A nomogram for predicting bladder dysfunction in patients with type 2 diabetes mellitus: a retrospective study

Introduction

Type 2 diabetes mellitus (T2DM) is among the most prevalent chronic diseases globally, affecting approximately 529 million individuals as of 2021. By 2050, this number is projected to exceed 1.31 billion (GBD 2021 Diabetes Collaborators, 2023). Diabetic bladder dysfunction (DBD) is a common and serious complication of T2DM that significantly impacts patients’ physical and mental health (Xu et al., 2017). Studies indicate that 25–87% of T2DM patients experience varying degrees of DBD (Wittig et al., 2019). Also referred to as diabetic neurogenic bladder (Moller, 1976), DBD is characterized by impaired bladder sensory nerves, reduced detrusor muscle contractility, increased bladder capacity, and elevated residual urine volume (Wittig et al., 2019). This condition can lead to upper urinary tract damage, including pyelonephritis, hydronephrosis, and ureteral dilation (Nseyo & Santiago-Lastra, 2017). Unfortunately, once diagnosed, DBD is typically irreversible, and treatment outcomes are often unsatisfactory (Bree & Santiago-Lastra, 2020). Therefore, early intervention is crucial for managing DBD and improving patient outcomes.

Currently, urodynamic tests, post-void residual urine (PVR) measurement, and symptom assessment scales are the primary methods for diagnosing and evaluating DBD. Guidelines recommend these three methods as tools for assessing DBD risk (Guo et al., 2020; Ginsberg et al., 2021). However, several challenges exist in clinical practice. Both urodynamic tests and PVR measurement rely on specialized medical equipment, and urodynamic tests are invasive procedures, limiting their widespread use, particularly in primary healthcare settings. Additionally, while symptom assessment scales are easy to use, their objectivity and accuracy are significantly influenced by subjective factors, posing further limitations. Therefore, there is an urgent need to develop a convenient, objective, and user-friendly tool for predicting DBD risk.

However, there are limited predictive tools for assessing the risk of DBD in T2DM patients. In response to this gap, the present study comprehensively and systematically collected relevant data from T2DM patients using a retrospective study design. For the first time, general clinical data, renal function, hepatic function, lipid profiles, immune function, and urine and blood indices were all included in the analysis. This approach not only broadened the scope of information but also enhanced the accuracy of the predictive model. The findings aim to provide an effective tool for identifying high-risk groups for DBD, thereby aiding in the prevention and reduction of DBD incidence.

Materials and Methods

Study population

This retrospective study was conducted at two comprehensive hospitals in Shenzhen, China, utilizing electronic medical records from Southern Medical University Shenzhen Hospital between January 2020 and August 2023 to identify T2DM cases. These records formed the training set, while the validation set was composed of data from patients treated at Shenzhen Third People’s Hospital from March 2022 to October 2023.

Inclusion criteria: (1) Diagnosed with T2DM (ICD-10 code E11.900); (2) comprehensive medical records documenting both bladder symptoms and T2DM-related manifestations.

Exclusion criteria: (1) Duration of diabetes <5 years; (2) neuropathy due to non-diabetic causes (e.g., spinal cord injury, multiple sclerosis (MS), stroke, spina bifida, Parkinson’s disease); (3) urinary tract infection (UTI) within the past month; (4) prostate-related conditions (e.g., benign prostatic hyperplasia (BPH), prostate cancer, history of prostate surgery); (5) pelvic conditions (e.g., pelvic organ prolapse, history of pelvic surgery); (6) acute metabolic complications of diabetes (e.g., diabetic ketoacidosis, hyperosmolar hyperglycemic state); (7) severe dysfunction of the heart, liver, lungs, or kidneys. Only data from the initial admission were included. The exclusion process is illustrated in Fig. 1.

Results

Development of the predictive model

In a multivariate logistic regression analysis, HbA1c, PCP-2h, DPN, TCO₂, PAB, T-Bil, I-Bil, IgE, URBC, UI and UR were identified as independent predictive factors for DBD, and a nomogram was constructed. The method for using the nomogram is provided in Appendix 1. Detailed results can be found in Table 4 and Fig. 2. Notably, T-Bil emerged as a protective factor against DBD.

