Serum alpha-1-microglobulin as a predictor of multiple complications in type 2 diabetes mellitus patients

Abstract

BACKGROUND

Poor glycaemic control in patients with type 2 diabetes mellitus (T2DM) is often accompanied by multiple complications, including diabetic nephropathy (DN), diabetic retinopathy (DR), diabetic peripheral neuropathy (DPN), and cardiac structural abnormality left ventricular hypertrophy (LVH). Early identification of high-risk populations for these complications and the implementation of intervention measures are crucial for improving patient outcomes. Serum alpha-1-microglobulin (α1-MG), a multifunctional protein synthesized by the liver and lymphocytes, has been considered a potential biomarker of diabetes-related diseases in recent years.

AIM

To investigate the associations of serum α1-MG with DN, DR, DPN, and LVH in T2DM patients and its predictive value.

METHODS

This retrospective study included 5045 T2DM patients. The study participants were stratified into quartiles according to their serum α1-MG levels. Multivariate logistic regression, restricted cubic spline, and explainable machine learning models were employed for risk assessment and feature importance evaluation.

RESULTS

Increased α1-MG levels were observed in patients with DN, DR, DPN, and LVH (all P < 0.001). Multivariate logistic regression revealed that each standard deviation increase in α1-MG was associated with an 84% increase in DN risk (OR: 1.84, 95%CI: 1.62-2.10, P < 0.001), a 17% increase in DR risk (OR: 1.17, 95%CI: 1.07-1.28, P < 0.001), a 14% increase in DPN risk (OR: 1.14, 95%CI: 1.03-1.27, P = 0.014), and a 28% increase in LVH risk (OR: 1.28, 95%CI: 1.18-1.38, P < 0.001). Subgroup analyses and machine learning confirmed the associations of elevated α1-MG with these complications in T2DM patients.

CONCLUSION

Elevated serum α1-MG levels were independently associated with increased risks of DN, DR, DPN, and LVH in T2DM patients, suggesting its potential as a predictive biomarker.

Key Words: Alpha-1-microglobulin; Microvascular complications; Cardiac complications; Type 2 diabetes mellitus; Machine learning models

Core Tip: Research on the relationship between serum alpha-1-microglobulin (α1-MG) and multiple complications in patients with type 2 diabetes mellitus (T2DM) is limited. Our study revealed a significant correlation between elevated α1-MG levels and the risk of diabetic nephropathy, diabetic retinopathy, diabetic peripheral neuropathy, and left ventricular hypertrophy. By combining traditional statistics with machine learning, we established the diagnostic value of α1-MG for microvascular and cardiac complications, demonstrating its superior performance in early risk prediction. These findings indicate the potential of α1-MG as a biomarker for predicting multiple complications in patients with T2DM.

INTRODUCTION

Diabetes mellitus is a prevalent chronic metabolic disorder that has emerged as a global public health challenge, with the number of affected individuals projected to rise to 783.2 million by 2045, with type 2 diabetes mellitus (T2DM) patients accounting for 90%[1]. In patients with diabetes, prolonged instability of blood glucose levels has been shown to lead to tissue damage across multiple organ systems, resulting in the development of various complications, predominantly microvascular and cardiac in nature[2]. Microvascular complications, including diabetic retinopathy (DR), diabetic nephropathy (DN), and diabetic peripheral neuropathy (DPN), are hallmark manifestations of T2DM that arise from a common pathological basis characterized by widespread microvascular sclerosis and dysfunction, which ultimately results in impaired blood supply, neurotrophic vascular abnormalities, and consequent multiorgan damage[3-6]. Prolonged hyperglycaemia, chronic inflammation, and oxidative stress may also lead to myocardial fibrosis and cardiomyocyte hypertrophy, promoting the development of cardiac complications, primarily left ventricular hypertrophy (LVH)[7]. LVH, which has a reported prevalence of 32%-71% in patients with T2DM, is a characteristic cardiac structural abnormality of diabetic cardiomyopathy[8]. These complications of T2DM can lead to blindness, renal failure, nerve damage, and cardiovascular events, which severely impact patients' quality of life[9]. Early detection and intervention are critical for reducing the risk of both the onset and progression of these complications[10].

Alpha-1-microglobulin (α1-MG), a glycoprotein with a molecular weight ranging from approximately 26000 to 33000 Daltons, is predominantly synthesized in the liver, kidneys, and immune cells and plays essential biological roles, including antioxidant activity, immune regulation, and renal protection[11-14]. Existing studies have identified α1-MG as a potential biomarker for the early diagnosis and treatment of several diseases. Elevated serum levels of α1-MG have been reported in patients with preeclampsia, suggesting its potential use as a biochemical marker for predicting the onset of this condition[15]. Serum α1-MG may also serve as a valuable biomarker for assessing renal function, evaluating the efficacy of haemodialysis, and stratifying the prognosis of chronic kidney disease patients receiving renal replacement therapy[16,17]. Furthermore, studies have suggested the role of α1-MG as a biomarker for the early detection of DN[18-20]. Since urinary α1-MG levels are directly related to albuminuria in T2DM patients and the expression levels of α1-MG in peripheral blood are significantly greater in DN patients, these findings indicate its potential as a predictive biomarker for DN.

