For the next several weeks I plan to use posting on the blog as a platform for teaching the concept of an adjustment set. It is important to keep in mind that this concept is already well established in the literature of directed acyclic graphs of causal models, and is a feature in the platform DAGitty ( I think it is best to use an example and I will start with an example I am familiar with and is not mired in the technicalities and complexities of a physiological DAG. The emphasis of these posts is to teach the use of DAGs with this model as an example, not to teach about the causal structure of HRQOL in patients with HF (although the latter will occur a bit).

Students came to my office the other day (I forgot to ask permission to use their names) and were asking whether KBP as an approach could be applied to other areas of PT, and how I saw it moving ahead practically speaking - i.e. assuming the theoretical foundations are acceptable to people, what does it mean, what changes, what happens? (I am paraphrasing of course - or at least presenting what I heard :) ) As I presented at CSM and explain in the voiceover of that presentation there are several steps - several things that can change, but as I mentioned in the previous post, they all are based on the generation, sharing, testing, modification and use of graphical causal models as representations of clinical knowledge.

Therefore, an important step for KBP is to teach about these models so others can start generating and sharing them for testing and modification. The DAG at   is based on the paper by Krethong et al (2008). For this model I have simply identified “Functional Status” as the primary cause we are interested in investigating and we are interested in its effect on health related quality of life (HRQOL). DAGitty automaticallydagitty-model colors the graphic with red indicating a biasing pathway (in other words, not knowing the information in red confounds our ability to draw a causal inference (inductively) from functional status to HRQOL (assuming the causal structure is correct). DAGs are a representation of a causal network and they allow us to visualize the causal structure, or at least our understanding of the causal structure and our assumptions. A causal path in a DAG starts at cause and contains arrows (directed) toward the effect (outcome). A causal path can have several  intermediary steps (a causal chain). The above DAG has a direct causal path between Functional Status and HRQOL. Bio/Physio Status  has two causal paths with HRQOL, one that passes through Functional Status and one that passes through Symptom Status. There is also something more complicated going on with Bio/Physio Status as there is a causal path from Bio/Physio Status to Symptom Status to Functional Status to HRQOL. Take a look at the post called Exacerbating Factor - 2 , where we see the classical condition of confounding. Based on this structure, Bio/Physio Status can be considered a confounder in the relationship between Symptom Status and Functional Status, and could therefore be an exacerbating factor or a possible eliminator (if you recall these definitions of cause included the same basic causal structure, they were words we give based on what we observe or expect to occur for a particular confounder).

If you go to the link you can see in the top right window that symptom status and social are the minimum adjustment set. If you put your mouse over these nodes of the graphs and type “a” to adjust for them you will notice that all the red arrows turn black. The minimum adjustment set has provided the set of variables that must be adjusted on (we need information about them) to make a causal inference free of confounding. Notice we do not need to adjust for Bio/Physio Status since if we know Symptom Status and Functional Status we know both consequences of Bio / Physio Status. So, while Bio / Physio Status remains a variable we do not know about, it does not confound our assessment between Functional Status and HRQOL.

More to come…..