Risk factors and a predictive model of diabetic foot in hospitalized patients with type 2 diabetes

Abstract

BACKGROUND

The risk factors and prediction models for diabetic foot (DF) remain incompletely understood, with several potential factors still requiring in-depth investigations.

AIM

To identify risk factors for new-onset DF and develop a robust prediction model for hospitalized patients with type 2 diabetes.

METHODS

We included 6301 hospitalized patients with type 2 diabetes from January 2016 to December 2021. A univariate Cox model and least absolute shrinkage and selection operator analyses were applied to select the appropriate predictors. Nonlinear associations between continuous variables and the risk of DF were explored using restricted cubic spline functions. The Cox model was further employed to evaluate the impact of risk factors on DF. The area under the curve (AUC) was measured to evaluate the accuracy of the prediction model.

RESULTS

Seventy-five diabetic inpatients experienced DF. The incidence density of DF was 4.5/1000 person-years. A long duration of diabetes, lower extremity arterial disease, lower serum albumin, fasting plasma glucose (FPG), and diabetic nephropathy were independently associated with DF. Among these risk factors, the serum albumin concentration was inversely associated with DF, with a hazard ratio (HR) and 95% confidence interval (CI) of 0.91 (0.88-0.95) (P < 0.001). Additionally, a U-shaped nonlinear relationship was observed between the FPG level and DF. After adjusting for other variables, the HRs and 95%CI for FPG < 4.4 mmol/L and ≥ 7.0 mmol/L were 3.99 (1.55-10.25) (P = 0.004) and 3.12 (1.66-5.87) (P < 0.001), respectively, which was greater than the mid-range level (4.4-6.9 mmol/L). The AUC for predicting DF over 3 years was 0.797.

CONCLUSION

FPG demonstrated a U-shaped relationship with DF. Serum albumin levels were negatively associated with DF. The prediction nomogram model of DF showed good discrimination ability using diabetes duration, lower extremity arterial disease, serum albumin, FPG, and diabetic nephropathy (Clinicaltrial.gov NCT05519163).

Key Words: Type 2 diabetes; Diabetic foot; Nonlinear association; Prediction model; Retrospective cohort

Core Tip: The risk factors and prediction models for diabetic foot (DF) remain inconclusive, and various predictors may contribute to its development as type 2 diabetes progresses. In this study, we analyzed 6301 inpatients with type 2 diabetes in China from January 2016 to December 2021. Our findings revealed that fasting plasma glucose was positively associated with DF, both at low and high levels, while albumin showed a negative association. Additionally, prolonged duration of diabetes, lower extremity arterial disease, and diabetic nephropathy were related to DF. Based on these insights, targeted prevention strategies are needed for predictors with varying implications.

INTRODUCTION

Diabetic foot (DF) is characterized by infection of the lower extremity, ulceration, and deep tissue damage, which primarily results from neuropathy and varying degrees of vascular disease in individuals with diabetes[1]. This condition represents a major complication arising from unmanaged diabetes. DF ulcer (DFU) is the most prevalent form of DF, and it contributes to high rates of amputation and mortality[2]. Recent studies have indicated that the incidence of DF ranges from 1.9% to 8.1%[3-6], with a recurrence rate of approximately 31.6%[4]. Furthermore, DF poses significant threats to the physical and mental well-being of patients, whereby it adversely affects their quality of life; moreover, it exerts a substantial burden on society. The prediction of DF and the implementation of early preventative measures for patients with diabetes are crucial[7]. Moreover, the early identification of patients who are at high risk of DF and the proactive management of risk factors are critical for halting the progression of diabetes and decreasing the likelihood of hospitalization, amputation, and death[8,9].

A systematic review and meta-analysis revealed that complications of diabetes, such as diabetic nephropathy, diabetic retinopathy, and diabetic peripheral neuropathy (DPN), are risk factors for DFU[3]. Furthermore, research has identified factors such as fasting blood glucose (FPG), duration of diabetes, insulin levels, serum bilirubin, fibrinogen, D-dimer, and red blood cell distribution width coefficient of variation (RDW-CV) as being predictive markers for DF[10-12]. Despite these findings, several research areas regarding this condition remain underexplored. Notably, studies on DF in China are sparse[2], and the integration of new indicators or a focus on diverse populations may reveal unknown risk factors. Furthermore, many predictive models target recurrent DF, with few models addressing new-onset cases of the disease[3,5,12,13]. Finally, the nonlinear relationships between continuous variables and DF have not been fully elucidated[12].