However, research on the associations between serum α1-MG and other diabetic complications, including DR and DPN, as well as cardiac complications, has been limited. In addition, traditional statistical methods might neglect complex nonlinear relationships in biomarker data, so advanced methods such as machine learning are needed to improve prediction accuracy and feature prioritization[21]. Current research shows that regularized regression methods such as least absolute shrinkage and selection operator (LASSO) perform well in biomedical data analysis[22]. LASSO can promote variable selection and model regularization, making it very suitable for identifying clinically meaningful predictors in the context of complex and multifactorial outcomes such as diabetic complications. In addition, integrated tree-based methods such as extreme gradient boosting (XGBoost) and random forest (RF) are well suited for capturing nonlinearities and interactions between clinical variables and often outperform traditional models in disease risk prediction[23]. Furthermore, support vector machine (SVM) have demonstrated robust performance in classifying complex phenotypes of various diabetic complications[24]. Combining these cutting-edge algorithms not only enhances predictive performance but also aligns with current trends in precision medicine and data-driven clinical research[25,26].

The aim of this study was to explore the associations between serum α1-MG levels and DN, DR, DPN, and LVH in patients with T2DM and to employ traditional statistical methods combined with machine learning to evaluate its potential as a predictive biomarker for early intervention and improved management.

MATERIALS AND METHODS

Patients

This retrospective cross-sectional study, conducted at Southwest Hospital’s Department of Endocrinology (2018-2023), was approved by the Southwest Hospital Human Research Ethics Committee (KY2024007). The diagnosis of T2DM adhered to the criteria established by the American Diabetes Association (2020). Patients were excluded if they met the following criteria: (1) Were less than 18 years of age; (2) Were pregnant; (3) Presented with acute complications of diabetes mellitus; (4) Lacked complete α1-MG data; or (5) Were diagnosed with severe valvular heart disease, atrial fibrillation, renal diseases (excluding DN), end-stage renal disease, stroke, infectious diseases, haematological disorders, or other malignant conditions. Ultimately, the study included a total of 5045 T2DM patients who met the inclusion criteria.

Clinical data

Baseline clinical data, including age, sex, height, weight, body mass index (BMI), systolic blood pressure (SBP), diastolic blood pressure (DBP), alcohol consumption history, smoking history, hypertension history, hyperlipidaemia history, coronary heart disease (CHD) history, antihypertensive drugs, lipid-lowering drugs, and antidiabetic drugs, were extracted from the electronic medical records of patients. A smoking history of “yes” was defined as current or former regular smoking, whereas “no” was defined as a lifetime history of never engaging in regular smoking[27]. Similarly, an alcohol consumption history of “yes” was defined as current or former regular alcohol use, and “no” was defined as lifelong abstinence or only occasional drinking without a pattern of regular use[28]. BMI was computed as weight (kg) divided by height squared (m²) and expressed as kg/m². The laboratory parameters included fasting blood glucose (FBG), glycated haemoglobin (HbA1c), total cholesterol (TC), triglyceride (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), alanine aminotransferase (ALT), aspartate aminotransferase (AST), uric acid (UA), estimated glomerular filtration rate (eGFR), α1-MG, white blood cell (WBC), C-reactive protein (CRP), and left ventricular ejection fraction (LVEF) levels. Following an overnight fast of at least 8 hours, all blood samples were collected and immediately analysed via standardized protocols.

Definition

DN was defined as a urinary albumin-to-creatinine ratio ≥ 30 mg/g[9], following the exclusion of nondiabetic renal parenchymal diseases and urinary tract infections. DR was assessed via high-resolution fundus photography, with bilateral eye imaging scans obtained for each patient[29], and the results were interpreted by experienced ophthalmologists. DPN was diagnosed when two or more of the following five diagnostic criteria were met: (1) Impaired temperature sensation; (2) Reduced or absent sensation in the feet detected by nylon filament testing; (3) Abnormal vibration sensation; (4) Absent ankle reflexes; and (5) Slowing of two or more nerve conduction velocities[30,31]. In patients whose clinical presentation could not be fully explained by hypertension or coronary artery disease, LVH was diagnosed via a high-resolution ultrasound system by certified echocardiography specialists.

Statistical analysis

Statistical analyses and graphing in this study were performed using RStudio (version 2023.12.0 +369). Normally distributed continuous variables are reported as the mean ± SD, and between-group differences were compared via independent samples t tests. Nonnormally distributed continuous variables are reported as medians with interquartile ranges (25%-75%), and group comparisons were conducted using the Mann-Whitney U test. Categorical variables are expressed as frequencies and percentages, with multiple-group comparisons analysed with χ² tests. Multivariate logistic regression was employed in T2DM patients to evaluate the independent associations between α1-MG and the occurrence of DR, DN, DPN, and LVH. The analysis was adjusted for potential confounding variables, including age, sex, BMI, smoking, alcohol consumption, SBP, DBP, hypertension, hyperlipidaemia, CHD, antihypertensive drugs, lipid-lowering drugs, antidiabetic drugs, FBG, HbA1c, TC, TG, HDL-C, LDL-C, UA, eGFR, ALT, AST, CRP, and LVEF. Multivariate-adjusted restricted cubic spline (RCS) regression was performed to investigate the dose-response associations between serum α1-MG levels and DR, DN, DPN, and LVH. Subgroup analyses were conducted to further assess the stability and reliability of the primary findings. Statistical significance was defined as a P value < 0.05.

