Truace
Subscribe

Application of machine learning prediction models for in-hospital mortality in diabetic sepsis patients: a study based on the MIMIC IV database

Sepsis complicated by diabetes mellitus represents a significant clinical challenge in intensive care units (ICUs), associated with substantial morbidity and mortality. Accurate prognostication of in-hospital mortality in this complex patient population is crucial for optimizing clinical decision making, resource allocation, and patient management strategies. We conducted a retrospective cohort study utilizing data from the Medical Information Mart for Intensive Care (MIMIC-IV) database. The study focused on adu…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-2217· cite & verify
Application of machine learning prediction models for in-hospital mortality in diabetic sepsis patients: a study based on the MIMIC IV database
FreeStyle libre am Oberarm und Auslesegerät-3 by Thirunavukkarasye-Raveendran. CC BY 4.0 · https://creativecommons.org/licenses/by/4.0

In brief

Researchers used the MIMIC-IV intensive care database to study 7716 adult ICU patients diagnosed with both sepsis and diabetes mellitus, of whom 1357 (17.6%) died in hospital. They compared six machine learning algorithms and the SOFA score on how well each discriminated patients who died from those who survived.

Better prognosis for this complex group could support clinical decisions and resource allocation, and SHAP values were used to show which features drove predictions. However, the results come from a retrospective single database, and the authors say multi-center validation and prospective trials are still needed to test generalizability and clinical impact.

Main points

  1. The cohort included 7716 adult ICU patients with concurrent sepsis and diabetes mellitus, drawn from MIMIC-IV.
  2. In-hospital mortality occurred in 1357 patients, or 17.6% of the cohort.
  3. Six algorithms were tested: logistic regression, KNN, random forest, GBDT, XGBoost and LightGBM.
  4. GBDT performed best with an AUC of 0.792 (95% CI: 0.769-0.814), above the SOFA score.
  5. SHAP analysis ranked clinical and demographic predictors to make the model more interpretable.

The gain

A gradient boosting model identified in-hospital mortality risk among ICU patients with sepsis and diabetes more accurately than other algorithms and the SOFA score in a retrospective cohort.

The rundown

The study was a retrospective cohort analysis built on the Medical Information Mart for Intensive Care database. It targeted adults in intensive care who had both sepsis and diabetes, a group the authors describe as a significant clinical challenge with substantial morbidity and mortality.

Model discrimination was measured with the area under the receiver operating characteristic curve. The gradient boosting decision tree model scored highest among the machine learning approaches and also exceeded the conventional SOFA score used as a comparison.

To address interpretability, the team applied SHAP values to explain the relative importance of predictor variables. The authors suggest this could facilitate adoption of such tools in clinical practice, though they have not tested real-world implementation.

Sources

  1. Peer-reviewedBMC Medical Informatics and Decision Making2026-10-22

The debate