Abstract
Based on the ability of our machine learning algorithm to establish an effective risk score, we propose that comprehensive assessments of SBP, DBP, FBG, TG, gender, age, and smoking should be included in the screening process for MAU.
Introduction
Microalbuminuria (MAU) serves as a critical early indicator of systemic endothelial damage, strongly associated with chronic kidney disease (CKD), stroke, and myocardial infarction. While the 24-hour urine collection test remains the "gold standard" for detection, its practical application in large-scale public health screenings is limited by its cumbersome nature. Consequently, there is a pressing need for a predictive model that utilizes common clinical data to identify high-risk individuals. This study aims to develop a simplified, data-driven risk model specifically tailored for the Chinese population to facilitate early intervention.
Methods
This cross-sectional study involved 3,294 participants from Ningde City, Fujian Province, China, ranging in age from 16 to 93 years. The cohort included 1,294 men and 2,000 women, excluding individuals with Type-1 diabetes, severe kidney disease, or pregnancy.
The research workflow integrated traditional statistics with machine learning techniques. Data preprocessing involved multiple imputation via the mice package in R to address missing values. A two-stage logistic regression approach was employed: an initial model identified potential predictors ($P < 0.1$), which were subsequently refined in a second stage to isolate the most significant features ($P < 0.05$). The dataset was split into training and validation sets (7:3 ratio).
Results and Key Insights
The modeling process identified seven primary clinical predictors for MAU: Systolic Blood Pressure (SBP), Diastolic Blood Pressure (DBP), Fasting Blood Glucose (FBG), Triglycerides (TG), sex, age, and smoking. Notably, elevated fasting blood glucose emerged as a significant predictor even in the absence of a formal diabetes diagnosis, suggesting the model's sensitivity to pre-diabetic states. Demographic factors such as female gender and advanced age were also found to significantly increase MAU prevalence.
The model demonstrated robust performance, particularly in its ability to minimize false positives. In the validation set, it achieved a specificity of 0.9 and an overall accuracy of 0.63. The Positive Predictive Value (PPV) was 0.55, and the Negative Predictive Value (NPV) was 0.65.
Calibration curves indicated a strong alignment between predicted and actual probabilities, confirming the reliability of the risk score. For instance, the data revealed that every 10-year increase in age correlates with a roughly 9% increase in the odds of significant triglyceride elevation and subsequent MAU risk.
Discussion and Limitations
By focusing on variables routinely collected during standard physical examinations, this model provides a cost-effective tool for initial risk assessment without requiring specialized laboratory tests. However, several limitations exist. The study's single-center design in southeast China necessitates further validation across diverse geographic and ethnic populations. Additionally, the cross-sectional nature of the data establishes correlation rather than causation. Future research should involve longitudinal studies and explore more complex machine learning architectures to potentially enhance predictive accuracy while maintaining clinical interpretability.
Summary updated