Integrating traditional Chinese medicine syndromes with clinical variables to predict lower extremity arterial disease in type 2 diabetes

Table 1 Baseline characteristics of included patients

Indicator	LEAD group (n = 647)	Non-LEAD group (n = 371)	P value
Age (year)	63 (55, 72)	49 (40, 57)	< 0.001
SBP (mmHg)	132 (122, 145)	130 (120, 142)	0.007
DBP (mmHg)	80 (74, 88)	82 (76, 90)	0.024
Sex			0.410
Male	399 (61.67)	219 (59.03)
Female	248 (38.33)	152 (40.97)
Hypertension	386 (59.66)	141 (38.01)	< 0.001
Coronary heart disease	20 (3.09)	5 (1.35)	0.129
Fatty liver	182 (28.13)	140 (37.74)	0.002
Diabetic nephropathy	176 (27.20)	60 (16.17)	< 0.001
Abnormal complexion	410 (63.37)	203 (54.72)	0.008
Obesity	92 (14.22)	73 (19.68)	0.027
Emaciation	86 (13.29)	23 (6.20)	< 0.001
Dark tongue	513 (79.29)	265 (71.43)	0.007
Red tongue	330 (51.00)	207 (55.80)	0.158
Pale tongue	118 (18.24)	69 (18.60)	0.962
Teeth-marked tongue	8 (1.24)	5 (1.35)	0.897
Tongue with petechiae	6 (0.93)	3 (0.81)	0.991
Cracked tongue	7 (1.08)	5 (1.35)	0.843
Limb numbness	325 (50.23)	144 (38.81)	< 0.001
Limb pain	149 (23.03)	62 (16.71)	0.022
Dizziness and palpitation	293 (45.29)	131 (35.31)	0.002
Fatigue	335 (51.78)	166 (44.74)	0.035
Chest tightness	157 (24.27)	86 (23.18)	0.716
Blurred vision	357 (55.18)	174 (46.90)	0.013
Polyphagia	90 (13.91)	69 (18.60)	0.062
Frequent urination	229 (35.39)	146 (39.35)	0.234

Data are expressed as median (Q1, Q3) for continuous variables and n (%) for categorical variables. Continuous and ordinal variables were compared using the Mann-Whitney U test (with ties correction). P values were pooled from five multiple imputations using Rubin’s rules. LEAD: Lower extremity arterial disease; SBP: Systolic blood pressure; DBP: Diastolic blood pressure.

Table 2 Laboratory indicators of included patients

Indicator	LEAD group (n = 647)	Non-LEAD group (n = 371)	P value
Hemoglobin (g/L)	135 (123, 147)	141 (129, 152)	< 0.001
RBC count (× 1012/L)	4.47 (4.02, 4.81)	4.69 (4.33, 5.04)	< 0.001
Platelet count (× 109/L)	187 (151, 227)	197 (156, 236)	0.083
Urine glucose			0.281
-	388 (59.97)	214 (57.68)
±	27 (4.17)	9 (2.43)
1+	31 (4.79)	15 (4.04)
2+	183 (28.28)	121 (32.61)
3+	18 (2.78)	12 (3.23)
Urine WBC (× 109/L)	5 (2, 16)	5 (2, 15)	0.395
Urine RBC (× 109/L)	5 (2, 11)	5 (3, 12)	0.492
Urine protein			0.004
-	594 (91.81)	358 (96.50)
1+	19 (2.94)	5 (1.35)
2+	20 (3.09)	4 (1.08)
3+	14 (2.16)	3 (0.81)
4+	0 (0.00)	1 (0.27)
FBG (mmol/L)	7 (6, 10)	8 (6, 11)	0.024
HbA1c (%)	7.90 (6.80, 9.60)	8.10 (6.80, 9.80)	0.611
TG (mmol/L)	1.38 (0.97, 2.10)	1.59 (1.02, 2.47)	0.007
TC (mmol/L)	4.42 (3.81, 5.24)	4.69 (4.05, 5.44)	< 0.001
LDL-C (mmol/L)	2.456 (1.645, 3.055)	2.620 (1.935, 3.275)	0.003
HDL-C (mmol/L)	1.10 (0.90, 1.42)	1.14 (0.90, 1.44)	0.600
Serum creatinine (μmol/L)	62.7 (53.1, 75.5)	56.6 (48.6, 66.8)	< 0.001
BUN (mmol/L)	5.95 (4.98, 7.24)	5.55 (4.65, 6.47)	< 0.001
ALT (U/L)	17 (13, 24)	20 (14, 32)	< 0.001
AST (U/L)	17.8 (14.0, 22.0)	18.0 (14.0, 24.0)	0.037

