Machine learning-based prediction of postoperative mortality risk after abdominal surgery

Table 1 Univariate analysis of postoperative death in patients undergoing abdominal surgery, mean ± SD/n (%)

Index	Death (n = 52)	Survival (n = 178)	t/χ2	P value
Sex			0.025	0.875
Male	27 (51.92)	95 (53.37)
Female	25 (48.08)	83 (46.63)
Age, years	59.21 ± 5.26	57.34 ± 6.14	1.990	0.048
BMI, kg/m2	23.50 ± 2.73	23.89 ± 2.95	-0.839	0.403
Emergency surgery			1.525	0.217
Yes	19 (36.54)	48 (26.97)
No	33 (63.46)	130 (73.03)
Previous abdominal surgery			0.073	0.787
Yes	14 (26.92)	45 (25.28)
No	38 (73.08)	133 (74.72)
Smoking			0.303	0.582
Yes	10 (19.23)	40 (22.47)
No	42 (80.77)	138 (77.53)
Drinking			0.014	0.904
Yes	24 (46.15)	81 (45.51)
No	28 (53.85)	97 (54.49)
Hypertension			0.115	0.735
Yes	5 (9.62)	19 (10.67)
No	47 (90.38)	159 (89.33)
Diabetes			0.031	0.861
Yes	18 (34.62)	59 (33.15)
No	34 (65.38)	119 (66.85)
Surgical approach			7.209	0.007
Laparoscopic	23 (44.23)	111 (62.36)
Laparotomy/conversion	29 (55.77)	67 (37.64)
ASA classification			10.011	0.040
I	3 (5.77)	9 (5.06)
II	27 (51.92)	130 (73.03)
III	12 (23.08)	26 (14.61)
IV	9 (17.31)	8 (4.49)
V	1 (1.92)	5 (2.81)
Abnormal ECG			1.792	0.181
Yes	27 (51.92)	73 (41.01)
No	25 (48.08)	105 (58.99)
Preoperative complications			8.136	0.004
Yes	16 (30.77)	26 (14.61)
No	36 (69.23)	152 (85.39)
SBP, mmHg	127.94 ± 7.53	128.59 ± 6.04	-0.642	0.522
DBP, mmHg	74.44 ± 5.27	75.20 ± 6.15	-0.809	0.420
D-dimer, mg/L	1.18 ± 0.35	1.13 ± 0.20	1.120	0.264
PT, second	12.29 ± 3.15	11.90 ± 2.38	0.948	0.344
APTT, second	32.33 ± 2.83	27.04 ± 3.46	10.077	< 0.001
FBG, g/L	3.31 ± 0.74	3.38 ± 0.89	-0.503	0.616
AST, IU/L	43.50 ± 14.51	37.46 ± 9.72	3.489	0.001
BUN, mmol/L	5.63 ± 0.63	5.53 ± 0.69	0.993	0.322
TBA, μmol/L	10.55 ± 2.32	10.12 ± 1.96	1.323	0.187
ALB, g/L	34.99 ± 5.94	35.81 ± 5.45	-0.934	0.351
A/G ratio	1.24 ± 0.29	1.32 ± 0.27	-1.800	0.073
Cre, μmol/L	69.61 ± 9.07	68.25 ± 8.09	1.036	0.301
TBIL, μmol/L	26.07 ± 3.82	20.38 ± 4.42	8.400	< 0.001
Blood calcium, mmol/L	2.17 ± 0.14	2.16 ± 0.17	0.447	0.655
Blood sodium, mmol/L	140.13 ± 3.35	140.98 ± 3.29	-1.654	0.099
Blood potassium, mmol/L	3.82 ± 0.38	3.81 ± 0.40	0.103	0.918
NEUT, × 109/L	5.13 ± 0.79	4.98 ± 0.56	1.486	0.139
LYMPH, × 109/L	1.21 ± 0.25	1.28 ± 0.33	-1.393	0.165
WBC, × 109/L	11.84 ± 3.94	9.72 ± 3.44	3.774	< 0.001
PLT, × 109/L	191.35 ± 63.09	194.91 ± 50.82	-0.420	0.675
Hb, g/L	116.75 ± 5.89	116.40 ± 5.78	0.384	0.701
Operation type
Emergency	2 (3.85)	1 (0.56)	3.104	0.541
Urgent	9 (17.31)	36 (20.22)
Semi-urgent	12 (23.08)	45 (25.28)
Non-urgent	11 (21.15)	28 (15.73)
Elective	18 (34.62)	68 (38.20)
Operation time, hour	2.52 ± 1.09	2.54 ± 1.03	-0.103	0.918

