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 (besides all of my cardiopulmonary examples), 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 :) ) In short, the answer is yes, and I have several ideas of how this can move forward practically.
As I presented at CSM and explain in the voiceover of that presentation there are several steps - several things that we can do and that can change. They are all based on the generation, sharing, testing, modification and use of graphical causal models as representations of clinical knowledge.
Since the first step is generation of graphical causal models using DAGs I will spend time in the next several weeks identifying resources, teaching and providing examples of DAGs so that anyone with an inclination to start developing them specific to physical therapy practice can in fact start.
The overall plan, from the CSM presentation, is that for the use of DAGs:
As a framework for study development and interpretation of data as contributing to the causal model
To identify causal constraints for dynamic inference
For reasoning about signs and symptoms
For reasoning at the level of the causal model for diagnosis, intervention selection and explanation
For predicting changes from the causal model
Steps ahead include:
Build a repository of models based on clinical practice and current knowledge (can incorporate as part of CPG process but with a much broader approach including physiological and biomechanical mechanistic studies)
Publish reviews that include covariates based on DAGs, using DAGs to identify minimal adjustment sets (and abduction sets)
- Quantify complexity of clinical cases based on the minimal sufficient adjustment sets (and abduction sets) for clinically relevant unbiased causal inferences
- To be useful to practice, a study or systematic review should make explicit contributions to a clinically useful causal model
Overall, DAGs provide a framework for developing causal models as clinical knowledge, but also provide the logical structure to consider the dynamic inferential process used both with developing models through study, and using the models in practice.
As you can see - it all really starts with generation of causal models. So the first step for me here on the blog will be guide those willing in that process. There are three parts to learning how to generate a causal model. There is: 1. the syntax - the use of a system such as DAGitty with all of its features; 2. the semantics (or analytics), what different causal structures mean and why (the logical implications of the models); and 3. the translational (or syntopical) which is how to go from knowledge to a model, and from study results to a model - basically how to move from various sources of data, understanding, general premises, accepted universals to an interconnected causal model. These levels really align quite nicely with the system for reading set forth by Mortimer Adler and Charles Van Doren in their 1940 classic (and still in print) how to read a book (elementary, analytical, syntopical).
I will also start investigating approaches to systematizing model sharing through a repository system of some sort (i.e. GitHub, Zotero, or perhaps simply directly on DAGitty as it allows for online storage and sharing of DAGitty models) since that will be really important for the syntopical steps of generating a model.