Causal direction matters: Fibromyalgia and napping

Causal direction matters. Statistical associations do not test causal direction. Like an algebra equation, statistical associations are bi-directional. But as a clinician, understanding causal direction really matters a lot, even when (if) the association is due to a truly bi-directional relationship. If you want to make recommendations about what to do, or what not to do; if you want to evaluate the severity of a sign or symptom, the causal structure and therefore direction matters.

Today I want to chat about this recently published paper:

Theadom, A., Cropley, M., & Kantermann, T. (2015). Daytime napping associated with increased symptom severity in fibromyalgia syndrome. BMC Musculoskeletal Disorders, 16(1), 13.

It is freely available as an open source article in the BMC Musculoskeletal Disorders journal. It is well written and overall a very interesting study. It is exploratory - and I have always enjoyed exploratory studies. It explores the association between napping and fibromyalgia (FMS) severity through it’s signs and symptoms.  Not just napping, but the frequency (within the sample and within a person) and duration of napping in patients; and as the authors tell us the are exploring “any associations” between daytime napping and symptoms of FMS. The study is cross sectional - each subject (of the ~1000 subjects) completed one survey, at one point in time about their FMS symptoms (including fatigue), and their general napping habits and tendencies.

In the discussion the authors point out that: “These findings suggests that napping behaviour may be more intrinsically linked to FMS symptoms than other demographic factors, however causality was unable to be determined within this study. This highlights the potential unique characteristics of napping behaviour in clinical populations.”

Agreed. Due to the study being observational and cross sectional they are unable to determine causality. Using entirely self report questionnaires does not help either since all variables are inherently related to one another by the person reporting them. But these methodological factors do not mean that there isn’t a causal structure underlying the findings; or that causal structure is an important reason that the authors are doing the study in the first place.

From reading the study there are appears to be two possible causal structures involving the core variables of interest, and they are not mutually exclusive.

The first is simply that FMS severity has a causal influence over the behavior of napping - that is FMS severity causes napping:


This is referred to in the paper several times, and they actually attempt to predict napping behavior based on FMS severity indicators. Whenever an investigator attempts to predict a (dependent) variable from a set of  (independent) variables, it is essentially their implicit statement that they believe there is a causal association (and they should expect causal associations).

The second is simply that napping behavior, which then has physiological implications, has a causal influence over FMS severity.


This is also implicitly referred to in the paper. In fact, it is implied very early on when justifying the underlying purpose of the study. “Despite the evidence of the beneficial effects of napping found in healthy adults in terms of improved alertness, emotional state and cognitive performance, the impact of napping in clinical populations has received little attention.” Here they are referring to napping as an intervention. They go on to speak about how clinicians are not sure what to recommend about napping to patients with fibromyalgia.

Clearly what we have here is either equivocation (napping is being used in a different sense in each causal structure) or there is a bi-directional causal structure. I believe that there is a bi-directional causal structure. Now - DAGs are “directed acyclic graphs” and they force us to consider the mechanisms at play. In my pre-DAG life I would have depicted a bi-directional causal structure with a cycle, but the benefits of embodying the causal structure in a DAG, in my opinion, is that it forces me to consider the fact that causation happens over time and the cycles can be unfolded across time and consideration of different states of a variable.


This DAG depicts a rather complicated causal structure that only includes two variables, but across time the two variables are unfolded and expected to be in different states. If FMS severity influences napping, and then napping influences FMS severity then that new state of severity (FMS_Severity2) will influence napping again. This process continues on. Now - if napping does not influence FMS severity then the red arrows are removed, essentially unlinking napping from the future states of severity.

There are many implications from such an anticipated causal structure for analysis of data collected from a study; for planning a study; and for clinical practice. Is napping purely a behavioral sign of the severity of FMS, and an indication of how much it impacts activity and participation? Or is napping a viable intervention, which minimally reduces the progression of severity, or possibly reduces the severity?

This post is getting pretty long - in upcoming posts I will attempt to add other variables to this causal structure (bi-directional, unfolding) and further discuss the paper.

Leave a Comment