In this longitudinal, retrospective cohort study, we included 6301 hospitalized patients with type 2 diabetes from Shandong Provincial Qianfoshan Hospital (Shandong, China). Potential risk factors for DF were identified using the Cox model and least absolute shrinkage and selection operator (LASSO) analyses. A restricted cubic spline function was utilized to explore both linear and nonlinear associations between continuous variables and DF outcomes. Then the Cox model was employed to evaluate the relationships between the predictors and DF. This predictive model was designed to aid clinicians in identifying high-risk groups for DF among patients with type 2 diabetes and offers significant guidance for clinical prevention.

MATERIALS AND METHODS

Study population and data collection

The data for this study were collected from the Shandong Provincial Qianfoshan Hospital Healthcare Big Data Platform (SPQHHBDP). The SPQHHBDP integrates multisource data from hospital information systems, electronic medical records (EMRs), laboratory information management systems, picture archiving and communication systems, and nursing information systems. The encrypted personal identification number was used as a unique identifier to interlink each individual's data information in the abovementioned databases.

In this retrospective cohort study, data were collected from January 2016 to December 2021 and involved 19148 patients with diabetes aged 20 years and older, with a follow-up period of more than 1 year. The exclusion criteria included: (1) Any history of DF or amputation (n = 485); (2) Patients with type 1 diabetes (n = 73); (3) Being pregnant during the study period (n = 3622); (4) The presence of malignant tumors (n = 4454); (5) Severe organ failure including heart, liver, and kidney conditions (n = 1599); and (6) Missing data (n = 2614). Finally, 6301 hospitalized patients with type 2 diabetes were included in the analyses.

Potential risk factors

The baseline characteristics were potential risk factors for DF, with each factor being linked by an encrypted personal identification number. The detailed information is as follows: (1) Demographic indicators included age and sex; (2) Anthropometric examination indicators included body mass index, systolic blood pressure, and diastolic blood pressure; (3) Laboratory examination indicators were measured by laboratory specialists via standard clinical and laboratory protocols, which included FPG, plasma fibrinogen, hemoglobin, RDW-CV, lymphocyte count, white blood cell count (WBC), albumin, creatinine, alanine aminotransferase, serum aspartate aminotransferase, total bilirubin, indirect bilirubin, and serum uric acid; and (4) Disease- and lifestyle-related information, including diabetes duration, smoking status, and alcohol consumption, was recorded in the personal history in the EMR. Additionally, detailed histories of comorbidities and drug use, such as hyperlipidemia, hypertension, stroke, ischemic heart disease, lower extremity arterial disease, chronic venous insufficiency of the lower extremities, diabetic nephropathy, PN, and insulin use, were extracted from hospital diagnoses and personal histories in the EMR.

Outcome

DF was defined based on the self-reported disease history and clinical diagnosis. New-onset DF was diagnosed as the first DF occurrence during the study period.

Statistical analyses

The baseline characteristics are presented as the mean ± standard deviation if they were normally distributed, or medians (interquartile ranges) if they were not normally distributed. Categorical variables are expressed as numbers (percentages). Comparisons between the DF and non-DF groups were conducted via t-tests, Wilcoxon signed-rank tests, or χ² tests.

Variables with P < 0.05 in univariate Cox models were selected as candidates. The predictive variables were further refined using LASSO regression. Then a multivariate Cox regression model was implemented. Both linear and nonlinear relationships between continuous variables and the DF outcome were analyzed using a restricted cubic spline function with five knots adjusted for age and sex. When a nonlinear relationship was identified, continuous variables were classified based on relevant guidelines or at the quartile level. Kaplan-Meier survival curves, which are based on these classifications, were analyzed via the log-rank test. The predictive performance was assessed using receiver operating characteristic (ROC) curves and validated via 10-fold cross-validation. A nomogram was developed to predict new-onset DF at 2, 3, and 5 years, and decision curve analysis was subsequently performed via the developed prediction model.