Explainable machine learning method model

To investigate the associations between α1-MG levels and diabetic complications in patients with T2DM, we adopted an explainable and interdisciplinary machine learning framework integrating both clinical domain knowledge and advanced algorithmic techniques. In addition to conventional logistic regression analysis, we utilized four advanced machine learning algorithms, namely, LASSO, XGBoost, RF, and SVM, to construct a predictive model for diabetic complication outcomes. For model development and validation, the study cohort was randomly partitioned into training (70%) and testing (30%) datasets while preserving the distribution of key characteristics. To improve transparency and clinical interpretability, we utilized SHapley Additive exPlanations (SHAP)[26], a game theoretic interpretability method that provides: (1) Global feature importance ranking; (2) Granular visualization of how individual feature values influence risk; and (3) Identification of clinically meaningful cut-off points. SHAP values > 0 indicate a positive contribution towards the predicted outcome. This explainable AI approach significantly improves model transparency and clinical interpretability. All analyses were performed via Python (version 3.11.5) with scikit-learn (v0.24.2) and SHAP (v0.40.0).

RESULTS

Baseline characteristics of the study patients

This study included 5045 eligible participants categorized into quartiles (Q1-Q4) according to their serum α1-MG levels (Table 1). The median age was 55 years, with 56.59% being male. Compared with those in the lowest α1-MG quartile (Q1), participants in the highest quartile (Q4) were older and had a greater proportion of men (P < 0.001). Lifestyle factors also varied significantly, with smoking rates rising from 29.16% in Q1 to 41.67% in Q4 and alcohol consumption rates rising from 29.48% to 37.86% (P < 0.001). The comorbidity profile analysis demonstrated significant disparities among the α1-MG quartile groups. The prevalence of hypertension progressively increased from 52.54% in the lowest Q1 quartile to 81.67% in the highest Q4 quartile, whereas the CHD incidence showed a similar trend, increasing from 25.52% in the Q1 quartile to 36.43% in the Q4 quartile (all P < 0.001). Clinical parameters revealed elevated levels of SBP, DBP, BMI, FBG, HbA1c, TG, WBC, CRP, UA, ALT and AST, alongside reduced HDL-C, eGFR, and LVEF in the Q4 group compared with those in the Q1 group (all P < 0.001). In contrast, TC and LDL-C did not significantly differ among the quartile groups (P = 0.247 and P = 0.052, respectively). Compared with patients in the lowest α1-MG quartile group, patients in the highest α1-MG quartile group had an increased proportion of patients utilizing antihypertensive (P < 0.001) and lipid-lowering drugs (P = 0.024). However, no significant difference was observed in antidiabetic medication use among the quartile groups (P = 0.051). The rates of DR, DN, DPN, and LVH tended to increase across the α1-MG quartiles: DN increased from 22.35% in Q1 to 69.44% in Q4; DR increased from 26.23% to 47.86%; DPN increased from 57.61% to 77.54%; and LVH increased from 28.13% to 44.05%. As demonstrated in Supplementary Table 1, α1-MG was significantly positively correlated with SBP, DBP, BMI, TC, LDL-C, TG, HbA1c, FBG, UA, CRP, and WBC (all P < 0.001), whereas it was significantly negatively correlated with HDL-C, ALT, AST, eGFR, and LVEF (all P < 0.001). These associations remained statistically significant after adjusting for age.

Table 1 Baseline characteristics of participants by quartiles, n (%).

