Hi, I wondered why there is no death table included in the synthetic evaluation dataset - or have I missed this somewhere? Death data of the evaluation dataset would be very helpful for comparing the different models internally before submitting to the contest. Or is there some specific rationale for not releasing the evaluation death data? Thank you!

Created by O Mueller omueller
>For us it would be helpful to get the death table for the evaluation set. We know this are only synthetic data but it gives us hints about the correct working procedure of our models locally. Good to know, I'll work on putting this information into the next release of synthetic data. Thank you for the feedback. >So but to get it clear all data in the real data will be transformed to OMOP version 5.0 so there is a death data table in the training set and no hidden information in Concept ID or somewhere else like in the OMOP version 6.0 ? That's correct, we are using OMOP v5.0 so there is a death table in the training data and our gold-standard benchmark is derived from the death table in the evaluation dataset. There is no hidden information anywhere else.
For us it would be helpful to get the death table for the evaluation set. We know this are only synthetic data but it gives us hints about the correct working procedure of our models locally. So but to get it clear all data in the real data will be transformed to OMOP version 5.0 so there is a death data table in the training set and no hidden information in Concept ID or somewhere else like in the OMOP version 6.0 ? Best regards,
>I wondered why there is no death table included in the synthetic evaluation dataset - or have I missed this somewhere? Currently, there is no death table in the synthetic evaluation dataset. >Death data of the evaluation dataset would be very helpful for comparing the different models internally before submitting to the contest. We can include the death table if that would be helpful, but keep in mind, models trained and evaluated on synthetic data will not be accurate or generalizable. >Or is there some specific rationale for not releasing the evaluation death data? The only reason we didn't release it was because that table will not be available to your models in the inference stage. We were just trying to replicate our evaluation environment as closely as possible.

Synpuf evaluation dataset - death table missing? page is loading…