To validate the robustness of the risk estimates, a down-sampling method was employed, in which samples of 5301, 4301, 3301, and 2301 individuals were randomly selected to replicate the Cox models.

The detailed process of the analyses is shown in Figure 1. All of the analyses were performed using R software version 4.1.3. The “survival” package was used to fit the Cox model, the “rms” package was used to calculate the restricted cubic spline function, the “survminer” and “ggplot2” packages were used to plot the Kaplan-Meier curve, the “pROC” package was used to draw the ROC curve, and the “rmda” package was used for implementing decision curve analysis. Two-tailed P < 0.05 was considered statistically significant.

Figure 1 Process of analyses. DF: Diabetic foot; FPG: Fasting plasma glucose; LASSO: Least absolute shrinkage and selection operator.

RESULTS

A total of 6301 patients with type 2 diabetes were included in the study (Table 1). As of December 31, 2021, there were 75 cases of new-onset DF, with a median follow-up duration of 29.2 (12.2-72.8) months. The incidence density for DF was 4.5/1000 person-years. The mean age in the new-onset DF group was 67.0 ± 10.1 years, which was significantly greater than that in the control group (63.0 ± 11.4 years, P = 0.006). The proportion of females was similar between the new-onset DF group (44.0%) and the control group (43.9%), with no significant difference being observed (P = 0.935). The prevalence of disease history, including insulin usage, hypertension, stroke, lower extremity arterial disease, diabetic nephropathy, PN, and retinopathy, was significantly greater among patients with new-onset DF. Additionally, patients with DF had longer durations of diabetes. Patients with new-onset DF also exhibited higher levels of FPG, plasma fibrinogen, RDW-CV, WBC, serum creatinine, and serum uric acid compared to those without DF. Conversely, the levels of hemoglobin, albumin, alanine aminotransferase, total bilirubin, and indirect bilirubin were lower in patients with new-onset DF. Twenty variables associated with DF were identified via univariate Cox regression analysis (P < 0.05) and were selected as the candidate variables.

Table 1 Baseline characteristics of the studies population and hazard ratio (95% confidence interval) of univariate Cox model for new-onset diabetic foot, n (%).