Characteristic	Overall (n = 5045)	Q1 (n = 1262)	Q2 (n = 1265)	Q3 (n = 1258)	Q4 (n = 1260)	P value
Sex (male)	2855 (56.59)	602 (47.70)	751 (59.37)	720 (57.23)	782 (62.06)	< 0.001
Age (year)	55.00 (46.00, 64.00)	52.00 (42.25, 61.00)	54.00 (45.00, 63.00)	55.00 (46.25, 64.00)	57.00 (50.00, 67.00)	< 0.001
Smoking	1861 (36.89)	368 (29.16)	491 (38.81)	477 (37.92)	525 (41.67)	< 0.001
Alcohol consumption	1763 (34.95)	372 (29.48)	466 (36.84)	448 (35.61)	477 (37.86)	< 0.001
SBP (mmHg)	129.00 (117.00, 141.00)	125.00 (113.00, 137.00)	128.00 (116.00, 138.00)	130.00 (118.00, 142.00)	134.00 (120.00, 149.00)	< 0.001
DBP (mmHg)	80.00 (72.00, 89.00)	79.00 (71.00, 87.00)	80.00 (72.00, 88.00)	80.00 (72.00, 89.75)	81.00 (73.00, 90.00)	< 0.001
Hypertension	3368 (66.76)	663 (52.54)	802 (63.40)	874 (69.48)	1029 (81.67)	< 0.001
Hyperlipidaemia	3173 (62.89)	709 (56.18)	837 (66.17)	825 (65.58)	802 (63.65)	< 0.001
CHD	1538 (30.49)	322 (25.52)	370 (29.25)	387 (30.76)	459 (36.43)	< 0.001
BMI (kg/m2)	24.39 (22.04, 26.67)	23.76 (21.30, 26.05)	24.46 (22.21, 26.70)	24.65 (22.34, 27.05)	24.56 (22.48, 26.95)	< 0.001
FBG (mmol/L)	7.77 (6.12, 10.04)	7.22 (5.61, 9.50)	7.75 (6.13, 9.83)	7.94 (6.24, 10.40)	8.15 (6.47, 10.39)	< 0.001
HbA1c (%)	7.65 (6.40, 9.35)	7.15 (6.10, 9.00)	7.50 (6.40, 9.10)	7.75 (6.50, 9.40)	8.15 (6.85, 9.70)	< 0.001
TC (mmol/L)	4.53 (3.83, 5.28)	4.50 (3.79, 5.22)	4.54 (3.83, 5.28)	4.55 (3.88, 5.25)	4.52 (3.81, 5.38)	0.247
LDL-C (mmol/L)	2.79 (2.32, 3.31)	2.74 (2.27, 3.24)	2.81 (2.34, 3.35)	2.82 (2.36, 3.30)	2.81 (2.30, 3.37)	0.052
TG (mmol/L)	1.55 (1.10, 2.29)	1.27 (0.91, 1.89)	1.54 (1.11, 2.26)	1.62 (1.18, 2.41)	1.81 (1.29, 2.59)	< 0.001
HDL-C (mmol/L)	1.12 (0.95, 1.32)	1.19 (1.00, 1.41)	1.12 (0.95, 1.31)	1.10 (0.95, 1.27)	1.08 (0.92, 1.27)	< 0.001
WBC (109/L)	6.32 (5.29, 7.54)	5.86 (4.85, 7.13)	6.22 (5.23, 7.37)	6.37 (5.40, 7.54)	6.80 (5.71, 8.08)	< 0.001
CRP (mg/L)	2.92 (1.19, 9.94)	2.35 (0.99, 6.94)	2.35 (1.07, 7.21)	2.97 (1.20, 9.56)	4.38 (1.68, 18.99)	< 0.001
AST (IU/L)	21.40 (17.95, 26.60)	22.00 (18.20, 27.15)	21.80 (18.20, 26.85)	21.40 (18.00, 27.09)	20.55 (17.20, 25.06)	< 0.001
ALT (IU/L)	20.30 (15.00, 29.25)	20.85 (15.22, 30.70)	21.50 (16.00, 30.20)	20.70 (15.40, 30.58)	18.40 (13.70, 26.33)	< 0.001
UA (μmol/L)	332.50 (275.00, 399.50)	297.00 (248.39, 353.08)	324.00 (268.00, 379.50)	335.00 (282.00, 393.00)	386.00 (320.75, 455.25)	< 0.001
eGFR (mL/minute/1.73 m2)	96.31 (78.33, 109.40)	105.94 (95.44, 117.22)	101.10 (91.13, 112.08)	95.30 (81.28, 107.61)	65.27 (41.86, 89.28)	< 0.001
LVEF (%)	64.00 (60.00, 67.00)	64.00 (61.00, 68.00)	64.00 (60.00, 67.00)	64.00 (60.00, 67.00)	63.00 (59.00, 67.00)	< 0.001
Antihypertensive drugs	1040 (20.61)	196 (15.53)	245 (19.37)	282 (22.42)	317 (25.16)	< 0.001
Lipid-lowering drugs	1264 (25.05)	285 (22.58)	304 (24.03)	345 (27.42)	330 (26.19)	0.024
Antidiabetic drugs	1389 (27.53)	321 (25.44)	332 (26.25)	374 (29.73)	362 (28.73)	0.051
DN	1997 (39.58)	282 (22.35)	362 (28.62)	478 (38.00)	875 (69.44)	< 0.001
DR	1772 (35.12)	331 (26.23)	391 (30.91)	447 (35.53)	603 (47.86)	< 0.001
DPN	3351 (66.42)	727 (57.61)	808 (63.87)	839 (66.69)	977 (77.54)	< 0.001
LVH	1685 (33.40)	355 (28.13)	359 (28.38)	416 (33.07)	555 (44.05)	< 0.001
α1-MG (mg/L)	26.10 (21.61, 33.10)	19.02 (17.10, 20.49)	23.80 (22.70, 25.00)	29.13 (27.50, 30.86)	40.87 (36.26, 50.80)	-

Values are presented as median and interquartile range [25%, 75%].

SBP: Systolic blood pressure; DBP: Diastolic blood pressure; CHD: Coronary heart disease; BMI: Body mass index; FBG: Fasting blood glucose; HbA1c: Glycated haemoglobin; TC: Total cholesterol; LDL-C: Low-density lipoprotein cholesterol; TG: Triglycerides; HDL-C: High-density lipoprotein cholesterol; WBC: White blood cell; CRP: C-reactive protein; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; UA: Uric acid; eGFR: Estimated glomerular filtration rate; LVEF: Left ventricular ejection fraction; DN: Diabetic nephropathy; DR: Diabetic retinopathy; DPN: Diabetic peripheral neuropathy; LVH: Left ventricular hypertrophy; α1-MG: Alpha-1-microglobulin.