Table 4:

Multivariate logistic regression analysis of DBD.

Variables	β	OR (95% CI)	Z value	P-value
(Intercept)	−8.385	0.000 [0.000–0.001]	−9.485	<0.001*
HbA1c	0.097	1.101 [1.012–1.200]	2.232	0.026*
PCP-2h	0.089	1.093 [1.005–1.188]	2.074	0.038*
DPN	1.845	6.326 [3.861–10.651]	7.14	<0.001*
UMA-24h	0.000	1.093 [1.005–1.188]	1.485	0.137
TCO2	0.031	1.031 [1.013–1.070]	2.506	0.012*
PAB	0.011	1.012 [1.009–1.014]	11.29	<0.001*
T-Bil	−0.082	0.921 [0.844–0.985]	−2.167	0.03*
I-Bil	0.196	1.217 [1.100–1.376]	3.52	<0.001*
IgA	0.119	1.127 [0.971–1.310]	1.563	0.118
IgE	0.002	1.002 [1.001–1.002]	5.411	<0.001*
URBC	0.004	1.004 [1.001–1.008]	2.322	0.02*
UI	0.714	2.042 [1.330–3.157]	3.243	0.001*
UR	1.510	4.529 [2.869–7.263]	6.388	<0.001*

DOI: 10.7717/peerj.18872/table-4

Notes:

β is the regression coefficient.

OR, odds ratio; CI, confidence interval; HbA1c, Glycated Hemoglobin; DPN, Diabetic Peripheral Neuropathy; PCP-2h, Postprandial C-peptide 2-hour; UMA-24h, Urinary Microalbumin 24-hour; TCO₂, Total Carbon Dioxide; PAB, Prealbumin; T-Bil, Total Bilirubin; I-Bil, Indirect Bilirubin; URBC, Urinary Red Blood Cells; UI, Urinary Incontinence; UR, Urinary Retention; the asterisk (*) indicates that the result is statistically significant (P < 0.05).

Model performance

The discriminative ability of the predictive model was assessed using the ROC curve, which was analyzed for both the complete model and the top four individual factors ranked by AUC value. Detailed results are presented in Table 5, and the ROC curves are depicted in Fig. 3. The analysis yielded an AUC value of 0.897 (95% CI [0.874–0.920]) with an optimal cutoff point of 0.415, demonstrating a sensitivity of 0.832 and a specificity of 0.826. These findings underscore the model’s robust predictive ability for identifying the occurrence of DBD. Additionally, the Hosmer-Lemeshow test (P = 0.319) indicated a good model fit. The calibration curve (P = 0.782) showed a close alignment between the actual and predicted values, demonstrating that the model’s predictions of DBD risk in T2DM patients closely correspond to the observed risk, as illustrated in Fig. 4.

Table 5:

ROC curve analysis of factors and the predictive model with DBD.

Variable(s)	AUC	Specificity	Sensitivity	95% CI
Model	0.897	0.826	0.832	[0.874–0.920]
PAB	0.797	0.804	0.811	[0.761–0.832]
UR	0.658	0.797	0.519	[0.623–0.693]
DPN	0.640	0.445	0.835	[0.608–0.672]
URBC	0.618	0.409	0.786	[0.561–0.643]
TCO2	0.575	0.820	0.298	[0.532–0.618]
I-Bil	0.574	0.584	0.565	[0.531–0.617]
HbA1c	0.569	0.815	0.326	[0.526–0.613]
IgE	0.567	0.639	0.516	[0.523–0.611]
PCP-2h	0.562	0.728	0.389	[0.519–0.606]
UI	0.566	0.669	0.463	[0.530–0.602]
T-Bil	0.549	0.600	0.505	[0.507–0.592]

DOI: 10.7717/peerj.18872/table-5

Note:

AUC, area under the curve; CI, confidence interval; PAB, Prealbumin; UR, Urinary Retention; DPN, Diabetic Peripheral Neuropathy; URBC, Urinary Red Blood Cells; TCO₂, Total Carbon Dioxide; I-Bil, Indirect Bilirubin; HbA1c, Glycated Hemoglobin; PCP-2h, Postprandial C-peptide 2-hour; UI, Urinary Incontinence; T-Bil, Total Bilirubin.