Variables	Without diabetic foot (n = 6226)	Incident diabetic foot (n = 75)	HR (95%CI)	P value
Demographic indicators
Age, year	63.0 ± 11.4	67.0 ± 10.1	1.03 (1.01-1.05)	0.006
Sex
Male	3491 (56.1)	42 (56.0)	Reference	-
Female	2735 (43.9)	33 (44.0)	1.02 (0.65-1.61)	0.935
Clinical characteristics
DM duration
< 10 years	3571 (57.3)	21 (28.0)	Reference	-
10-19 years	1797 (28.9)	30 (40.0)	2.58 (1.48-4.51)	0.001
≥ 20 years	858 (13.8)	24 (32.0)	4.49 (2.50-8.07)	< 0.001
Insulin application	3277 (52.6)	56 (74.7)	2.90 (1.72-4.88)	< 0.001
Hyperlipidemia	775 (12.5)	9 (12.0)	0.87 (0.44-1.75)	0.705
Hypertension	3966 (63.7)	59 (78.7)	2.01 (1.16-3.49)	0.013
Stroke	1372 (22.0)	25 (33.3)	1.86 (1.15-3.00)	0.012
Ischemic heart disease	2732 (43.9)	42 (56.0)	1.57 (1.00-2.48)	0.051
Lower extremity arterial disease	210 (3.4)	15 (20.0)	8.06 (4.57-14.20)	< 0.001
Chronic venous insufficiency of the lower extremities	190 (3.1)	3 (4.0)	1.35 (0.43-4.28)	0.612
Diabetic nephropathy	356 (5.7)	20 (26.7)	6.31 (3.78-10.50)	< 0.001
Peripheral neuropathy	595 (9.6)	12 (16.0)	1.88 (1.01-3.49)	0.045
Retinopathy	512 (8.2)	11 (14.7)	2.01 (1.06-3.82)	0.032
Behavior and habits
Smoking	1893 (31.9)	25 (33.8)	1.10 (0.68-1.78)	0.697
Alcohol drinking	2050 (34.5)	22 (29.7)	0.82 (0.50-1.34)	0.426
Physical examination
BMI, kg/m2	21.6 ± 3.3	21.5 ± 2.8	0.99 (0.90-1.10)	0.863
SBP, mmHg	142.0 ± 21.0	143.7 ± 23.9	1.00 (0.99-1.02)	0.502
DBP, mmHg	80.8 ± 12.5	80.1 ± 12.3	1.00 (0.97-1.02)	0.738
Laboratory indicators
FPG, mmol/L	8.3 ± 3.2	10.1 ± 4.6	1.13 (1.08-1.19)	< 0.001
Plasma fibrinogen, g/L	3.0 ± 1.0	3.5 ± 1.3	1.47 (1.25-1.72)	< 0.001
Hemoglobin, g/L	135.5 ± 20.2	127.0 ± 25.4	0.98 (0.97-0.99)	< 0.001
RDW-CV, %	12.7 ± 1.2	13.0 ± 1.2	1.18 (1.05-1.32)	0.005
Lymphocyte count, 109/L	1.9 ± 0.7	1.7 ± 0.7	0.69 (0.48-0.98)	0.037
WBC, 109/L	7.1 ± 2.6	8.1 ± 3.6	1.12 (1.06-1.18)	< 0.001
Albumin, g/L	42.4 ± 5.1	38.6 ± 6.3	0.90 (0.87-0.93)	< 0.001
Serum creatinine, μmol/L, median (IQR)	66.0 (55.0-79.0)	77.0 (54.5-130.0)	1.01 (1.00-1.01)	< 0.001
Alanine aminotransferase, U/L, median (IQR)	16.6 (12.1-24.8)	13.8 (11.1-21.6)	0.97 (0.95-0.99)	0.022
Serum aspartate aminotransferase, U/L, median (IQR)	17.8 (14.6-22.9)	16.5 (14.3-20.8)	0.98 (0.96-1.01)	0.115
Total bilirubin, μmol/L	11.6 ± 6.7	9.7 ± 4.2	0.93 (0.88-0.98)	0.010
Indirect bilirubin, μmol/L	7.1 ± 4.2	5.7 ± 2.9	0.87 (0.80-0.95)	0.001
Serum uric acid, μmol/L	302.7 ± 90.0	311.0 ± 118.4	1.00 (0.99-1.01)	0.418

The characteristics are presented as the mean ± standard deviation if they were normally distributed, or medians (interquartile ranges) if they were not normally distributed. BMI: Body mass index; CI: Confidence interval; DBP: Diastolic blood pressure; DM: Diabetes mellitus; FPG: Fasting plasma glucose; HR: Hazard ratio; IQR: Interquartile range; RDW-CV: red blood cell distribution width coefficient of variation; SBP: Systolic blood pressure; WBC: White blood cell.

We further applied the LASSO method to select predictors. Five variables (diabetes mellitus duration, diabetic nephropathy, arterial disease of the lower extremities, albumin and FPG) were linked to the risk of new-onset DF, as depicted in Figure 2. The multivariable Cox proportional hazards model (Table 2) indicated that a longer duration of diabetes, low and high levels of FPG, lower extremity arterial disease, and diabetic nephropathy independently increased the risk of new-onset DF. Conversely, higher levels of albumin were associated with a reduced risk, with a hazard ratio (HR) and 95% confidence interval (CI) of 0.91 (0.88-0.95) (P < 0.001).

Figure 2 Predictive variable selection via the least absolute shrinkage and selection operator Cox regression model. The least absolute shrinkage and selection operator Cox regression model was constructed from 20 candidate variables, and the minimum standard was adopted to obtain the value of the super parameter λ by 10-fold cross-validation for variable selection. Dotted vertical lines were drawn at the optimal values by using the minimum criteria and the 1 standard error (SE) from the minimum criteria (the 1 - SE criteria).

Table 2 The predictive model of new-onset diabetic foot over the whole follow-up period in hospitalized patients with type 2 diabetes.

