Scholarlink

Background <p>Sepsis-associated acute kidney injury (SA-AKI) is a frequent complication in critically ill patients and remains difficult to identify at an early stage. Conventional indicators such as serum creatinine and severity scores often reflect established organ dysfunction rather than early renal risk. Whether routinely measured laboratory parameters can provide earlier and clinically useful risk information remains unclear.</p> Methods <p>We performed a retrospective cohort study using two large critical care databases (MIMIC-IV and eICU-CRD). Adult patients fulfilling Sepsis-3 criteria were included. AKI was defined according to KDIGO criteria. Clinical and laboratory variables obtained within the first 24&#xa0;h of ICU admission were analyzed. An interpretable machine learning framework was applied to identify key predictors and examine their associations with AKI risk.</p> Results <p>A total of 19,805 patients were included. Urine output, SOFA score, serum creatinine, and serum magnesium were consistently identified as important predictors. The final model achieved an AUC of approximately 0.84 across repeated validation strategies. Serum magnesium showed a non-linear association with AKI risk, with higher predicted risk at lower concentrations and attenuation of the association above approximately 2.0–2.1&#xa0;mg/dL. Multivariable regression analysis showed a consistent direction of association.</p> Conclusions <p>Serum magnesium was consistently associated with SA-AKI risk in this study. The observed non-linear pattern suggests that its relationship with AKI risk may not be adequately captured by linear models. These findings require confirmation in prospective studies.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Serum magnesium levels and early risk of sepsis-associated acute kidney injury: an interpretable machine learning analysis based on the MIMIC-IV and eICU databases

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月5日

查看全文
  • Qingyu Sun,
  • Chenyan Sun,
  • Xiaoxiao Sun,
  • Haomeng Xie,
  • Dong Ji,
  • Yinggang Zheng

摘要

Background

Sepsis-associated acute kidney injury (SA-AKI) is a frequent complication in critically ill patients and remains difficult to identify at an early stage. Conventional indicators such as serum creatinine and severity scores often reflect established organ dysfunction rather than early renal risk. Whether routinely measured laboratory parameters can provide earlier and clinically useful risk information remains unclear.

Methods

We performed a retrospective cohort study using two large critical care databases (MIMIC-IV and eICU-CRD). Adult patients fulfilling Sepsis-3 criteria were included. AKI was defined according to KDIGO criteria. Clinical and laboratory variables obtained within the first 24 h of ICU admission were analyzed. An interpretable machine learning framework was applied to identify key predictors and examine their associations with AKI risk.

Results

A total of 19,805 patients were included. Urine output, SOFA score, serum creatinine, and serum magnesium were consistently identified as important predictors. The final model achieved an AUC of approximately 0.84 across repeated validation strategies. Serum magnesium showed a non-linear association with AKI risk, with higher predicted risk at lower concentrations and attenuation of the association above approximately 2.0–2.1 mg/dL. Multivariable regression analysis showed a consistent direction of association.

Conclusions

Serum magnesium was consistently associated with SA-AKI risk in this study. The observed non-linear pattern suggests that its relationship with AKI risk may not be adequately captured by linear models. These findings require confirmation in prospective studies.