Machine learning-based prediction of in-hospital mortality for critically ill patients with sepsis-associated acute kidney injury

Figures & data

Table 1. Characteristics of the patients with sepsis associated AKI (SA-AKI).

Parameters	Total(n = 12,196)	Survival group(n = 9944)	Non-survival group(n = 2352)	p Value
Age (y)	67.0 ± 16.1	66.4 ± 16.2	69.5 ± 15.6	<0.001
Gender (male, %)	6995 (57.4)	5683 (57.7)	1312 (55.8)	0.090
BMI(kg/m2)	29.6 ± 7.8	29.8 ± 7.8	28.7 ± 7.8	<0.001
Baseline creatinine* (mg/dL)	1.0 ± 0.8	1.0 ± 0.7	1.1 ± 0.8	<0.001
Systolic blood pressure (mmHg)	115.4 ± 15.1	115.8 ± 14.9	113.6 ± 15.8	<0.001
Diastolic blood pressure (mmHg)	61.3 ± 10.4	61.3 ± 10.3	61.0 ± 10.8	0.132
Mean arterial Pressure (mmHg)	76.7 ± 10.1	76.9 ± 10.0	75.9 ± 10.6	<0.001
Respiratory rate (times per minute)	19.9 ± 4.1	19.6 ± 4.0	21.2 ± 4.4	<0.001
Temperature (°C)	36.9 ± 0.7	36.9 ± 0.6	36.7 ± 0.8	<0.001
SpO2 (%)	97.1 (2.2)	97.1 (2.0)	96.7 (2.7)	<0.001
AKI stage, n (%)				<0.001
Stage 1	2200 (18.0)	1991 (20.2)	209 (8.9)
Stage 2	5877 (48.2)	5150 (52.3)	727 (30.9)
Stage 3	4119 (33.8)	2703 (27.5)	1416 (60.2)
Urine output(mL/d)	1674.5 ± 1205.7	1751.8 ± 1201.3	1351.0 ± 1170.0	<0.001
Creatinine(mg/dL)	2.0 ± 1.7	1.9 ± 1.7	2.6 ± 1.9	<0.001
ARDS, n (%)	5687 (46.6)	4471 (45.4)	1216 (51.7)	<0.001
Myocardial infarction, n (%)	2308 (18.9)	1822 (18.5)	486 (20.7)	0.018
Congestive Heart failure, n (%)	4010 (32.9)	3155 (32.0)	855 (36.4)	<0.001
Cerebrovascular disease, n (%)	2101 (17.2)	1591 (16.2)	510 (21.7)	<0.001
Chronic pulmonary disease, n (%)	3398 (27.9)	2723 (27.7)	675 (28.7)	0.326
Diabetes, n (%)	3669 (30.1)	2995 (30.4)	674 (28.7)	0.098
Renal disease, n (%)	2394 (19.6)	1858 (18.9)	536 (22.8)	<0.001
Malignant cancer, n (%)	1611 (13.2)	1146 (11.6)	465 (19.8)	<0.001
Hypertension, n (%)	5450 (44.7)	4522 (45.9)	928 (39.5)	<0.001
Charlson comorbidity index	5.9 ± 2.9)	5.7 ± 2.8)	6.9 ± 3.0)	<0.001
Cardiac Surgery, n (%)	2147 (17.6)	2053 (20.9)	94 (4.0)	<0.001
Mechanical Ventilation, n (%)	11764 (96.5)	9463 (96.1)	2301 (97.8)	<0.001
Dialysis, n (%)	1147 (9.4)	653 (6.6)	494 (21.0)	<0.001
Vasopressin, n (%)	1873 (15.4)	1046 (10.6)	827 (35.2)	<0.001
Dopamine, n (%)	651 (5.3)	426 (4.3)	225 (9.6)	<0.001
Epinephrine, n (%)	1083 (8.9)	828 (8.4)	255 (10.8)	<0.001
Norepinephrine, n (%)	4831 (39.6)	3321 (33.7)	1510 (64.2)	<0.001
Antifungal medications, n (%)	3605 (29.6)	2639 (26.8)	966 (41.1)	<0.001
Diuretics, n (%)	9385 (77.0)	7720 (78.4)	1665 (70.8)	<0.001
ACEI/ARB drugs, n (%)	3624 (29.7)	3345 (34.0)	279 (11.9)	<0.001
SAPS II score	42.1 ± 14.1	40.3 ± 13.4	49.5 ± 14.8	<0.001
SOFA score	3.0 [2.0,5.0]	3.0 [2.0,4.0]	4.0 [2.0,5.0]	<0.001
GCS score	12.0 [7.0,14.0]	13.0 [9.0,14.0]	8.0 [3.0,13.0]	<0.001
Hemoglobin(g/dL)	9.9 ± 2.2	10.0 ± 2.2	9.7 ± 2.3	<0.001
Platelet(K/uL)	178.7 ± 104.7	179.9 ± 101.9	173.5 ± 115.7	0.015
White blood cell(K/uL)	16.2 ± 11.8	16.0 ± 11.1	17.3 ± 14.2	<0.001
Absolute lymphocyte count(K/uL)	1.4 [0.8,2.2]	1.4 [0.8,2.2]	1.2 [0.7,1.9]	<0.001
Anion gap(mEq/L)	16.9 ± 5.1	16.5 ± 4.9	18.8 ± 5.6	<0.001
Bicarbonate(mEq/L)	20.9 ± 5.0	21.2 ± 4.8	19.6 ± 5.6	<0.001
Glucose(mg/dL)	179.5 ± 102.9	176.1 ± 102.2	193.9 ± 104.6	<0.001
Lactate(mmol/L)	3.1 ± 2.7	2.9 ± 2.4	3.8 ± 3.4	<0.001
PO2(mm Hg)	82.8 ± 54.3	85.0 ± 53.7	73.6 ± 55.7	<0.001
PCO2(mm Hg)	49.3 ± 14.7	49.2 ± 14.2	49.7 ± 16.7	0.207
INR	1.4 [1.2,1.7]	1.3 [1.2,1.6]	1.5 [1.2,2.1]	<0.001
PT(sec)	18.4 ± 12.4	17.7 ± 11.5	21.3 ± 15.4	<0.001
APTT(sec)	33.9 [28.7,47.5]	33.3 [28.5,45.1]	37.8 [29.9,57.9]	<0.001
sodium(mmol/L)	144.8 ± 5.8	144.6 ± 5.4	145.7 ± 7.2	<0.001
potassium(mmol/L)	5.1 ± 0.9	5.1 ± 0.9	5.3 ± 1.0	<0.001
Length of hospital stay (days)	15.4 ± 14.3	16.0 ± 13.7	13.1 ± 16.4	<0.001
Length of ICU stay (days)	5.1 [3.1,9.5]	4.9 [3.1,9.1]	6.1 [3.7,10.9]	<0.001