Variable	HR (95%CI)	P value
Diabetes duration
< 10 years	Reference	-
10-19 years	1.86 (1.05-3.28)	0.032
≥ 20 years	2.96 (1.63-5.39)	< 0.001
Albumin	0.91 (0.88-0.95)	< 0.001
FPG
4.4-6.9 mmol/L	Reference	-
< 4.4 mmol/L	3.99 (1.55-10.25)	0.004
≥ 7.0 mmol/L	3.12 (1.66-5.87)	< 0.001
Diabetic nephropathy	2.23 (1.25-3.99)	0.007
Lower extremity arterial disease	6.21 (3.44-11.24)	< 0.001

CI: Confidence interval; HR: Hazard ratio; FPG: Fasting plasma glucose.

Using restricted cubic spline modeling, a nonlinear association was observed between FPG levels and new-onset DF (Figure 3A, P for linear < 0.001, P for nonlinear = 0.005). According to the FPG control goals outlined in the Chinese guidelines for type 2 diabetes[14], FPG levels were categorized into three groups: < 4.4 mmol/L, 4.4-6.9 mmol/L, and ≥ 7.0 mmol/L. After adjusting for the other variables in Table 2, compared with the moderate level (4.4-6.9 mmol/L), the HR and 95%CI for FPG levels of < 4.4 mmol/L and ≥ 7.0 mmol/L were 3.99 (1.55-10.25) (P = 0.004) and 3.12 (1.66-5.87) (P < 0.001), respectively. Kaplan-Meier survival curves for these FPG categories revealed significant differences in cumulative hazard (P < 0.001; Figure 4). Additionally, a negative correlation was observed between the serum albumin levels and the risk of new-onset DF (P for linear < 0.001, P for nonlinear = 0.910; Figure 3B).

Figure 3 Age- and sex- adjusted restricted cubic spline modeling of the relationships of fasting plasma glucose and albumin with the risk of new-onset diabetic foot. A: Effect of fasting plasma glucose (FPG) on the risk of new-onset diabetic foot; B: Effect of serum albumin on the risk of new-onset diabetic foot. HR: Hazard ratio.

Figure 4 Kaplan-Meier curves of new-onset diabetic foot cumulative hazard for patients with type 2 diabetes grouped by different levels of fasting plasma glucose. FPG: Fasting plasma glucose.

We developed a nomogram for predicting new-onset DF over 2, 3, and 5 years of follow-up based on variables that were identified from the Cox model (Supplementary Figure 1). The area under the curve (AUC) for the 3-year predictive model was 0.797 (Supplementary Figure 2), with a sensitivity and specificity of 0.707 and 0.757, respectively, and with a 10-fold cross-validation AUC of 0.782. Decision curve analysis revealed a certain net benefit across a range of threshold probabilities (Supplementary Figure 3). In the validation analyses, albumin, FPG, diabetic nephropathy, and lower extremity arterial disease had significant effects on DF in all of the randomly down sampled diabetic inpatients, which was similar to the total population (Table 3).

Table 3 The predictive model of new-onset diabetic foot over the whole follow-up period in hospitalized patients with type 2 diabetes using a down-sampling approach.

Variable	Sampled population, n = 5301		Sampled population, n = 4301		Sampled population, n = 3301		Sampled population, n = 2301
	HR (95%CI)	P value	HR (95%CI)	P value	HR (95%CI)	P value	HR (95%CI)	P value
Diabetes duration
< 10 years	Reference	-	Reference	-	Reference	-	Reference	-
10-19 years	1.87 (1.03-3.39)	0.041	1.80 (0.95-3.41)	0.072	2.68 (1.20-6.00)	0.016	3.45 (1.22-9.75)	0.019
≥ 20 years	2.42 (1.25-4.69)	0.008	1.93 (0.92-4.06)	0.082	2.31 (0.89-5.99)	0.085	2.65 (0.79-8.88)	0.113
Albumin	0.92 (0.88-0.96)	< 0.001	0.93 (0.89-0.97)	0.001	0.92 (0.87-0.96)	0.001	0.91 (0.86-0.97)	0.006
FPG
4.4-6.9 mmol/L	Reference	-	Reference	-	Reference	-	Reference	-
< 4.4 mmol/L	3.92 (1.43-10.74)	0.008	4.31 (1.45-12.80)	0.008	6.94 (1.84-26.17)	0.004	7.65 (1.87-31.29)	0.005
≥ 7.0 mmol/L	2.70 (1.40-5.24)	0.003	2.40 (1.19-4.86)	0.014	3.68 (1.41-9.56)	0.008	3.41 (1.13-10.22)	0.029
Diabetic nephropathy	2.47 (1.32-4.64)	0.005	2.96 (1.49-5.89)	0.002	3.30 (1.51-7.22)	0.003	3.48 (1.36-8.94)	0.009
Lower extremity arterial disease	5.15 (2.66-9.96)	< 0.001	3.84 (1.74-8.46)	0.001	3.28 (1.23-8.79)	0.018	4.34 (1.52-12.40)	0.006

