RESUMO
In view of the alarming increase in the burden of diabetes mellitus (DM) today, a rising number of patients with diabetic kidney disease (DKD) is forecasted. Current DKD predictive models often lack reliable biomarkers and perform poorly. In this regard, serum myoglobin (Mb) identified by machine learning (ML) may become a potential DKD indicator. We aimed to elucidate the significance of serum Mb in the pathogenesis of DKD. Electronic health record data from a total of 728 hospitalized patients with DM (286 DKD vs. 442 non-DKD) were used. We developed DKD ML models incorporating serum Mb and metabolic syndrome (MetS) components (insulin resistance and ß-cell function, glucose, lipid) while using SHapley Additive exPlanation (SHAP) to interpret features. Restricted cubic spline (RCS) models were applied to evaluate the relationship between serum Mb and DKD. Serum Mb-mediated renal function impairment induced by MetS components was verified by causal mediation effect analysis. The area under the receiver operating characteristic curve of the DKD machine learning models incorporating serum Mb and MetS components reached 0.85. Feature importance analysis and SHAP showed that serum Mb and MetS components were important features. Further RCS models of DKD showed that the odds ratio was greater than 1 when serum Mb was > 80. Serum Mb showed a significant indirect effect in renal function impairment when using MetS components such as HOMA-IR, HGI and HDL-C/TC as a reason. Moderately elevated serum Mb is associated with the risk of DKD. Serum Mb may mediate MetS component-caused renal function impairment.
