Article
Clinical expert-assigned ground truth for sepsis diagnosis in the Intensive Care Unit
Search Medline for
Authors
-
Franziska Holke
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Bianka Hahn
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Holger A. Lindner
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany; Mannheim Institute of Innate Immunoscience (MI3), Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Shigehiko Schamoni
- Department of Computational Linguistics, Heidelberg University, Heidelberg, Germany; Interdisciplinary Center for Scientific Computing (IWR), Heidelberg University, Heidelberg, Germany
-
Jörg Krebs
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Stephanie Nitsch
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Thomas Friedrich
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Anke Schwarz
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Peter-Tobias Graf
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Manfred Thiel
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
-
Verena Schneider-Lindner
- Department of Anesthesiology and Surgical Intensive Care Medicine, Medical Faculty Mannheim, Heidelberg University, Mannheim, Germany
| Published: | September 6, 2024 |
|---|
Outline
Text
Introduction: Due to its high mortality, sepsis remains one of the most challenging illnesses in intensive care units (ICUs). In the absence of specific biomarkers, multivariable models based on clinical characteristics can assist in early detection of septic patients. Retrospective determination of sepsis onset with consensus definition-based clinical criteria, however, had only fair agreement with expert judgement, i.e. ground truth, in our survey data [1]. Hence, we here employ ground truth for sepsis detection model development in mixed surgical ICU patients from routine data gathered in a 24-hour time span before sepsis onset. The selection of suitable controls poses an additional challenge, which we met by defining a treatment-time matched control group.
Methods: A cohort of non-septic patients on ICU admission was followed-up until sepsis onset, mortality or ICU discharge. We applied a nested-case-control design with risk set sampling to match septic cases with non-septic controls based on their length of stay (1:10 ratio). We used our Ground Truth for Sepsis Questionnaire (GTSQ), which is administered in the anaesthesiologic ICU of the University Medical Centre Mannheim since 2016, to determine the onset of sepsis. In the GTSQ, senior intensivists rate daily whether a non-septic ICU patient transitioned to sepsis. We diagnosed sepsis using 52 candidate variables from the last 24 hours before sepsis onset. These comprised vital signs, interventions, laboratory results, clinical scores (e.g. SAPS II (Simplified Acute Physiology Score)), and three dynamic SIRS (systemic inflammatory response syndrome) descriptors (mean λ, Δ, and C) [2]. We accounted for missing data with multiple imputation by chained equations (MICE) and split the dataset into a training (70%) and test (30%) subset. We selected the most important predictors via LASSO (least absolute shrinkage and selection operator) in the training dataset and applied those with a Cox regression model in the test dataset.
Results: Out of 1418 patients without sepsis on ICU admission, 390 (27.5%) developed sepsis during their ICU stay. Fifteen predictors were selected for the final Cox regression model. Overall, an AUC (Area Under the Curve) of 0.80 with sensitivity 0.79 and specificity 0.70 was achieved. Five predictors (SAPS II, body temperature, C-reactive protein, procalcitonin, and Δ) yielded statistically significant hazard ratios.
Conclusion: The sepsis detection model for mixed surgical ICU patients yielded a good overall performance as indicated by the AUC and balanced sensitivity and specificity. Furthermore, the model is based on 15 routinely available predictors, hence contributing to a straightforward clinical usability after successful external validation. We expect analyses of relevant patient subgroups (e.g. by referring department) to further improve diagnostic accuracy. Finally, we generally advocate valid sepsis labels for advancing early sepsis detection.
The authors declare that they have no competing interests.
The authors declare that a positive ethics committee vote has been obtained.
References
- 1.
- Lindner HA, Schamoni S, Kirschning T, Worm C, Hahn B, Centner FS, Schoettler JJ, Hagmann M, Krebs J, Mangold D, Nitsch S, Riezler S, Thiel M, Schneider-Lindner V. Ground truth labels challenge the validity of sepsis consensus definitions in critical illness. J Transl Med. 2022 Jan 15;20(1):27. DOI: 10.1186/s12967-022-03228-7
- 2.
- Lindner HA, Balaban Ü, Sturm T, Weiß C, Thiel M, Schneider-Lindner V. An algorithm for systemic inflammatory response syndrome criteria-based prediction of sepsis in a polytrauma cohort. Crit Care Med. 2016 Dec; 44(12):2199-2207. DOI: 10.1097/CCM.0000000000001955