The predictive Cox model adjusted diabetes duration, diabetic nephropathy, and lower extremity arterial disease as covariates. CI: Confidence interval; FPG: Fasting plasma glucose; HR: Hazard ratio.

DISCUSSION

Based on a retrospective cohort study of inpatients with type 2 diabetes, we established a robust predictive model for new-onset DF by using five readily available variables: duration of diabetes mellitus, diabetic nephropathy, arterial disease of the lower extremities, FPG, and albumin. The predictive accuracy was robust and similar to that reported in other studies[3,5,13]. To the best of our knowledge, this is the first study to demonstrate a U-shaped nonlinear association between FPG and DF. Our findings indicate that both low and high levels of FPG are linked to the progression of DF. Moreover, we highlighted the protective role of albumin in preventing the onset of DF disease. The identification of key risk factors and the development of predictive models are crucial for clinicians to recognize high-risk individuals and to implement early preventative measures based on the differential impacts of targeted risk factors.

This study identified several classic risk factors for DF among inpatients with type 2 diabetes. We observed that the risk of DF increased with the duration of diabetes, which aligns with findings from previous studies[5,15]. Our research confirmed that arterial disease of the lower extremities plays a crucial role in the development of DF, which is consistent with related studies[7]. The early detection of diabetes, timely management of its progression, and effective handling of diabetic complications are essential for preventing DF.

Our study revealed a novel U-shaped relationship between FPG and new-onset DF. Prior research on DF has primarily focused on the impact of glycosylated hemoglobin (HbA1c) rather than the nonlinear effect of FPG[3,5,13]. The clinical implications of FPG and HbA1c have been shown to differ. For example, the association between HbA1c and FPG is stronger at higher glucose levels. These factors are closely aligned in the diagnosis of diabetes but are not coincident in the prediction of intermediate hyperglycemia[16]. Moreover, the relationship between the HbA1c level and hypoglycemia in individuals with type 2 diabetes remains unclear[17]. Thus, our study investigated the relationship between FPG and new-onset DF. Hyperglycemia can lead to advanced glycation end-products accumulation[18] which contributes to impaired wound healing, vascular damage, neuropathy[18], and metabolic abnormalities, and results in the development of DF and its complications[19]. By accelerating atherosclerosis, inducing endothelial damage, provoking inflammatory reactions, and causing arterial constriction, hyperglycemia affects macro- and microvascular functionality[20], thus resulting in inadequate blood flow to the lower extremities and damage to peripheral nerves. In contrast, hypoglycemia is a frequent complication in diabetic patients receiving insulin treatment[21]. A previous study has identified hypoglycemia as being a potential independent risk factor for amputations in patients hospitalized with DF[22]. The impact of hypoglycemia on DF may be attributable to its association with increased physical morbidity[21], such as malnutrition, infections, severe illnesses, and vascular complications[22,23], as well as psychological challenges such as fear[21] and depression, which can exacerbate diabetic complications[24]. Furthermore, poor adherence to treatment, which is often linked to hypoglycemia[25], increases the risk of diabetic complications[26]. Although the precise biochemical mechanisms remain unclear, proactive efforts to prevent hypoglycemia are crucial in managing DF in patients with type 2 diabetes.

Notably, diabetic nephropathy was shown to be an effective predictor of DF. Our study revealed that patients with nephropathy have a 2.23-fold increased risk of developing new-onset DF compared with those without nephropathy, which is similar to the findings from previous research[4]. Uremia can lead to decreased immunity[27], which increases the risks of foot infection and nonhealing wounds. Additionally, hypertension and atherosclerosis[28] associated with diabetic nephropathy can reduce blood flow to the lower extremities, thus potentially accelerating the development of DF. Persistent hyperglycemia may also damage endothelial cells within the renal vasculature[29], thus leading to abnormal cellular function, chronic inflammation, and oxidative stress[30]. These factors can impair the microvascular endothelium, resulting in ischemia, tissue hypoxia, and microcirculation disorders, which subsequently induce neuropathy. Additionally, metabolic abnormalities such as electrolyte and acid-base imbalances caused by diabetic nephropathy[31] can exacerbate neuropathy and ischemic conditions, thus increasing the susceptibility of patients with diabetes to foot ulcers and infections.

