Table 2.
Manual validation of the CKD diagnosis and staging algorithm.
Group | Columbia University | Vanderbilt University | Mayo Clinic | Combined | ||||
---|---|---|---|---|---|---|---|---|
N Reviewed | PPV | N Reviewed | PPV | N Reviewed | PPV | N Reviewed | PPV | |
Controls | 62 | 0.968 | 20 | 0.950 | 20 | 1.000 | 102 | 0.971 |
Cases | 189 | 0.995 | 80 | 0.825 | 80 | 0.950 | 349 | 0.946 |
CKD Stage 1 | 20 | 0.900 | 10 | 0.600 | 10 | 1.000 | 40 | 0.850 |
CKD Stage 2 | 20 | 1.000 | 10 | 1.000 | 10 | 1.000 | 40 | 1.000 |
CKD Stage 3a | 20 | 1.000 | 10 | 1.000 | 10 | 1.000 | 40 | 1.000 |
CKD Stage 3b | 22 | 1.000 | 10 | 0.800 | 10 | 1.000 | 42 | 0.952 |
CKD Stage 4 | 23 | 0.913 | 10 | 1.000 | 10 | 1.000 | 43 | 0.953 |
CKD Stage 5 | 20 | 0.750 | 10 | 0.200 | 10 | 1.000 | 40 | 0.675 |
ESRD after transplant | 24 | 0.792 | 10 | 1.000 | 10 | 0.800 | 44 | 0.818 |
ESRD on dialysis | 40 | 0.700 | 10 | 1.000 | 10 | 0.900 | 60 | 0.783 |
The validations were performed by selecting 451 algorithm-defined cases and controls across all stages for blinded chart reviews by domain experts across the three independent validation sites: Columbia University, Vanderbilt University, and Mayo Clinic. We derived positive predictive values (PPVs) for controls and CKD cases combined and by disease stage. The overall diagnostic PPV was 95% (range 83–99%) for CKD cases and 97% (range 95–100%) for healthy controls.