Understanding whether the model is capturing biology is critical when thinking about applying it to IVF. If the model is primarily capturing socioeconomic correlates, those factors will (in most cases) be fixed for all embryos from a given pair of parents. The PRS needs to be weighting _biological_ risk conditioned on a fixed environment if its to be used ethically in this context.
Nature can still be important for even highly heritable traits. For example, the size of a person's vocabulary is highly heritable, but a feral human will have very limited vocabulary. Such traits still have a very strong biological basis.
Do you think no biological basis exists? Either way, they are still correlates, even if some aren't causative.
These models are still predictive when tested against control groups outside the training population.