We identified a novel impact of lower albumin levels on the development of DF. Albumin, which is a functional protein, plays a pivotal role in neutralizing free radicals and maintaining antioxidant effects. Low serum albumin levels may indicate poor nutritional status and systemic inflammation[32]. Prior research has shown that hypoalbuminemia (< 3.5 g/dL) can act as a biomarker for nonhealing DFU[33]. Yan et al[34] reported a negative correlation between albumin levels and DPN in Chinese patients with type 2 diabetes. Higher albumin levels have been found to be associated with a decreased risk of DPN[35], which can directly or indirectly lead to DF[32,36]. Low serum albumin levels may indicate poor nutritional status and systemic inflammation[32,37], thus exacerbating tissue breakdown, impeding recovery, reducing osmotic pressure, causing local tissue edema, and impairing skin barrier function. These mechanisms may elucidate the underlying pathophysiological processes of this condition.

We constructed a nomogram model to predict the 2-, 3-, and 5-year DF-free probabilities. The AUC of the nomogram was 0.797, with a 10-fold cross-validation score of 0.782, which indicated a sufficient and stable discrimination ability. This model can provide physicians with a reference for adjusting treatment strategies. Jiang et al[38] incorporated 12 factors to construct a nomogram for predicting DFU, achieving a good concordance index of 0.84 in the validation cohort. However, the only similarly identified risk factor was a longer duration of diabetes, which indicates the heterogeneity of DF predictors among patients with diabetes. Compared with our study, the nomogram developed by Lv et al[39] exhibited similar discrimination (AUC = 0.741 in the primary cohort) for DFU in patients with diabetes but identified different predictors. The diversity in the predictors and DF incidences[38,39] may be due to the absence of significant variables in various datasets, eras[40], climates, regions, medical environments, or study designs, as well as discrepancies in disease definitions[40]. Further studies incorporating multicenter and large-scale cohort data are needed to examine the varied conditions among patients with DF.

In this retrospective cohort study, we analyzed a group of hospitalized patients with type 2 diabetes by utilizing EMR data. The strengths of our study include a lengthy follow-up period of at least one year, a focus on new-onset DF events, and the identification of novel risk factors. However, there were several limitations. Primarily, the study was confined to a single center with a modest sample size, which may limit the generalizability of our predictive model without external data support. The sample included 6226 patients with diabetes and only 75 patients with DF, which may reduce statistical power, increase the variability of the results, and reduce the chance of detecting the true effect of predictors. Second, the reliance on EMR data from hospitalized patients at a single institution may have led to an underestimation of the DF rate. Patients may receive treatment in our hospital and then transfer treatment to other institutions or return to place of residence for treatment. As a result, individuals were lost to follow-up, and current data do not fully reflect patient outcomes. Third, patients with DF had higher age and a longer duration of diabetes mellitus which may lead to additional biases and influence the conclusion. Additionally, this study only included easily available variables during treatment in hospital, indicators such as HbA1c, C-reactive protein, and fungal foot infection had high missing rate which make it difficult to calculate precise risk estimate. Further studies are needed to enhance the applicability of this study.

CONCLUSION

We investigated the risk factors and established a predictive model for new-onset DF in a cohort of hospitalized patients with type 2 diabetes. A longer duration of diabetes, arterial disease of the lower extremities, low serum albumin levels, low and high levels of FPG, and diabetic nephropathy were independently associated with DF. FPG exhibited a U-shaped relationship with DF. The serum albumin concentration was negatively associated with DF. The Cox model demonstrated robust predictive ability for DF. The prediction nomogram model of DF showed good discrimination ability. Further studies are needed to validate the reproducibility of the results.

ACKNOWLEDGEMENTS

The authors thank SPQHHBDP for collecting data and researchers for participating in this study.