Data are expressed as median (Q1, Q3) for continuous variables and n (%) for categorical variables. Continuous and ordinal variables were compared using the Mann-Whitney U test (with ties correction). P values were pooled from five multiple imputations using Rubin’s rules. LEAD: Lower extremity arterial disease; RBC: Red blood cell; WBC: White blood cell; FBG: Fasting blood glucose; HbA1c: Glycosylated hemoglobin; TG: Triglycerides; TC: Total cholesterol; LDL-C: Low-density lipoprotein cholesterol; HDL-C: High-density lipoprotein cholesterol; BUN: Blood urea nitrogen; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase.

Table 3 Univariate and multivariate logistic regression models of predictors for lower extremity arterial disease

Variables	OR (95%CI)	P value	OR (95%CI) adjusted	P value adjusted	P value FDR
Age	5.799 (4.305-7.813)	< 0.001a	4.552 (3.339-6.207)	< 0.001a	< 0.001a
Hypertension	2.412 (1.856-3.135)	< 0.001a	1.788 (1.337-2.391)	< 0.001a	< 0.001a
Hemoglobin	3.649 (2.038-6.534)	< 0.001a	1.512 (0.718-3.183)	0.276	< 0.001a
RBC	4.281 (2.302-7.963)	< 0.001a	1.852 (0.853-4.020)	0.119	< 0.001a
BUN	2.160 (1.542-3.024)	< 0.001a	1.678 (1.157-2.435)	0.006a	< 0.001a
Emaciation	2.351 (1.457-3.794)	< 0.001a	1.900 (1.117-3.233)	0.018a	< 0.001a
Dark tongue	1.497 (1.115-2.011)	0.007a	1.194 (0.859-1.658)	0.292	0.008a
Fatigue	1.339 (1.036-1.730)	0.026a	1.345 (1.012-1.787)	0.041a	0.026a

^aP < 0.05.

Adjusted covariates in the multivariate model included age ≥ 60 years, hypertension, hemoglobin, red blood cell, blood urea nitrogen, emaciation, dark tongue, and fatigue. 95%CI: 95% confidence interval; OR: Odds ratio; FDR: False discovery rate; RBC: Red blood cell; BUN: Blood urea nitrogen.

Table 4 Performance comparison of prediction models based on five machine learning classifiers

Model	AUC	Accuracy	Sensitivity	Specificity	F1 score
XGBoost	0.763	0.712	0.651	0.820	0.743
Random forest	0.762	0.699	0.631	0.820	0.728
Logistic regression	0.758	0.686	0.605	0.829	0.711
LightGBM	0.754	0.703	0.641	0.811	0.733
GBDT	0.744	0.699	0.651	0.784	0.734

AUC: Area under the receiver operating characteristic curve; GBDT: Gradient boosting decision tree.

Table 5 Key numerical summary of model performance, calibration, and clinical utility

Metric	XGBoost
Nested 5 × 5 cross-validation AUC	0.746 ± 0.038
Bootstrap optimism-corrected AUC	0.745
Calibration slope	1.176-1.007
Calibration intercept	-0.104 to -0.040
Expected calibration error	0.044 to 0.034
Brier score (validation)	0.186
Split-ratio AUC (6:4/7:3/8:2)	0.758/0.762/0.755
PPV/NPV at threshold 0.5	0.737/0.607
PPV/NPV at threshold 0.7	0.863/0.551

Results pertain to the XGBoost model (area under the receiver operating characteristic curve = 0.763). AUC: Area under the receiver operating characteristic curve; PPV: Positive predictive value; NPV: Negative predictive value.