Validation of the predictive model

In this study, we validated the model using data from both the training and validation sets. In the training set, The predictive model underwent internally validated by Bootstrap resampling 1,000 times, ensuring its stability (Accuracy = 0.810, Kappa = 0.600). The AUC remained at 0.896 (95% CI [0.873–0.919]). In the validation set, the AUC value was 0.862 (95% CI [0.816–0.908]), with an optimal cutoff of 0.493, a sensitivity of 0.710, and a specificity of 0.894. These findings suggest that the nomogram model effectively distinguishes between the risk of DBD and Non-DBD occurrence during early screening.

Calibration curves were used to assess the degree of model fit and prediction accuracy. The calibration curve represents a scatter plot of the actual probability of occurrence vs. the predicted probability (Chen et al., 2024). The closer the calibration curve is to a fitted straight line, the better the expected value matches the measured value, indicating greater accuracy of the model (Chen et al., 2024). The results show that the calibration curves fit well with the actual curves, suggesting that the predicted probability of DBD aligns with the exact probability for both data sets (see Figs. 5A, 5B).

The clinical utility of the nomogram

Figures 6A, 6B illustrate the clinical utility and net benefit of the DBD predictive model across different high-risk thresholds. In the training set (Fig. 6A), the decision curve analysis (DCA) demonstrates that the model provides a net benefit within a threshold probability range of 3% to 73%, indicating its strong predictive performance over a wide range of clinical scenarios. In the validation set (Fig. 6B), the DCA results reveal that the model offers a net benefit within a threshold probability range of 4% to 71%. This range is slightly narrower than that observed in the training set, reflecting the model’s consistent yet slightly reduced performance in the external validation cohort. Overall, the DCA results highlight the robust clinical applicability of the DBD predictive model in identifying high-risk patients, with meaningful net benefits observed in both the training and validation sets.

Figures 7A, 7B provides a comprehensive assessment of the clinical efficacy of the DBD risk prediction nomogram across both training and validation sets. In the training set (Fig. 7A), when the threshold probability exceeds 79%, the subgroup identified as high-risk DBD by the nomogram closely corresponds to the actual high-risk DBD population. This alignment demonstrates the nomogram’s strong predictive accuracy and clinical relevance in this setting.

Similarly, in the validation set (Fig. 7B), the nomogram exhibits robust performance. Threshold probabilities exceeding 80% show a strong association between the predicted high-risk group and the actual high-risk DBD population.

Discussion

In this study, a risk prediction model for DBD was developed and internally validated as well as externally validated. Through multivariate analysis, several key variables were included in the final model: HbA1c, PCP-2h, DPN, TCO₂, PAB, T-Bil, I-Bil, IgE, URBC, UI and UR. The performance and clinical relevance of the prediction model were comprehensively evaluated using various analytical methods. The results demonstrated that the model exhibits good efficacy and clinical applicability in predicting DBD.

Surprisingly, this study found that 78.25% of patients with DBD did not have DN, and the occurrence of DBD was not significantly associated with the presence of DN (P = 0.789). This suggests that DBD may be an independent complication of diabetes rather than an adjunct manifestation of DN, revealing the multifaceted impact of diabetes on the urinary system. Previous studies have indicated that DN typically results from diabetes-induced damage to the renal microvasculature and renal tubules (Yu & Bonventre, 2018), whereas DBD is likely primarily caused by autonomic neuropathy. Diabetic neuropathy, even without renal impairment, can directly affect bladder nerve control, leading to bladder dysfunction (Daneshgari et al., 2009). This finding presents new challenges for clinical management: traditionally, DN and DBD have been considered linked complications, with bladder dysfunction often accompanying renal impairment. However, this study demonstrates that DBD can occur independently of DN, necessitating the inclusion of DBD screening and early intervention in the management of diabetes patients, even in the absence of DN symptoms.

