, usually pronounced "dags") among graphical models, this paper shows how graphical models are used routinely in causal inference (4-9). Among causal inference models, the DAG is recognized as a simple tool that facilitates discussion among researchers, and since the turn of this century it has been used frequently at international conferences and research meetings. Graduate schools of public health, including those at Harvard University and the University of North Carolina, now teach DAGs as part of epidemiological theory. In addition, the third edition of Modern Epidemiology, published in 2008, devotes an entire chapter to DAGs as an overview (4) (Footnote 1). Using DAGs allows researchers to respond flexibly to protocol revisions and helps make study results easier to understand. In Japan, however, DAGs are hardly being used to their full potential, and there are no review articles introducing DAGs in the medical field. This paper outlines the basic knowledge needed to construct DAGs and shows that they are useful and simple for selecting and organizing confounding factors and for making hypotheses concrete in causal inference. It also raises the problems of entering covariates into a model as potential confounders without sufficient consideration (10).

Causal inference in medicine: Part II. Japanese Journal of Hygiene((Suzuki E, Komatsu H, Yorifuji T, Yamamoto E, Doi H, Tsuda T. Causal inference in medicine: Part II. Japanese Journal of Hygiene. 2009;64(4):796-805. doi:10.1265/jjh.64.796))

So Takei videos (Gairoku ch)

Part 1

Gairoku ch - Tell me about your life -. So Takei at 39, his days as an unknown newcomer / A once-in-a-lifetime gamble in front of SMAP's Masahiro Nakai /.; 2020. Accessed March 12, 2021. https://www.youtube.com/watch?v=sWR92b6Qd2U

Part 2

Gairoku ch - Tell me about your life -. So Takei lying low in a Nishi-Azabu bar... / With Takaaki Ishibashi, Gokuraku Tombo's Yamamoto, and M-1 champion Untouchable's Shibata /.; 2020. Accessed March 12, 2021. https://www.youtube.com/watch?v=JqkhCxka-Vg

Part 3

Gairoku ch - Tell me about your life -. So Takei's history with Shinobu Sakagami / Returning to the entertainment industry for revenge, carrying the wishes of his older brother who died young of terminal cancer.; 2020. Accessed March 12, 2021. https://www.youtube.com/watch?v=PcfMClWeLf4