Association between serum α1-MG and the risk of DR, DN, DPN, and LVH

Serum α1-MG levels were significantly greater in the DN, DR, DPN, and LVH groups than in their respective control groups, as assessed by the Mann-Whitney U test (Figure 1, P < 0.001 for all comparisons). Multivariate logistic regression analysis was performed to investigate the independent associations between serum α1-MG levels and the risk of DN, DR, DPN, and LVH in patients with T2DM (Table 2). After adjusting for comprehensive covariates in Model 3, each increase in the serum α1-MG level was significantly associated with an 84% increased risk of DN (OR: 1.84, 95%CI: 1.62-2.10, P < 0.001), a 17% increased risk of DR (OR: 1.17, 95%CI: 1.07-1.28, P < 0.001), a 14% increased risk of DPN (OR: 1.14, 95%CI: 1.03-1.27, P = 0.014), and a 28% increased risk of LVH (OR: 1.28, 95%CI: 1.18-1.38, P < 0.001). When α1-MG levels were categorized into quartiles (Q1-Q4), patients in the highest quartile (Q4) demonstrated significantly increased risks of multiple complications compared with those in the reference Q1 group following full adjustment in Model 3, with significant associations observed for DN (OR: 2.47, 95%CI: 1.96-3.12, P < 0.001), DR (OR: 1.61, 95%CI: 1.29-2.01, P < 0.001), DPN (OR: 1.44, 95%CI: 1.11-1.87, P = 0.005) and LVH (OR: 1.48, 95%CI: 1.18-1.84, P < 0.001).

Figure 1 Serum α 1-microglobulin levels in patients with and without different diabetic complications. A: With and without diabetic nephropathy; B: With and without diabetic retinopathy; C: With and without diabetic peripheral neuropathy; D: With and without left ventricular hypertrophy. ^aP < 0.001. DN: Diabetic nephropathy; DR: Diabetic retinopathy; DPN: Diabetic peripheral neuropathy; LVH: Left ventricular hypertrophy; α1-MG: Alpha-1-microglobulin.

Table 2 The associations between serum alpha-1-microglobulin level and the risk of complications occurrence in type 2 diabetes mellitus patients.

Complications		Model 1		Model 2		Model 3
		OR (95%CI)	P value	OR (95%CI)	P value	OR (95%CI)	P value
DN	Continuous (per SD)	3.31 (2.99, 3.67)	< 0.001	2.89 (2.60, 3.21)	< 0.001	1.84 (1.62, 2.10)	< 0.001
	Quartiles (Q1-Q4)
	Q1	Reference		Reference		Reference
	Q2	1.39 (1.16, 1.67)	< 0.001	1.17 (0.97, 1.41)	0.108	1.08 (0.88, 1.31)	0.470
	Q3	2.13 (1.79, 2.54)	< 0.001	1.72 (1.43, 2.07)	< 0.001	1.28 (1.05, 1.56)	0.014
	Q4	7.90 (6.61, 9.46)	< 0.001	5.90 (4.88, 7.15)	< 0.001	2.47 (1.96, 3.12)	< 0.001
	P for trend	< 0.001		< 0.001		< 0.001
DR	Continuous (per SD)	1.40 (1.32, 1.49)	< 0.001	1.32 (1.24, 1.41)	< 0.001	1.17 (1.07, 1.28)	< 0.001
	Quartiles (Q1-Q4)
	Q1	Reference		Reference		Reference
	Q2	1.26 (1.06, 1.50)	0.009	1.11 (0.93, 1.33)	0.247	1.11 (0.92, 1.33)	0.287
	Q3	1.55 (1.31, 1.84)	< 0.001	1.33 (1.11, 1.59)	0.002	1.23 (1.02, 1.49)	0.031
	Q4	2.58 (2.19, 3.05)	< 0.001	2.08 (1.74, 2.49)	< 0.001	1.61 (1.29, 2.01)	< 0.001
	P for trend	< 0.001		< 0.001		< 0.001
DPN	Continuous (per SD)	1.41 (1.31, 1.53)	< 0.001	1.21 (1.12, 1.30)	< 0.001	1.14 (1.03, 1.27)	0.014
	Quartiles (Q1-Q4)
	Q1	Reference		Reference		Reference
	Q2	1.30 (1.11, 1.53)	0.001	1.02 (0.85, 1.22)	0.835	1.01 (0.83, 1.22)	0.956
	Q3	1.47 (1.25, 1.73)	< 0.001	1.09 (0.91, 1.31)	0.339	0.98 (0.79, 1.20)	0.812
	Q4	2.54 (2.14, 3.02)	< 0.001	1.64 (1.36, 1.90)	< 0.001	1.44 (1.11, 1.87)	0.005
	P for trend	< 0.001		< 0.001		0.043
LVH	Continuous (per SD)	1.38 (1.30, 1.47)	< 0.001	1.35 (1.26, 1.44)	< 0.001	1.28 (1.18, 1.38)	< 0.001
	Quartiles (Q1-Q4)
	Q1	Reference		Reference		Reference
	Q2	1.01 (0.85, 1.20)	0.889	1.06 (0.88, 1.29)	0.522	0.98 (0.81, 1.19)	0.863
	Q3	1.26 (1.07, 1.50)	0.007	1.21 (1.01, 1.47)	0.043	1.08 (0.89, 1.31)	0.459
	Q4	2.01 (1.71, 2.37)	< 0.001	1.90 (1.58, 2.30)	< 0.001	1.48 (1.18, 1.84)	< 0.001
	P for trend	< 0.001		< 0.001		0.001

Model 1: Unadjusted; Model 2: Adjusted for age, sex, body mass index (BMI), smoking, drinking, systolic blood pressure (SBP), diastolic blood pressure (DBP), hypertension, hyperlipidaemia, and coronary heart disease (CHD); Model 3: Adjusted for age, sex, BMI, smoking, alcohol consumption, SBP, DBP, hypertension, hyperlipidaemia, CHD, antihypertensive drugs, lipid-lowering drugs, antidiabetic drugs, fasting blood glucose, glycated haemoglobin, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, uric acid, estimated glomerular filtration rate, alanine aminotransferase, aspartate aminotransferase, white blood cell, C-reactive protein, and left ventricular ejection fraction. DN: Diabetic nephropathy; DR: Diabetic retinopathy; DPN: Diabetic peripheral neuropathy; LVH: Left ventricular hypertrophy.

