Less time for blogging the next couple of weeks, so some shorter posts….
Pam Davis-Keen and Sarah Hart bring up an interesting question about genomic (or, for Davis-Keen, image-based) prediction: What if, for benign clinical purposes, you wanted to predict dyslexia in young children, so you could get services to them early? Surely there isn’t anything darkly eugenic about that.
I see what they mean. I am (at least officially) a clinician, and if I am wondering if a patient is schizophrenic or bipolar I would certainly be interested in knowing something about family history. But that is the point– that kind of clinical prediction is based on the phenotype, and how much would say a schizophrenia GPS add to that? Very very little (remember we are conditioning on the parental phenotype) and nowhere close to anything clinically useful.
So if what you really want is just raw prediction of dyslexia you might as well throw everything you have at it, and I am sure parental and child phenotypes, plus demographic things like SES, would play a much bigger role in the prediction equation than the GPS.
But the actual impulse to get the GPS actually comes from somewhere other than a simple desire to do a good job predicting, and this is where it gets tricky. We have an intuition that dyslexia is more than just a phenotypic phenomenon, that it has some deeper existence as a real thing, a Meehlian taxon, at a genetic or neurological level of analysis. How did Paige put it? We think that dyslexia is a biological construct, and therefore have an intuition that the dyslexia we identify with a GPS will be the real genomic thing, that it will get at the actual essence of dyslexia in a way that just predicting that parents with dyslexia are likely to have kids with dyslexia will not.
The question is how we are going to feel when our GPS starts identifying way more poor kids than rich kids as dyslexic (I presume rates of dyslexia are higher among the disadvantaged; the fact is I don’t know very much about dsylexia). Does that show that the reason disadvantaged kids are at increased risk is that they carry more dyslexia genes? Or does it mean that social risk is a ginger child route between the genome and the pathology?
So I guess my conclusion about this issue is that if you really mean it about only being interested in prediction (which I think is actually rare) and if a GPS is a useful predictor over and above phenotypic data (which they generally are not), then fine, go ahead and predict. But beware the essentialist conclusion that the GPS is getting you closer to the real deal, unless you have evidence (again, in the form of mechanism) that such is actually the case.
I have a chapter about defining psychopathology at lower and higher levels of analysis. Let me know if you are interested and I’ll send it to you.