📄 medicine

Association Between Urinary Cadmium Levels and Tubular Injury Biomarkers with Development of a Machine Learning-Based Predictive Model in Non-Dialysis Patients with Chronic Kidney Disease: a retrospective study

This retrospective study of 278 non-dialysis chronic kidney disease patients developed and validated a machine learning-based nomogram incorporating urinary tubular injury biomarkers (NAG, RBP), diabetes status, and GFR to predict high urinary cadmium exposure, revealing significant non-linear associations between cadmium and specific biomarkers while highlighting diabetes as a susceptibility factor.

Original authors: xi chen, tian tang, yuxue liu, rong zhou, xuejing guan

Published 2026-09-25

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: xi chen, tian tang, yuxue liu, rong zhou, xuejing guan

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The kidneys are the body's silent filters, working tirelessly to clean the blood and remove waste. When these organs are damaged, the first signs are often subtle and hidden deep within the microscopic structures that do the actual work. Among the many threats to kidney health, cadmium stands out as a particularly stubborn enemy. This heavy metal, found in soil, water, and some foods, is not easily broken down by the body. Once absorbed, it travels to the kidneys and settles in the tiny tubes responsible for reabsorbing useful nutrients. There, it can remain for decades, slowly wearing down the cells that keep the filtration system running. For people who already have chronic kidney disease, this hidden burden can be especially dangerous, accelerating damage that might otherwise go unnoticed until it is too late.

Scientists have long known that cadmium harms the kidneys, but they have struggled to find a simple way to identify which patients are at the highest risk before irreversible damage occurs. Traditional blood tests often miss early injury because the kidneys can compensate for a long time before showing clear signs of failure. This gap in knowledge leaves many patients vulnerable, exposed to environmental toxins without a clear warning system. Researchers in Shanghai recently set out to bridge this gap by looking for a new kind of early warning signal. They focused on specific proteins and enzymes found in urine that act as messengers, leaking out only when the kidney's inner tubes are under stress. By combining these biological clues with modern data analysis, they aimed to create a practical tool that could help doctors spot high-risk patients earlier and intervene before the damage becomes permanent.

The study, conducted at Yangpu Hospital, looked back at the medical records of 278 adults with chronic kidney disease who were not yet on dialysis. The researchers divided these patients into two groups based on their levels of cadmium in the urine: those with low exposure and those with high exposure, defined as having two micrograms of cadmium per gram of creatinine or more. They then compared the two groups to see how their bodies were reacting to the metal. The results were striking. Patients with high cadmium levels showed significantly higher amounts of specific injury markers in their urine, including retinol-binding protein and an enzyme called N-acetyl-beta-D-glucosaminidase. These markers are like tiny canary-in-the-coal-mine signals, indicating that the kidney's reabsorption tubes were struggling to do their job. Interestingly, the researchers also found that patients with diabetes were much more likely to be in the high-exposure group, suggesting that having diabetes might make the kidneys more vulnerable to the toxic effects of cadmium.

To make sense of these complex relationships, the team used advanced computer modeling techniques to sift through the data and find the most reliable predictors. They discovered that the relationship between cadmium exposure and kidney injury was not always a straight line. For one marker, retinol-binding protein, the risk increased steadily as the levels rose, acting like a reliable gauge. For another, the enzyme NAG, the risk jumped sharply once a certain threshold was crossed, then leveled off, suggesting that the kidney can handle a small amount of stress before a tipping point is reached. By combining these biological markers with information about whether a patient had diabetes and how well their kidneys were filtering blood, the researchers built a visual prediction tool. This tool, known as a nomogram, allows a doctor to input a patient's specific data and receive a calculated probability of high cadmium exposure.

The model proved to be quite accurate, correctly distinguishing between high and low exposure groups in most cases. It showed that the combination of elevated NAG levels, the presence of diabetes, and specific levels of retinol-binding protein provided the clearest picture of who was at risk. The researchers noted that while the tool is promising, it is not yet a final solution for every clinic. It requires further testing in larger groups of people to confirm its reliability across different populations. However, the findings offer a compelling new direction for care. They suggest that for patients with chronic kidney disease, especially those with diabetes, checking for these specific urine markers could reveal a hidden threat that standard tests miss. By identifying these high-risk individuals sooner, doctors might be able to take steps to reduce exposure and protect the remaining kidney function, turning a silent, slow-moving danger into a manageable condition.

Technical Summary: Association Between Urinary Cadmium Levels and Tubular Injury Biomarkers in Non-Dialysis CKD Patients

Problem Statement
Cadmium (Cd) is a pervasive nephrotoxic environmental pollutant that accumulates in renal proximal tubules with a biological half-life of 10–30 years. While low-level exposure is linked to chronic kidney disease (CKD), early detection remains challenging because cadmium-induced damage is clinically silent until irreversible interstitial fibrosis occurs. Non-dialysis CKD patients are particularly vulnerable due to reduced renal functional reserve. Currently, there is a lack of individualized, clinically applicable tools to assess the risk of elevated urinary cadmium exposure in this specific population. Existing studies often rely on single-marker analyses, fail to account for non-linear dose–response relationships, and lack predictive models for stratifying high-risk individuals.

