Background and objectives SERPINA4 has been identified as a potential diagnostic biomarker for diabetic nephropathy (DN) in our previous research. This study aims to develop electrochemiluminescence immunoassay (ECLIA) methods for the detection of SERPINA4 and to establish a diagnostic model that incorporates additional indicators for DN.Materials and methods Antibodies utilized in the ECLIA for the detection of SERPINA4 were labelled with ruthenium and biotin, respectively. The reliability of ECLIA was evaluated based on its linear range, precision, and hook effect. A total of 28 indicators were collected from 98 patients, including SERPINA4/UCr, diabetic retinopathy (DR), and duration of diabetes mellitus. A diagnostic model was developed employing Random Forest, Support Vector Machine (SVM), and Naive Bayes algorithms. The performance of the model was assessed using metrics such as area under the curve (AUC), precision, recall, and F1 score; ultimately selecting the best-performing model for final diagnosis.Result The ECLIA method established in this study for urinary SERPINA4 demonstrates a linearity range from 7.5 ng/mL to 16,000 ng/mL, with within-run precision (CV%) values of 0.25% and 3.78%. The diagnostic model developed using random forest exhibits optimal performance, achieving an AUC of 0.89, accuracy of 90%, sensitivity of 100%, and specificity of 70%. The top five variables ranked by importance are serum creatinine, microalbumin, SERPINA4/UCr ratio, systolic blood pressure, and total urine protein.Conclusion A method for the detection of urinary SERPINA4 using ECLIA has been successfully established. The combination of SERPINA4/UCr with other clinical indicators demonstrated strong performance in the diagnostic model developed through the random forest algorithm.