Predictive model and risk analysis for outcomes in diabetic foot ulcer using eXtreme Gradient Boosting algorithm and SHapley Additive exPlanation

Table 1 Descriptive statistics and test analysis of variables, mean (SD)

Variables	Non-amputation (n = 392)	Minor amputation (n = 157)	Major amputation (n = 50)	P value
Demographic data
Gender
Female	129 (32.9)	36 (22.9)	14 (28.0)	0.067
Male	263 (67.1)	121 (77.1)	36 (72.0)
Age: year	68.28 (9.99)	67.06 (9.20)	68.88 (9.25)	0.336
Diabetes duration (years)
Diabetes duration	17.13 (9.43)	18.66 (9.00)	18.89 (8.99)	0.138
Medical history
Hypertension
No	152 (38.8)	71 (45.2)	15 (30.0)	0.128
Yes	240 (61.2)	86 (54.8)	35 (70.0)
Hyperlipidemia
No	279 (71.2)	118 (75.2)	36 (72.0)	0.640
Yes	113 (28.8)	39 (24.8)	14 (28.0)
Coronary heart disease
No	145 (37.0)	41 (26.1)	23 (46.0)	0.012
Yes	247 (63.0)	116 (73.9)	27 (54.0)
Heart attack
No	354 (90.3)	145 (92.4)	44 (88.0)	0.604
Yes	38 (9.7)	12 (7.6)	6 (12.0)
Atrial fibrillation
No	370 (94.4)	149 (94.9)	48 (96.0)	0.881
Yes	22 (5.6)	8 (5.1)	2 (4.0)
Cerebrovascular disease
No	280 (71.4)	107 (68.2)	31 (62.0)	0.343
Yes	112 (28.6)	50 (31.8)	19 (38.0)
Chronic lung disease
No	355 (90.6)	139 (88.5)	44 (88.0)	0.705
Yes	37 (9.4)	18 (11.5)	6 (12.0)
Renal insufficiency
No	38 (9.7)	35 (22.3)	12 (24.5)	< 0.001
Yes	354 (90.3)	122 (77.7)	37 (75.5)
Clinical and laboratory data
WBC (109/L)	8.38 (4.53)	9.88 (5.00)	11.45 (6.34)	< 0.001
CRP (mg/L)	41.51 (51.90)	73.31 (77.84)	99.43 (88.93)	< 0.001
PCT (ng/mL)	0.25 (0.63)	0.44 (1.01)	0.67 (1.19)	< 0.001
HbA1c (%)	8.07 (2.50)	8.43 (2.02)	8.38 (1.86)	0.218
Mb (ng/mL)	51.27 (68.78)	52.15 (77.34)	96.16 (123.04)	< 0.001
Tn (ng/mL)	8.64 (68.71)	4.70 (13.29)	10.44 (34.59)	0.720
CK-MB (μg/mL)	1.89 (1.97)	1.63 (1.17)	2.08 (1.48)	0.180
Cr (μmol/L)	107.44 (138.29)	144.60 (191.22)	167.62 (202.93)	0.006
BUN (mmol/L)	7.77 (4.50)	8.92 (5.40)	9.47 (7.07)	0.010
Albumin (g/L)	37.04 (5.74)	34.75 (5.48)	31.79 (5.21)	< 0.001
TC (mmol/L)	4.07 (1.21)	3.84 (1.15)	3.72 (1.93)	0.055
TG (mmol/L)	1.48 (0.89)	1.32 (0.68)	1.25 (0.73)	0.038
LDL-C (mmol/L)	2.37 (0.91)	2.38 (0.90)	2.12 (0.77)	0.166
Medication history
Antibiotic use before hospitalization
No	194 (49.5)	56 (35.7)	16 (32.0)	0.002
Yes	198 (50.5)	101 (64.3)	34 (68.0)
Wound characteristics
Wagner classification
0-3	298 (76.0)	53 (33.8)	14 (28.0)	< 0.001
4/5	94 (24.0)	104 (66.2)	36 (72.0)
Range
≤ 5 cm	307 (78.3)	122 (77.7)	23 (46.0)	< 0.001
≥ 5 cm	85 (21.7)	35 (22.3)	27 (54.0)
Osteomyelitis
No	196 (50.0)	21 (13.4)	3 (6.0)	< 0.001
Yes	196 (50.0)	136 (86.6)	47 (94.0)
Vascular intervention surgery
No	310 (79.1)	87 (55.4)	33 (66.0)	< 0.001
Yes	82 (20.9)	70 (44.6)	17 (34.0)

P values represent the results of intergroup trend analyses among the three groups, with P < 0.05 deemed significant. All continuous variables have been standardized, and binary variables with 0 representing no and 1 representing yes. WBC: White blood cell; CRP: C-reactive protein; PCT: Procalcitonin; HbA1c: Glycated hemoglobin; Mb: Myoglobin; Tn: Troponin; CK-MB: Creatine kinase isoenzyme; Cr: Creatinine; BUN: Blood urea nitrogen; TC: Total cholesterol; TG: Triglycerides; LDL-C: Low-density lipoprotein cholesterol.

Table 2 Parameter optimization results of eXtreme Gradient Boosting model

Parameter	Default	Value
Booster	Gbtree	Gbtree
Objective	Reg: Linear	Binary: Logistic
Eta	0.3	0.1
Max_depth	6	6
Eval_metric	Logloss	Logloss
Colsample_bytree	1.0	0.5
Subsample	1.0	0.5
Min_child_weight	1.0	1.0

Table 3 Various evaluation indicators of different models in the test set

Evaluation_metrics	Non amputation (%)	Minor amputation (%)	Major amputation (%)
Sensitivity	75.7	78.1	98.7
Specificity	77.8	69.4	88.9
NPV	66.0	61.0	88.9
PPV	84.8	83.8	98.7

NPV: Negative predictive value; PPV: Positive predictive value.