Predicting patient outcomes from the sequence of their diagnoses
A systematic review of deep learning studies that predict outcomes using sequences of diagnosis codes, and how well they generalise and explain themselves.
The question
A patient’s record holds a sequence of diagnosis codes over time. Can deep learning use that sequence to predict what happens next, and can we trust those predictions to work elsewhere and to explain themselves?
What we did
We led a systematic review of the existing deep learning studies that use sequences of diagnosis codes as a key predictor of patient outcomes. We also examined how well these models generalise and how explainable they are.
What we found
Deep learning on diagnosis sequences shows remarkable promise. Using several types of features and including the time intervals between events improved prediction. The main remaining challenges are generalisation and explainability.
Why it matters
Tackling those two challenges is what will let such models improve health care outcomes in practice.