SNOMED CT prediction has been evaluated only on the International edition using American English clinical notes. The model may not perform equally well in other languages or country editions. We encourage you to evaluate it against your own data and use case. If you notice anything that doesn’t look right, please let us know so we can improve it.
Upcoming improvements
This is our first release of SNOMED CT, and we will continuously improve it. We are currently working on the following improvements:- Reducing over-coding of procedures that aren’t clinically relevant to the encounter.
- Predicting a negation attribute, so each coded finding or disorder indicates whether it is affirmed or denied. For example, “the patient denies fever” will carry a negation attribute rather than being coded simply as “fever”.
- Predicting the subject of each code, so you can distinguish who a condition applies to. For example, “diabetes” will indicate whether it applies to the patient or to a family member.
- Predicting temporality, so each code indicates whether a condition is current or historical. For example, distinguishing an active myocardial infarction from a past one.
We have only evaluated this implementation on data from a limited set of specialties. This may not generalize to all hospitals. Contact us to share feedback or report issues.