Evaluation of a tumor marker gastrin-releasing peptide precursor in the patients with kidney injuries.

Gastrin-releasing peptide precursor (ProGRP) is a bioactive precursor of GRP and might play an important role as an emerging tumor marker in early cancer diagnosis. It might also be abnormal in the nonmalignant disease and renal function abnormalities. The present study was undertaken to investigate the changes of ProGRP levels in patients with kidney injuries, especially with chronic kidney disease (CKD), determine the upper reference intervals and clinical diagnostic value of ProGRP in CKD, and thus help oncologists in interpreting ProGRP levels and making clinical judgments of malignances. 676 individuals were enrolled in this cross-sectional study and divided into five groups: healthy control (n=194), CKD (n=272), nephrotic syndrome (NS) (n=137), antineutrophil cytoplasmic antibody (ANCA)-associated vasculitis (AAV) (n=41), and urinary tract infection (UTI) (n=32). A total of 27 features including age, gender, and 25 laboratory markers were analyzed. Machine learning algorithms were built for the diagnostic models of CKD. Statistical analysis was performed by R software. It was shown that serum ProGRP level in CKD was significantly higher than that in healthy controls, UTI and NS ( < 0.01). The upper reference limit of ProGRP was 188.42 pg/ml for CKD, 245.40 pg/ml for CKD IV-V, and 97.25 pg/ml for NS. Compared with the healthy control, the level of serum ProGRP in CKD stages II, III, IV-V was significantly increased and elevated progressively with CKD grade ( < 0.01). Random Forest (RF) model works best among 4 building machine learning algorithms. 5 vital indicators, ProGRP, estimated glomerular filtration rate (eGFR), urea, albumin (ALB), and direct bilirubin (DBIL), were selected to establish RF model for diagnosing CKD with an area under the curve (AUC) of 0.96 (95% confidence interval [CI]: 0.94-0.97) and high sensitivity (0.89) and specificity (0.92). This study demonstrates that the level of ProGRP in patients with CKD, nephrotic syndrome or AAV, was significantly higher than that in the healthy population. The machine learning model of ProGRP with DBIL, eGFR, ALB, and urea, could provide good clinical value for CKD evaluation.