BMI: Body mass index; ECG: Electrocardiogram; ASA: American Society of Anesthesiologists; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; PT: Prothrombin time; APTT: Activated partial thromboplastin time; FBG: Fibrinogen; AST: Aspartate aminotransferase; BUN: Blood urea nitrogen; TBA: Total bile acid; ALB: Albumin; A/G: Albumin/globulin ratio; Cre: Creatinine; TBIL: Total bilirubin; NEUT: Neutrophils; LYMPH: Lymphocytes; WBC: White blood cell; PLT: Platelets; Hb: Hemoglobin.

Table 2 Multivariate logistic regression analysis of postoperative death in patients undergoing abdominal surgery

Index	B	SE	Wald	P value	OR	95%CI
Age	-0.609	0.146	17.291	< 0.001	0.544	0.408-0.725
Surgical approach	0.305	0.495	0.380	0.538	1.357	0.514-3.582
Preoperative complications	-0.172	0.605	0.081	0.776	0.842	0.257-2.753
APTT	0.533	0.108	24.360	< 0.001	1.705	1.379-2.107
AST	0.172	0.062	7.681	0.006	1.188	1.052-1.341
TBIL	0.174	0.078	4.962	0.026	1.190	1.021-1.387
ASA classification	0.119	0.300	0.158	0.691	1.127	0.626-2.030
WBC	0.183	0.074	6.154	0.013	1.201	1.039-1.388
Constant	4.880	5.475	0.794	0.373	131.566

OR: Odds ratio; CI: Confidence interval; APTT: Activated partial thromboplastin time; AST: Aspartate aminotransferase; TBIL: Total bilirubin; ASA: American Society of Anesthesiologists; WBC: White blood cell.

Table 3 Comparison of predictors between the training and validation cohorts, mean ± SD

Index	Training cohort (n = 161)	Validation cohort (n = 69)	t	P value
Age	57.67 ± 5.80	57.99 ± 6.46	-0.364	0.716
APTT	28.16 ± 4.00	28.41 ± 3.99	-0.424	0.672
AST	38.41 ± 10.75	39.80 ± 12.33	-0.860	0.391
TBIL	21.44 ± 4.75	22.20 ± 5.24	-1.081	0.281
WBC	10.33 ± 3.59	9.88 ± 3.84	0.856	0.393

APTT: Activated partial thromboplastin time; AST: Aspartate aminotransferase; TBIL: Total bilirubin; WBC: White blood cell.

Table 4 Model prediction efficiency

Model	AUC (95%CI)	Accuracy	Sensitivity	Specificity	Recall	Precision
Nomogram	0.908 (0.824-0.992)	0.899	0.962	0.688	0.962	0.911
Decision-tree	0.874 (0.785-0.963)	0.855	0.849	0.875	0.849	0.957
Random-forest	0.928 (0.869-0.987)	0.841	0.906	0.625	0.906	0.889
Gradient-boosting tree	0.907 (0.837-0.976)	0.870	0.906	0.750	0.906	0.923
Support vector machine	0.983a,2,4 (0.959-1.000)	0.884	0.925	0.750	0.925	0.925
Naive Bayes	0.807a,1,3,4,5 (0.702-0.911)	0.797	0.811	0.750	0.811	0.915

^aP < 0.05.

¹P vs nomogram.

²P vs decision-tree model.

³P vs random-forest model.

⁴P vs gradient-boosting tree model.

⁵P vs support vector machine model.

AUC: Area under the receiver operating characteristic curve; CI: Confidence interval.