*Baseline creatinine was defined as the lowest creatinine 48 h before admission.

BMI: body mass index; AKI: acute kidney injury; ARDS: Acute Respiratory Distress Syndrome; SAPS II: Simplified Acute Physiology Score; SOFA: Sequential Organ Failure Assessment; GCS: Glasgow Coma Scale; ACEI/ARB, angiotensin- converting enzyme inhibitor/angiotensin receptor antagonist; PaO₂: partial pressure of oxygen; PaCO₂: partial pressure of carbon dioxide; PT: prothrombin time; INR: international normalized ratio; APTT: activated partial thromboplastin time

Figure 1. The comparison of ROC curves of the mortality prediction models of seven machine learning algorithms for patients with SA-AKI.

Figure 2. Calibration curve of the RF model for prediction of short-term mortality in patients with SA-AKI.

Table 2. The comparison of performance among the prediction models resulted from seven machine learning algorithms that predict the risk of mortality in patients with SA-AKI.

Model	AUC (95%CI)	Accuracy	Precision	F1 scores*	Delong test
Logistic regression	0.756 (0.732–0.779)	0.816	0.569	0.258	<0.001
SVM Linear/rbf	0.718/0.723 (0.691–0.743/0.694– 0.752)	0.809/0.828	1.00/0.671	0.008/0.309	<0.001/<0.001
Naive Bayesian	0.764 (0.741–0.786)	0.795	0.446	0.354	<0.001
XGBoost	0.796 (0.774–0.817)	0.823	0.569	0.414	0.815
Random forest	0.798 (0.774–0.821)	0.832	0.661	0.372	Ref.
K-nearest neighbors	0.689 (0.663–0.714)	0.805	0.487	0.359	<0.001
Decision tree	0.635 (0.611–0.659)	0.765	0.395	0.409	<0.001

*F1 scores= 2* ((precision*recall)/(precision + recall)).

Figure 3. SHAP summary plot of the 11 clinical features of the RF model for prediction of short-term mortality in patients with SA-AKI. GCS: Glasgow Coma Scale score; aki_stage: AKI stage; SAPS-II: Simplified Acute Physiology Score; lymphocyte, absolute lymphocyte count; BMI: body mass index.

Availability of data and materials

The datasets analyzed for this study can be found in the MIMIC-IV (https://mimic.mit.edu/) [11].