The results of this study indicate that HbA1c (OR: 1.101, 95% CI [1.012–1.200]) and DPN (OR: 6.326, 95% CI [3.861–10.651]) play crucial roles in predicting DBD. HbA1c, an important indicator of long-term glycemic control, reflects chronic hyperglycemia in T2DM patients. Elevated HbA1c levels promote the accumulation of advanced glycation end products, which impair the structure and function of microvasculature, leading to microvascular complications (Khalid, Petroianu & Adem, 2022). Studies have shown that microvascular complications not only affect the blood supply and oxygenation of nerves but ultimately result in nerve fiber degeneration and apoptosis, triggering DPN (Tesfaye & Selvarajah, 2012). Notably, DPN extends beyond peripheral nervous system damage to affect the autonomic nervous system, damaging sympathetic and parasympathetic nerves. This directly impairs bladder sensory and motor nerve functions, leading to reduced bladder sensation and disrupted voiding reflexes (Feldman et al., 2019). Therefore, effective glycemic control and early intervention for neuropathy are paramount.

Univariate analysis in this study revealed that PCP-2h levels in DBD patients were significantly lower than those in Non-DBD patients, similar to previous research findings (Saisho, 2016). Elevated PCP-2h levels indicate good responsiveness of pancreatic β-cells, helping to reduce postprandial blood glucose levels and potentially lowering the risk of DBD (Tai et al., 2016; Saisho, 2016). However, surprisingly, multivariate analysis in our study identified PCP-2h levels (OR: 1.093, 95% CI [1.005–1.188]) as a risk factor for the occurrence of DBD. To gain deeper insight into this phenomenon, a comparative analysis was conducted between significant variables from the univariate analysis and PCP-2h levels. The results revealed a significant confounding effect of HbA1c on PCP-2h levels. This finding indicates that while PCP-2h levels can reflect pancreatic β-cell functionality to some extent (Saisho, 2016), their relationship with DBD is neither direct nor singular but rather influenced by the complex state of glycemic control in patients. This suggests a complex interaction between PCP-2h levels, HbA1c, and the development of DBD, which necessitates further investigation in future studies.

This study revealed a significant finding: higher TCO₂ levels are significantly associated with an increased risk of DBD (OR: 1.031, 95% CI [1.013–1.070]). This suggests that acid-base balance may play a crucial role in the pathogenesis of DBD. In patients with T2DM, persistent hyperglycemia due to insulin resistance and insufficient insulin secretion can lead to metabolic acidosis (Farwell & Taylor, 2008). This condition may increase the production of free radicals, resulting in oxidative stress. Oxidative stress not only directly damages cell membranes, DNA, and intracellular proteins (Powell & Gehring, 2023), but it may also harm nerve cells and bladder wall cells, thereby impairing the normal functions of nerves and the bladder (Xu et al., 2022). This finding provides new perspectives and potential intervention targets for the prevention and treatment of DBD.

Our results emphasize the importance of liver function indicators in assessing the risk of DBD. The findings show that elevated levels of PAB (OR = 1.012, 95% CI [1.009–1.014]) and I-Bil (OR = 1.217, 95% CI [1.100–1.376]) are associated with the occurrence of DBD. This may suggest a potential role of liver dysfunction in the pathogenesis of DBD, possibly leading to metabolic disorders, increased oxidative stress, and enhanced inflammatory responses (Michael et al., 2000; Szasz et al., 2016). These factors collectively exert adverse effects on cells and tissues (Powell & Gehring, 2023), particularly affecting the normal functions of the nervous and urinary systems. Additionally, an interesting observation was made in this study: elevated T-Bil levels (OR = 0.921, 95% CI [0.844–0.985]) are associated with a reduced risk of DBD. This finding contradicts the notion that I-Bil is a risk factor for DBD but may reflect the complex physiological roles of bilirubin. The observed phenomenon may be attributed to the natural antioxidant properties of bilirubin, which help alleviate diabetes-related oxidative stress (Vítek, 2012), potentially reducing the risk of microvascular damage and neuropathy (Tesfaye & Selvarajah, 2012; Khalid, Petroianu & Adem, 2022).

It is noteworthy that elevated IgE levels play a potential role in predicting DBD (OR = 1.002, 95% CI [1.001–1.002]). The increase in IgE may reflect immune system activation and disruption of mucosal barriers under prolonged hyperglycemic conditions, a phenomenon particularly common in patients with T2DM. This immune response could facilitate pathogen invasion, triggering localized immune reactions that cause damage to bladder tissues (Velloso, Eizirik & Cnop, 2013; Khalid, Petroianu & Adem, 2022). Moreover, elevated IgE levels may exacerbate bladder dysfunction by promoting inflammation, which contributes to increased damage and fibrosis in the bladder smooth muscle (Wang et al., 2016; Borsodi et al., 2023). These findings provide a new perspective, suggesting that immune regulation may play a more significant role in the pathogenesis of DBD, warranting further exploration in future research.