Furthermore, RCS analyses were conducted (Figure 2). The levels of α1-MG were related to the onset of DN, DR, DPN, and LVH (all overall P < 0.05). There was a significant nonlinear association between α1-MG levels and the risk of DN and DR (Figure 2A and B, nonlinear P < 0.05). In contrast, the associations of α1-MG with DPN and LVH were more closely aligned with a linear pattern (Figure 2C and D). The RCS results revealed that when the level of α1-MG exceeded 25.85 mg/L, the incidence of DN, DR, DPN, and LVH in T2DM patients gradually increased.

Figure 2 Association of the serum alpha-1-microglobulin level with the risk of different diabetic complications according to restricted cubic splines. A: Diabetic nephropathy; B: Diabetic retinopathy; C: Diabetic peripheral neuropathy; D: Left ventricular hypertrophy. The models were adjusted for age, sex, body mass index, smoking, alcohol consumption, systolic blood pressure, diastolic blood pressure, hypertension, hyperlipidaemia, coronary heart disease, antihypertensive drugs, lipid-lowering drugs, antidiabetic drugs, fasting blood glucose, glycated haemoglobin, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, uric acid, estimated glomerular filtration rate, alanine aminotransferase, aspartate aminotransferase, white blood cell, C-reactive protein, and left ventricular ejection fraction. reference values for estimation were set at median values of alpha-1-microglobulin. α1-MG: Alpha-1-microglobulin.

Subgroup analyses

To further investigate the associations between α1-MG levels and microvascular and cardiac complications, we performed stratified subgroup analyses by age, sex, BMI, hypertension, hyperlipidaemia, and CHD history (Table 3). A more pronounced association between α1-MG levels and DN was observed in subjects aged ≥ 60 years (P for interaction = 0.017) and in patients with hypertension (P for interaction = 0.003). The association between α1-MG levels and DR was significant in patients without CHD (P for interaction < 0.001). No significant interactions were detected between the α1-MG level and DPN. The association of α1-MG with LVH was stronger in men (P for interaction = 0.004).

Table 3 Subgroup analysis of the relationship between serum alpha-1-microglobulin level and complications risk in type 2 diabetes mellitus patients.

	DN (95%CI)	P value	P for interaction	DR (95%CI)	P value	P for interaction	DPN (95%CI)	P value	P for interaction	LVH (95%CI)	P value	P for interaction
Age			0.017			0.392			0.978			0.887
< 60	1.80 (1.54, 2.12)	< 0.001		1.12 (1.01, 1.25)	0.039		1.04 (0.92, 1.18)	0.524		1.28 (1.16, 1.41)	< 0.001
≥ 60	1.98 (1.58, 2.48)	< 0.001		1.22 (1.04, 1.42)	0.011		1.16 (0.94, 1.43)	0.177		1.28 (1.09, 1.49)	0.002
Sex			0.335			0.059			0.379			0.004
Male	1.76 (1.48, 2.08)	< 0.001		1.09 (0.98, 1.22)	0.114		1.09 (0.95, 1.25)	0.213		1.30 (1.15, 1.47)	< 0.001
Female	1.97 (1.59, 2.44)	< 0.001		1.31 (1.13, 1.53)	< 0.001		1.18 (0.99, 1.40)	0.059		1.11 (0.98, 1.25)	0.102
BMI			0.598			0.867			0.549			0.352
< 25	1.83 (1.54, 2.18)	< 0.001		1.15 (1.02, 1.29)	0.027		1.07 (0.93, 1.23)	0.342		1.20 (1.06, 1.36)	0.004
≥ 25	1.85 (1.51, 2.25)	< 0.001		1.20 (1.05, 1.37)	0.007		1.20 (1.02, 1.42)	0.028		1.29 (1.15, 1.46)	< 0.001
Hypertension			0.003			0.060			0.707			0.464
Yes	2.00 (1.71, 2.35)	< 0.001		1.15 (1.05, 1.27)	0.003		1.15 (1.01, 1.30)	0.033		1.29 (1.19, 1.41)	< 0.001
No	1.48 (1.16, 1.89)	0.001		1.29 (1.02, 1.64)	0.033		1.06 (0.83, 1.34)	0.637		1.10 (0.86, 1.42)	0.433
Hyperlipidaemia			0.564			0.270			0.360			0.575
Yes	1.91 (1.61, 2.25)	< 0.001		1.23 (1.10, 1.38)	< 0.001		1.15 (1.00, 1.32)	0.053		1.25 (1.12, 1.38)	< 0.001
No	1.79 (1.44, 2.22)	< 0.001		1.06 (0.92, 1.23)	0.426		1.13 (0.96, 1.34)	0.149		1.21 (1.05, 1.39)	0.010
CHD			0.169			< 0.001			0.690			0.954
Yes	1.61 (1.24, 2.10)	< 0.001		1.06 (0.92, 1.22)	0.460		1.01 (0.82, 1.24)	0.944		1.23 (1.05, 1.43)	0.008
No	1.91 (1.64, 2.22)	< 0.001		1.26 (1.13, 1.41)	< 0.001		1.19 (1.05, 1.35)	0.007		1.31 (1.19, 1.45)	< 0.001

Stratified by age, sex, body mass index (BMI), hypertension, hyperlipidaemia, and coronary heart disease (CHD). Adjusted for age, sex, BMI, smoking, alcohol consumption, systolic blood pressure, diastolic blood pressure, hypertension, hyperlipidaemia, CHD, antihypertensive drugs, lipid-lowering drugs, antidiabetic drugs, fasting blood glucose, glycated haemoglobin, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, uric acid, estimated glomerular filtration rate, alanine aminotransferase, aspartate aminotransferase, white blood cell, C-reactive protein, and left ventricular ejection fraction. DN: Diabetic nephropathy; DR: Diabetic retinopathy; DPN: Diabetic peripheral neuropathy; LVH: Left ventricular hypertrophy; BMI: Body mass index; CHD: Coronary heart disease.

