Spurious correlations are common in medical models not idiosyncratic like in other fields. Which makes it hard to use models in clinics. Doctors are smart abo

Spurious correlations are common in medical models not idiosyncratic like in other fields. Which makes it hard to use models in clinics. Doctors are smart about this. Every providier we spoke to about survival models based on biomarker data did not want supervised learning models. The explainability, trust, authorization just aren't there and the risk of misuse is too high. For now, research is a better use case, but less commercial funding. It is not hopeless, but your basic LSTM is probably not going to revolutionize medicine.