Hello. Thank you for the excellent work.
When conducting case studies, I had a concern on the preprocessing code. The PyHealth task preprocessing code on mortality and readmission was based on "HADM_ID" not the "ROW_ID". This made the sequence mixed, for example a patient was dead in her/his second visit, but again was alive in the third visit. Also for the readmission task, the time difference should have been the difference between the next visit's encounter time and current visit's discharge time. The original code calculated the time difference for the two encounter times.
This leads to a difference of rates for positive labels, and might lead to a change in results.
Can I have some explanation regarding these issues.
Thank you.
Hello. Thank you for the excellent work.
When conducting case studies, I had a concern on the preprocessing code. The PyHealth task preprocessing code on mortality and readmission was based on "HADM_ID" not the "ROW_ID". This made the sequence mixed, for example a patient was dead in her/his second visit, but again was alive in the third visit. Also for the readmission task, the time difference should have been the difference between the next visit's encounter time and current visit's discharge time. The original code calculated the time difference for the two encounter times.
This leads to a difference of rates for positive labels, and might lead to a change in results.
Can I have some explanation regarding these issues.
Thank you.