Machine learning for detecting the diagnostic value of α1-MG in diabetic complication risk assessment

To investigate the associations between α1-MG levels and diabetic complications (DN, DR, DPN, and LVH) in patients with T2DM, we employed both conventional logistic regression and advanced machine learning approaches, including LASSO, XGBoost, RF, and SVM, for feature selection. Receiver operating characteristic curve analysis was performed to evaluate the diagnostic performance of each prediction model (Supplementary Figure 1). On the basis of the area under the curve (AUC) values from the validation set, both the XGBoost and RF models were selected for predicting each complication outcome (Supplementary Table 2). SHAP analysis was implemented to visualize variable importance in the prediction models and determine the diagnostic significance of α1-MG. The results demonstrated that serum α1-MG consistently ranked among the top 10 most important variables across all complication models as shown in Figure 3A-D, where red and blue indicate higher and lower feature values, respectively. Positive SHAP values indicate an increased risk, with higher values correlating with a greater likelihood of complication occurrence (Figure 3A-D). SHAP dependence plots revealed that α1-MG values exceeding zero were associated with elevated risks of DN, DR, DPN, and LVH (Figure 3E-H). Individual feature contributions were further illustrated through SHAP force plots (Supplementary Figure 2).

Figure 3 Associations of serum alpha-1-microglobulin with the risk of diabetic complications revealed by SHapley Additive exPlanations summary and dependence plots. A: SHapley Additive exPlanations (SHAP) summary plot for diabetic nephropathy (DN); B: SHAP summary plot for diabetic retinopathy (DR); C: SHAP summary plot for diabetic peripheral neuropathy (DPN); D: SHAP summary plot for Left ventricular hypertrophy (LVH); E: SHAP dependence plot for DN; F: SHAP dependence plot for DR; G: SHAP dependence plot for DPN; H: SHAP dependence plot for LVH. In Figure 3A-D, each dot represents a patient, and the X-axis shows the SHAP value of the corresponding feature. For continuous variables, dot color indicates feature values, ranging from blue (low) to red (high). For categorical variables, blue denotes female sex or positive status, and red denotes male sex or negative status. In Figure 3E-H, the red dashed line represents SHAP = 0, and black arrows indicate threshold points. When the SHAP value of alpha-1-microglobulin exceeds zero, it indicates an increased risk of DN, DR, DPN, and LVH. XGBoost: Extreme gradient boosting; SHAP: SHapley Additive exPlanations; DN: Diabetic nephropathy; DR: Diabetic retinopathy; DPN: Diabetic peripheral neuropathy; LVH: Left ventricular hypertrophy; α1-MG: Alpha-1-microglobulin.

DISCUSSION

Diabetic complications refer to a spectrum of disorders that arise as a consequence of prolonged diabetes mellitus. Microvascular complications are prevalent chronic manifestations of diabetes mellitus, whereas LVH has emerged as a critical cardiovascular complication in T2DM patients[2]. Early detection and intervention are critical for delaying disease progression and improving patient outcomes[10]. Circulating biomarkers[32], including markers related to glycation and oxidative stress [e.g., advanced glycation end products (AGEs) and asymmetric dimethylarginine], endothelial dysfunction (e.g., tumour necrosis factor-α and interleukin-6) and noncoding RNAs (e.g., miR-221 and miR-146a)[33-36], have demonstrated potential for facilitating early detection and targeted management of diabetes-related complications. This study is the first systematic evaluation of the associations between serum α1-MG levels and microvascular and cardiac complications (e.g., LVH) in patients with T2DM. Our findings demonstrated a significant positive correlation between elevated α1-MG levels and the prevalence of these complications, which remained independent after adjusting for multiple confounding factors, suggesting that α1-MG might play a pivotal role in the onset and progression of diabetic complications in T2DM patients.

Previous studies have demonstrated that the serum α1-MG level serves as a biomarker correlated with the histopathological severity of renal impairment and functions as an early predictor of the risk of DN[19,20]. Nevertheless, the comprehensive relationship between serum α1-MG and diabetic complications remains incompletely characterized. In this study, the serum α1-MG levels were significantly greater in T2DM patients with complications. Notably, the incidence of each complication was markedly elevated in the highest α1-MG quartile (Q4). Multivariate logistic regression analysis further revealed a significant dose-dependent association between the α1-MG quartiles and complication risk, with maximal risk elevation observed in the highest quartile. These results indicate that α1-MG might not only serve as a marker of renal injury but also be associated with retinopathy and neuropathy, as well as cardiac structural abnormalities. Moreover, previous studies have established that α1-MG has antioxidant properties and is capable of scavenging free radicals and protecting cells from oxidative stress damage under physiological conditions[13,37]. However, in the chronic hyperglycaemic environment of T2DM, persistent oxidative stress might exceed the scavenging capacity of α1-MG, thereby diminishing or abolishing its antioxidant function[38]. The positive correlation between α1-MG and the inflammatory factors CRP and HbA1c observed in this study suggested that elevated α1-MG levels might reflect chronic hyperglycaemic exposure while implicating inflammatory and oxidative stress pathways in its pathological mechanisms. This increase in α1-MG might exacerbate microvascular damage through synergistic interactions with reactive oxygen species and AGEs[33,39]. Nevertheless, experimental studies are warranted to further validate causality and elucidate the specific molecular pathways involved.