Methodology
This retrospective, single-center study enrolled 278 patients with stage 1–5 non-dialysis CKD from Yangpu Hospital (Tongji University School of Medicine) between January 2019 and October 2022. Patients were stratified into low-exposure ( $n=180$ ) and high-exposure ( $n=98$ ) groups based on a urinary cadmium cut-off of 2 µg/g·Cr.

The study employed a multi-stage analytical framework integrating machine learning and advanced statistical modeling:

Data Collection: Baseline demographics, clinical history (including diabetes and hypertension), renal function (eGFR, Cystatin C), and urinary tubular injury biomarkers ( $\beta_2$ -microglobulin, Retinol-Binding Protein [RBP], N-acetyl- $\beta$ -D-glucosaminidase [NAG], $\alpha_1$ -microglobulin, and microalbumin) were collected.
Correlation and Dose-Response Analysis: Spearman rank correlation assessed variable associations. Restricted Cubic Spline (RCS) regression models were utilized to evaluate non-linear dose–response relationships between continuous variables and high cadmium exposure.
Variable Selection: LASSO (Least Absolute Shrinkage and Selection Operator) logistic regression with 10-fold cross-validation was applied to screen for predictive variables and handle multicollinearity.
Model Construction: A multivariable logistic regression model was built using variables selected by LASSO. A visualized clinical risk nomogram was constructed based on the regression coefficients.
Validation and Evaluation: Model performance was assessed using the Area Under the Curve (AUC) for discrimination, Hosmer–Lemeshow tests and calibration plots for calibration, and Decision Curve Analysis (DCA) for clinical utility. Internal validation was performed using the bootstrap method with 1,000 resamples.

Key Results

Demographic and Clinical Differences: The high-exposure group exhibited a significantly higher prevalence of Type 2 Diabetes Mellitus (T2DM) compared to the low-exposure group (42.86% vs. 28.33%, $P=0.02$ ). No significant differences were found in age, sex, smoking index, or GFR between groups.
Biomarker Elevation: Urinary levels of $\beta_2$ -MG, RBP, NAG, and microalbumin were significantly elevated in the high-exposure group ( $P < 0.001$ ). $\alpha_1$ -MG showed no significant difference.
Dose-Response Patterns (RCS):
- NAG: Showed a significant positive, non-linear association with high cadmium exposure, characterized by a saturation effect ( $P_{nonlinear} < 0.001$ ).
- RBP: Exhibited a significant, approximately linear positive association ( $P_{nonlinear} = 0.069$ ).
- $\beta_2$ -MG: Displayed a J-shaped non-linear association ( $P_{nonlinear} = 0.038$ ).
Predictive Model: LASSO regression identified four independent predictors: T2DM, RBP, GFR, and NAG. The final multivariable logistic model confirmed these as independent predictors:
- NAG positivity (OR = 3.88, 95% CI 1.99–7.56)
- T2DM (OR = 1.81, 95% CI 1.00–3.28)
- RBP (OR = 1.03, 95% CI 1.01–1.06)
- GFR (OR = 1.01, 95% CI 1.00–1.02)
Model Performance: The constructed nomogram achieved an AUC of 0.77 (95% CI 0.71–0.83). Calibration was excellent (Hosmer–Lemeshow $P = 0.989$ , Brier score 0.18). DCA indicated a positive net benefit across threshold probabilities of 0.05 to 0.82.

Significance and Claims
The authors claim that this study provides the first integrated analysis using machine learning to systematically compare multiple tubular biomarkers for predicting high urinary cadmium exposure specifically in the non-dialysis CKD population. Key contributions include:

Identification of Optimal Indicators: RBP and NAG are identified as effective indicators for identifying high-risk cadmium exposure, with distinct dose–response patterns (linear for RBP, non-linear/saturating for NAG).
Susceptible Subgroup: The study highlights patients with comorbid diabetes as a potentially susceptible subgroup to cadmium nephrotoxicity, warranting closer monitoring.
Clinical Tool: The developed nomogram demonstrates satisfactory calibration and moderate discrimination, serving as a practical reference for early screening and risk stratification of cadmium-related tubular injury.
Methodological Advancement: The combination of LASSO for variable selection and RCS for non-linear characterization offers a robust strategy for environmental epidemiology, overcoming limitations of traditional single-marker or linear analyses.

Limitations and Future Directions
The authors explicitly acknowledge several limitations: the cross-sectional design precludes causal inference; the single-center nature limits generalizability; and the model relies on internal validation, lacking external validation in large-scale cohorts. Additionally, the model predicts high urinary cadmium status (an intermediate indicator) rather than hard clinical endpoints like ESRD. The authors conclude that while the model is a useful adjunctive tool, further multicenter prospective studies are required to validate its predictive value and explore whether reducing cadmium exposure can slow CKD progression.

Technical Summary: Association Between Urinary Cadmium Levels and Tubular Injury Biomarkers in Non-Dialysis CKD Patients

More like this