Background
Methods
Data collection
AKI definition
Study design
Predicting AKI
Detecting AKI
Experimental methodology
Results
Patient characteristics
Variable | AKI (2258) | Non-AKI (23263) |
P-value |
---|---|---|---|
Demographics | |||
Age | 75.2 ± 9.5 | 75.3 ± 9.6 | 0.799 |
BMI | 29.7 ± 8.2 | 28.4 ± 7.3 |
<0.0001
|
Race = White | 1974 (87.4 %) | 20940 (90.0 %) |
<0.0001
|
Race = Black | 198 (8.8 %) | 1571 (6.8 %) |
<0.0004
|
Race = Other | 27 (1.2 %) | 240 (1.0 %) | 0.5331 |
Sex = Female | 1101 (48.8 %) | 12426 (53.4 %) |
<0.0001
|
Sex = Male | 1154 (51.1 %) | 10791 (46.4 %) |
<0.0001
|
Tobacco Use = Never | 767 (34.0 %) | 8601 (37.0 %) |
<0.005
|
Tobacco Use = Quit | 1281 (56.7 %) | 12118 (52.1 %) |
<0.0001
|
Tobacco Use = Yes | 193 (8.5 %) | 2360 (10.1 %) |
<0.02
|
Alcohol Use | 705 (31.2 %) | 8037 (34.6 %) |
<0.002
|
Family History | 37 (1.64 %) | 442 (1.90 %) | 0.4280 |
Laboratory Values | |||
BUN | 25.81 ± 14.27 | 17.45 ± 8.63 |
<0.0001
|
AST | 35.49 ± 29.02 | 27.79 ± 18.7 |
<0.0001
|
Troponin | 3.98 ± 3.13 | 1.31 ± 1.12 |
<0.0001
|
Blood Bilirubin | 0.77 ± 0.56 | 0.60 ± 0.37 |
<0.0001
|
Platelet Count | 214.33 ± 92.98 | 215.11 ± 85.71 | 0.686 |
Heart Rate | 84.2 ± 19.0 | 81.3 ± 18.7 |
<0.0001
|
Temperature | 98.2 ± 1.2 | 98.1 ± 1.6 | 0.08 |
BP systolic | 140 ± 32 | 138 ± 28 |
<0.01
|
BP diastolic | 72 ± 16 | 71 ± 14 |
<0.04
|
Medications | |||
ACE Inhibitors | 1219 (54.0 %) | 10869 (46.7 %) |
<0.0001
|
ARB | 422 (18.7 %) | 3851 (16.6 %) |
<0.02
|
NSAIDS | 821 (36.36 %) | 8619 (37.05 %) | 0.5312 |
Lipid Lowering Drugs | 1539 (68.16 %) | 15233 (65.48 %) |
<0.02
|
Diuretics | 1750 (77.50 %) | 13654 (58.69 %) |
<0.0001
|
ACE Inhibitors or NSAIDS or Diuretics | 2052 (90.88 %) | 19111 (82.15 %) |
<0.0001
|
ARB or ACE Inhibitors or NSAIDS or Diuretics | 2076 (91.94 %) | 19580 (84.17 %) |
<0.0001
|
K Sparing | 284 (12.58 %) | 1786 (7.68 %) |
<0.0001
|
Aminoglycoside Antibiotics | 241 (10.67 %) | 1986 (8.54 %) |
<0.001
|
Radiocontrast Dyes | 1536 (68.02 %) | 14229 (61.17 %) |
<0.0001
|
Cisplatin | 31 (1.37 %) | 267 (1.15 %) | 0.3963 |
Acyclovir | 119 (5.27 %) | 1170 (5.03 %) | 0.6539 |
Comorbidities | |||
Prior AKI | 378 (16.74 %) | 1470 (6.32 %) |
<0.0001
|
Diabetes | 223 (9.88 %) | 1410 (6.06 %) |
<0.0001
|
Hyperlipidemia | 392 (17.36 %) | 3127 (13.44 %) |
<0.0001
|
Hypercalcemia | 32 (1.417 %) | 222 (0.95 %) |
<0.05
|
Thrombocytopenia | 103 (4.56 %) | 567 (2.43 %) |
<0.0001
|
Hypertension | 453 (20.06 %) | 3518 (15.12 %) |
<0.0001
|
Heart Failure | 272 (12.05 %) | 1442 (6.20 %) |
<0.0001
|
Coronary Artery Disease | 250 (11.07 %) | 1670 (7.18 %) |
<0.0001
|
Disorders of Lipoid Metabolism | 134 (5.93 %) | 1413 (6.07 %) | 0.8265 |
Pancreatitis | 51 (2.26 %) | 342 (1.47 %) |
<0.005
|
Rhabdomyolysis | 19 (0.84 %) | 164 (0.70 %) | 0.5464 |
Congestive Heart Failure | 272 (12.04 %) | 1442 (6.20 %) |
<0.0001
|
Sepsis | 230 (10.19 %) | 1324 (5.69 %) |
<0.0001
|
Respiratory Failure | 363 (16.08 %) | 1271 (5.46 %) |
<0.0001
|
Classification results
Method | Prediction at 24 h (95 % CI) | Detection (95 % CI) |
---|---|---|
Logistic Regression | 0.660 (0.647–0.673) |
0.743 (0.732–0.755)
|
Naïve Bayes |
a0.654 (0.639–0.669) |
a0.699 (0.684–0.715) |
Decision Trees |
a0.639 (0.627–0.651) |
a0.725 (0.717–0.733) |
Support Vector Machine |
a0.621 (0.609–0.633) |
a0.692 (0.682–0.702) |
Ensemble |
0.664 (0.651–0.676)
|
a0.738 (0.727–0.748) |
Ablation study
All (95 % CI) | Exclude Labs (95 % CI) | Exclude Meds (95 % CI) | Exclude Comorbidities (95 % CI) | |
---|---|---|---|---|
Prediction at 24 h | 0.660 | 0.656 | 0.647 | 0.625a
|
(0.647–0.673) | (0.643–0.668) | (0.633–0.662) | (0.611–0.638) | |
Detection | 0.743 | 0.668a
| 0.728a
| 0.705a
|
(0.732–0.755) | (0.652–0.683) | (0.707–0.749) | (0.686–0.724) |
Learning curves
Odds ratios
Discussion
AKI detection vs. AKI prediction
Within time from admission (hours) | Number of AKI cases |
---|---|
6 | 8 |
12 | 92 |
24 | 476 |
36 | 725 |
48 | 1166 |
60 | 1313 |
120 | 1796 |
240 | 2103 |
Through end of stay (total) | 2258 |