The risk of diabetes complications is influenced by factors such as age, blood pressure, and metabolic status[40-42]. Our subgroup analyses revealed heterogeneous associations between serum α1-MG levels and microvascular complications and between serum α1-MG levels and LVH across different baseline characteristics. Specifically, the link between α1-MG and DN was stronger in elderly patients (aged over 60 years) with hypertension, and we observed a statistically significant interaction between α1-MG and hypertension status in the DN subgroup (P for interaction = 0.003). These findings suggest that the impact of α1-MG on DN risk may be amplified in hypertensive individuals. Importantly, however, even among patients without hypertension, each SD increase in α1-MG remained significantly associated with DN risk (OR = 1.48, 95%CI: 1.16-1.89, P = 0.001), indicating an independent effect of α1-MG. In contrast, among patients without a history of CHD, the association between α1-MG and DR was stronger. Although few studies have explored the relationship between α1-MG and cardiac complications, we observed that the association between α1-MG and LVH was particularly strong in men, indicating that sex might affect the impact of α1-MG on cardiac structural abnormalities. These findings suggest that the role of α1-MG in microvascular complications and LVH varies depending on patients’ baseline characteristics, suggesting that hormonal differences, adipose tissue distribution and metabolic stress potentially modulate the contribution of α1-MG to disease progression. Although stratification reduced sample sizes and influenced effect sizes, the consistent direction of the results reinforced the reliability of these observational analyses.

Additionally, although self-supervised contrastive learning has recently shown significant potential in representation learning from medical imaging and high-dimensional omics data[43], its applicability to structured tabular data with explicit outcome labels is currently limited. Therefore, we selected a prediction model that combines traditional logistic regression with advanced machine learning algorithms (LASSO, XGBoost, RF, and SVM) and validated it with AUC metrics. The feature attribution through SHAP analysis revealed that the contribution of the serum α1-MG level to the risk of complications in all the models was consistent and dose dependent. The latest developments in machine learning may further enhance predictive performance and interpretability. For example, Transformer-based time modelling has shown promising prospects in longitudinal clinical data because it captures remote dependencies that do not require manual feature engineering[44,45]. Future prospective studies that include continuous measurement of serum α1-MG levels may benefit from time modelling to more effectively capture disease progression patterns and delayed biomarker effects. In addition, the federated learning framework allows us to collaborate on model training across multiple clinical sites while protecting patient data privacy[46]. These emerging paradigms can complement existing methods and provide promising directions for future research.

The dose-dependent associations of α1-MG with DN, DR, DPN, and LVH support its potential integration into T2DM management. A higher level of α1-MG (over 25.85 mg/L) could serve as a threshold value for alerting patients to conduct early screening for diabetic complications via albuminuria tests, fundus photography, etc. High-risk patients might benefit from intensified glycaemic control or SGLT2 inhibitor treatment, which have been shown to reduce oxidative stress and microvascular damage[47]. Routine α1-MG monitoring is suggested as a strategy to stratify patients for personalized care, thereby improving outcomes and aligning with the principles of precision medicine in diabetes management.

Several limitations warrant acknowledgement. First, this was a retrospective single-centre study based on electronic medical records, which may have introduced patient selection bias and limitations in data completeness. In particular, detailed quantitative information on time-dependent variables such as smoking and alcohol consumption was not consistently available. As a result, these variables were recorded as binary indicators (yes/no), which may have led to misclassification bias and limited the precision of risk factor assessment. Nevertheless, the large sample size may have partially mitigated these limitations. Second, the cross-sectional design did not reflect the causal relationship between elevated levels of α1-MG and complications, despite multivariate adjustments and subgroup analyses enhancing the reliability of the associations between them. Additionally, owing to the limitations of the existing data, such as the lack of specific oxidative stress markers or endothelial function parameters, the conclusions were not sufficiently grounded in the mechanism. The relationship between α1-MG and diabetic complications requires further investigation, potentially through in vivo and in vitro experimental systems, to elucidate the underlying mechanisms involved. Moreover, establishing clinical thresholds for α1-MG and assessing its predictive value in combination with other biomarkers could enhance its utility as a predictive biomarker, thereby improving management strategies for patients with T2DM.

CONCLUSION

In conclusion, this study identified elevated serum α1-MG levels as an independent risk factor for DN, DR, DPN, and LVH in T2DM patients, with the strongest association observed for DN. These findings suggest a potential role of α1-MG in the pathogenesis and progression of microvascular and cardiac complications in patients with T2DM and highlight its potential clinical utility as a predictive marker of complications.

ACKNOWLEDGEMENTS

The authors would like to acknowledge the support of Professor Sen Fang from Nanjing Shensen Technology Co., Ltd. for the contributions of data processing and analysis.