Re: Integrating Complex Systems Thinking into Epidemiologic Research : Epidemiology

Secondary Logo

Journal Logo



Integrating Complex Systems Thinking into Epidemiologic Research

Lofgren, Eric T.; Marshall, Brandon D.; Galea, Sandro

Author Information
doi: 10.1097/EDE.0000000000000680
  • Free

To the Editor:

We read with interest Dr. Ashley Naimi’s recent commentary1 about systems science. We welcome the increased attention to the issue but have a different perspective on several of the points raised. They are as follows.

The commentary characterizes systems science as a new field. The use of systems models to study infectious diseases was adopted and refined early in the 20th century,2 and for natural systems generally is far older. Computationally intensive models are recent; they no more make systems science an immature field than accessible Markov chain Monte Carlo procedures do Bayesian inference.

From this perspective, concerns about the inherent ambiguity of the term “nonlinear” take on a different light. In systems science, a “nonlinear system” has a clear meaning: a system where the output(s) are not proportional to any linear combination of its inputs.3 This meaning eliminates R1 through R3 of Dr. Naimi’s examples, all of which are linear systems, albeit with complex inputs, and with it the ambiguity. The imprecise and conflated use of the term Dr. Naimi points to is the unfortunate yet inevitable consequence of two fields with developed vocabularies encountering each other. Epidemiology is well acquainted with this phenomenon: the word “model” may be used to mean “conceptual,” “regression,” “mathematical,” an experiment involving Caenorhabditis elegans, “prototype,” and others. Nonetheless, we agree that those who employ this term should so with improved clarity.

We disagree with Dr. Naimi’s concerns about the exchangeability of systems models. He is absolutely correct that the inherent exchangeability of systems models is narrow and specific. Systems models substitute the exchangeability assumption for an assumption that the model’s representation of reality is accurate. There is a subtle, but critical, difference between these two assumptions. There are settings, especially for large-scale, multilevel questions, where no exchangeable population exists.4 Therein lies the primary niche for systems models. Attempting to address the question of model accuracy is, itself, a vibrant and active field within systems science and requires increased attention in epidemiology.

We agree with Dr. Naimi that more discourse needs to take place between those working with systems and causal inference models, and that systems thinking is not exclusive to either. However, we do not believe that this dialog will emerge from casting these techniques as mutually exclusive choices. Both techniques have advantages and disadvantages, and both have crucial roles to play in advancing population health.

Eric T. Lofgren

Paul G. Allen School for

Global Animal Health

Washington State University

Pullman, WA

[email protected]

Brandon D. Marshall

Department of Epidemiology

Brown School of Public Health

Brown University

Providence, RI

Sandro Galea

School of Public Health

Boston University

Boston, MA


1. Naimi AICommentary: integrating complex systems thinking into epidemiologic research. Epidemiology. 2016;27:843–847.
2. Smith DL, Battle KE, Hay SI, Barker CM, Scott TW, McKenzie FERoss, Macdonald, and a theory for the dynamics and control of mosquito-transmitted pathogens. PLoS Pathog. 2012;8:e1002588.
3. Rickles D, Hawe P, Shiell AA simple guide to chaos and complexity. J Epidemiol Community Health. 2007;61:933–937.
4. Lofgren ET, Halloran ME, Rivers CM, et alOpinion: mathematical models: a key tool for outbreak response. Proc Natl Acad Sci U S A. 2014;111:18095–18096.
Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.