Causal Structure Matters: Fibromyalgia and napping

In the last post (here) we considered causal direction with a particular example of fibromyalgia (FMS) and napping. We unfolded what could be considered a bi-directional causal relationship. By unfolding I mean taking a feedback loop and recognizing that time passes with each loop, so unfolding the loop across time creates a sequence of events, state changes in the variables involved.

With this post I want to present more of the causal structure based on variables included in the study by Theadom et al (2015).

Including all the variables in which they tested univariate (one variable at a time), and showed significant associations between “napping” and “non napping” in patients with FMS results in a very large model. I have also included possible interconnections between some of these variables - for example, co-morbidities with sleeping and other signs / symptoms; sleeping with fatigue and memory, etc.  The causal structure is quite complex:

dagitty-model-3

Note that FMS and napping are not, themselves, connected. This study did not assess whether there was a relationship between FMS and napping. It assumed there was a relationship and in a sample of only patients with FMS the had a lot that napped. But to really demonstrate an association between FMS and napping the study would have to have collected data on people without FMS. That is why I have not (and will not) include a directed edge between FMS and napping, it was not the studies intent to demonstrate that - it is simply assessing the intermediaries between FMS and napping within a sample of only patients with FMS.

This structure is too complex to attempt to unfold over time, but we would expect that napping would have an impact on some of these variables within a person. This study did not (could not) demonstrate whether napping has a restorative effect on fatigue. They had a large sample of patients with FMS and divided them by regular napping and not regular napping. They they demonstrated  patients with FMS that napped had greater fatigue, more sleep problems, more pain, etc. But we have no idea how much fatigue the nappers would have had if they did not nap. Even a relationship between severity of symptoms and napping within the napping group would not provide that understand. For that understanding we need an experimental design.

Taking the above and making it simpler by collapsing several of the tested variables into general categories would be helpful.

In the causal structure below (a bit more simple) the variables that are white are “adjusted for” in the study (observed). I have added one variable that has been shown to be related to FMS, and is easy to rationalize that it would be related to napping (and several other variables) and this is general class called social / environmental. This generally includes anything from the social setting, or environment. Job setting, stressors, time demands and commitments. These can impact FMS, and can certainly impact napping, I am thinking of reducing the ability to nap, for example. Social / environmental is “grey” in that it was not observed. But we see lots of red edges - these are biasing edges, confounding paths in an attempt to understand the relationships of the structure. To remove the red edges, you must observe (adjust for) the social / environmental factors.

dagitty-FMS

With this many variables causal structures can get overwhelming - hence the tendency to not explicitly include them. The models presented above do not even attempt to demonstrate the possible relationship between napping and possible variables (such as fatigue). But causal structures matter, they should influence the design, analysis and interpretation of studies. They form the basis of what we strive to call our “knowledge” for use in practice.

I will wrap up today with one possible alternative explanation to the findings of this study that people that napped more had more severe FMS based on the symptoms reported. h you the analysis they have done does not, cannot, rule out this hypothesis (remember, by rejecting a null you have not accepted your directional hypothesis, you have only rejected the null; many alternatives are possible that are just as commensurate with the data as your proposed directional).   What if the FMS patients that socially cannot nap, they have obligations that keep them active during the day, engaged, participatory, and they enjoy them so much they do not want to stop (yes, even some jobs can be like that) that being active and engaged and participatory reduces the symptoms of their FMS? How do we possibly test these alternative hypotheses?

This study has done a great job starting the conversation, it was a well done exploratory study and I hope the authors continue to pursue this line of inquiry. I look forward to their future work.

Leave a Comment