Web EpiMonitor

News of the publication of the 6^th edition of The Dictionary of Epidemiology reached our offices this month. From the promotional material for the book, we learned that the 6^th edition reflects multiple changes in the thinking about and practice of epidemiology. The Dictionary has always been a window on the world of epidemiology, so we took advantage of the new release to interview Miquel Porta to get his views about some of the latest trends.

EM:  You mention that a methodological "revolution" is ongoing and deeply changing how we think about research and assess validity of findings in epidemiology. Not everyone will know what you mean by this statement. Can you expand on your statements a bit more for our readers? 

Porta: In my view, a methodological “revolution” is  ongoing. That is, fundamental changes in theory, concepts and practice are taking place. This renewal is deeply changing how we conceive epidemiological and clinical research, and how we assess the validity of findings.

I think a good way to understand what is happening is to read, think, and assess the changes in the definitions of terms as, for example, collider, M-bias, causal diagram, backdoor (biasing path), instrumental variable, negative controls, inverse probability weighting, attributable fraction, bias, selection bias, confounding, residual confounding, interaction, cumulative and density sampling, Berkson’s bias, Simpson’s paradox, overadjustment, identifiability, transportability, positivity, ignorability, collapsibility, exchangeable, g-estimation, and marginal structural models.

 Other important changes are occuring on concepts such as risk, rate, risk ratio, risk set, open population, test hypothesis, null hypothesis, causal null, causal inference, generalizability, representativeness, missing data, standardization, frequentist statistics, immortal time bias, nonmonotonic, potential outcome, sample space, or false discovery rate.

 Of course, a dictionary is not a textbook. But a new generation of textbooks is emerging, and two good examples probably are:

--- Hernán MA, Robins JM. Causal inference. New York: Chapman & Hall / CRC; 2015.

--- VanderWeele TJ. Explanation in causal inference: Methods for mediation and interaction. New York: Oxford University Press; 2015.

 Another good way to assess what is changing is to follow the work of leaders such as James Robins, Miguel Hernán, Sander Greenland, or Tyler VanderWeele.

EM:  You also state that the revolution is having a very big impact on research production and use. Can you give examples or say more about these consequences?

Porta: Pick up any issue of the main epidemiologic journals and you will find several examples of what is going on. All these titles come from just one such issue:

--- Causal models and learning from data: integrating causal modeling and statistical estimation.

--- Methodological challenges in Mendelian randomization.

--- Obesity paradox: conditioning on disease enhances biases in estimating the mortality risks of obesity.

--- Meat intake and reproductive parameters among young men.

--- Education and cause-specific mortality: the mediating  role of differential exposure and vulnerability to behavioral risk factors.

--- Racial residential segregation and preterm birth: built environment as a mediator.

I suspect that the ongoing changes are having a high impact on the production of scientific evidence in the health sciences, but so far nobody has quantified such an impact, as it is natural.

It is also understandable that not everybody among the senior generations seems to be aware of this. Exceptions abound, of course. They include James Robins and Sander Greenland themselves, as well as Jan Vandenbroucke and Allen Wilcox, for example.

One substantive example may be that some “obesity paradoxes” are disappearing as many “apparent paradoxes” had an explanation. In this case, it seems, conditioning on a collider was introducing a selection bias.