Causation for dummies – wot no maths – is how I am going to label this recent book. I have only just started reading it but the clear link with issues that arise in clinical practice, research and medicolegal practice are evident in the first few pages.
Many of the examples discussed in the book are medical issues such as the randomised controlled trial.
It is about causal inference.
This is something I often struggle with in diagnosing children with neurological conditions particularly children with relatively non-specific or prevalent features. For example, in clinical practice, I increasingly receive molecular genetics reports for children I have investigated that do not fully exclude or conclude that a particular detected gene variant is “the cause” of the child’s condition (the dreaded “variant of uncertain significance”).
From the clinical end, I often then have to look again in more detail at the child’s condition (the phenotype), perhaps do some other biochemical and physiological investigations, to see if there is something more specific about thetthe child’s condition to make the variant into something of less uncertain significance.
In medicolegal practice, this issue often comes up where a child has relatively prevalent neurodevelopmental problems such as borderline cognitive and academic abilities, or attentional and behaviour difficulties, which the parents ascribe to a salient event in pregnancy, delivery or childhood but usually is seen in children with no such adverse events. What I often find on looking at the salient event in more detail is that the event could potentially (hypothetically) cause permanent injuries but usually does not.
What I then have to do is to look for more specific findings – usually the neuroimaging – for more evidence that the salient event was causative.
Another interesting concept raised in the book is the “counter factual” – the what if? scenarios. This is called the “but for” question in medicolegal practice. As the author points out, this requires a causal model that works forward from baseline observations.
A concept new to me introduced in the book is the “estimand” and whether it can be solved with the observations we can access. Usually it is easy to look at data we are presented with, which tends to be things which are easy to measure, rather than what is in the causal model. For example having to rely on heart rate and blood pressure measurements to time the onset of achte hypoxic ischaemic brain injury rather than a real time high spatial and temporal resolution non invasive biochemical measure of neural cell death.
Unfortunately in real life its harder than that since we often have multifactorial contributions to causation – which include environmental and social factors – rather than a single overwhelming cause.
I will update with my new found wisdom when I have finished reading the book.