This study is the first to reveal that elevated URBC count (OR = 1.004, 95% CI [1.001–1.008]) is significantly associated with an increased risk of DBD in patients with T2DM. High URBC counts reflect microvascular damage or an inflammatory state within the urinary system. In T2DM patients, chronic hyperglycemia is a key factor leading to endothelial damage and microvascular complications (Stehouwer, 2018; Khalid, Petroianu & Adem, 2022). Such microvascular damage can result in inadequate blood supply to the bladder wall, impairing bladder contraction and voiding capability (Wang et al., 2023). Furthermore, a high-glucose environment promotes the increase of reactive oxygen species, triggering oxidative stress and cellular damage (Khalid, Petroianu & Adem, 2022). This process can induce inflammatory responses, exacerbating urinary tract irritation and contributing to further urinary system damage. These combined factors may ultimately lead to the development of DBD.

This study highlights the critical role of UI (OR: 2.042, 95% CI [1.330–3.157]) and UR (OR: 4.529, 95% CI [2.869–7.263]) in DBD. UI, a common symptom of DBD, aligns with previous studies (Daneshgari et al., 2009), suggesting that diabetes-induced neuropathy may impair bladder storage function, thus leading to UI. Furthermore, the occurrence of UR is closely linked to diabetic neuropathy, particularly in cases involving autonomic nerve damage, which impairs bladder emptying and exacerbates bladder dysfunction (Feldman et al., 2019). Our findings confirm UI and UR as independent risk factors for DBD, indicating multiple dysfunctions in bladder storage and emptying in diabetic patients. These findings not only deepen our understanding of the mechanisms underlying bladder symptoms in diabetes, but also provide crucial theoretical grounds for clinical interventions and patient management, emphasizing the importance of early bladder function assessment and intervention in diabetic patients.

The strength of this study lies in the use of a large-scale retrospective data system to analyze risk factors associated with DBD in patients with T2DM, and the successful development of a predictive model. First, the model integrates not only traditional diabetes-related indicators but also multidimensional data such as liver and kidney function markers and immune biomarkers, thereby improving predictive accuracy and clinical applicability. Second, the model demonstrated strong predictive power and calibration through both internal and external validation, Further supporting its potential clinical utility. Moreover, this study not only analyzed traditional hyperglycemia markers but also explored the potential role of novel biomarkers in the pathogenesis of DBD, providing new directions and intervention targets for future research. Finally, the study incorporated the temporal sequence between T2DM and the onset of DBD, enhancing the interpretability of the results.

We also acknowledge the limitations of this study. First, as a retrospective study, it relies on the detailed records of medical case files, with the presence of predictive factors and prognostic outcomes dependent on these historical data. However, this reliance may introduce inherent information bias. Second, convenience sampling was used to extract data from the electronic medical records of two hospitals. Although this method allows for rapid sample collection, it may introduce selection bias and fail to comprehensively represent the characteristics of all T2DM patients, potentially leading to the underestimation or overestimation of certain groups. Additionally, the study data were sourced from only two general tertiary hospitals in Shenzhen. While these hospitals have a diverse patient population, the geographic limitation to Shenzhen restricts the generalizability of the results. To improve the external validity of the findings, future studies should adopt more stringent random sampling methods, such as stratified random sampling, and expand the data source to include patients from different regions and healthcare levels, covering a broader spectrum of T2DM patients. Furthermore, future research could consider a multi-center, prospective design to further explore causal relationships, analyze potential influencing factors, and validate the generalizability and stability of the model across different patient populations, thereby providing more reliable scientific evidence.

Conclusion

This study developed and both internally and externally validated a risk prediction model for DBD in patients with T2DM, and constructed a nomogram. The model demonstrated excellent discrimination, calibration, and clinical validity, providing healthcare professionals with an effective tool to reduce the incidence of DBD and improve patients’ quality of life.

Supplemental Information