This PR improves the implementation of the KDIGO guidelines for MIMIC-IV.
The big changes:
Refactored the UO query with better syntax for calculating a sum over the last X hours (no change in content)
Added the aki_stage_smoothed column which carries forward the highest AKI stage for 6 hours (arbitrary but seems reasonable). Helps smooth over the aki_stage alternating between the creatinine/UO derived stage, e.g.
time
uo
cr
aki_stage
aki_stage_smoothed
18:00:00
0.3833
3
3
19:30:00
0.8
0
3
Rewrote the UO query to more closely align with the ranges given by KDIGO - stage 1 uses UO from 6-12 hours, stage 2 uses UO from >= 12 hours, and stage 3 uses UO from >12 hours and >= 24 hours. Difference in UO stages across stays is below:
aki_stage_uo
n_row_updated
n_row_before
n_stay_updated
n_stay_before
NULL
1043972
628376
73181
73151
0
2222462
2586664
60458
68250
1
185797
135580
37718
27675
2
385465
358460
29103
27513
3
173559
195757
8317
9355
Seems to result in a lot more stage 1 AKI - but I think the logic is better. Would welcome any thoughts.
This PR improves the implementation of the KDIGO guidelines for MIMIC-IV.
The big changes:
aki_stage_smoothed
column which carries forward the highest AKI stage for 6 hours (arbitrary but seems reasonable). Helps smooth over theaki_stage
alternating between the creatinine/UO derived stage, e.g.NULL
Seems to result in a lot more stage 1 AKI - but I think the logic is better. Would welcome any thoughts.