Freely available datasets have become an invaluable tool to propel data-driven research, especially in the field of critical care medicine. However, the number of datasets available is limited. This leads to the repeated reuse of datasets, inherently increasing the risk of selection bias. Additionally, the need arose to validate insights derived from one dataset with another. In 2023, the Salzburg Intensive Care database (SICdb) was introduced. SICdb offers insights in currently 27,386 intensive care admissions from 21,583 patients. It contains cases of general and surgical intensive care from all disciplines. Amongst others SICdb contains information about: diagnosis, therapies (including data on preceding surgeries), scoring, laboratory values, respiratory and vital signals, and configuration data. Data for SICdb (1.0.6) was collected at one single tertiary care institution of the Department of Anesthesiology and Intensive Care Medicine at the Salzburger Landesklinik (SALK) and Paracelsus Medical University (PMU) between 2013 and 2021. This article aims to elucidate on the characteristics of the dataset, the technical implementation, and provides analysis of its strengths and limitations.

Intensive Care Medicine (ICM), titled "Trash in/Trash out? Using Routinely Collected Clinical Data for Data Science in the ICU: Con. " with great interest. However, we believe that in certain aspects, their perspective does not fully reflect the broader potential offered by large-scale intensive care unit (ICU) datasets collected through clinical routine. In the rapidly evolving world of data science, the application of artificial intelligence (AI) promises to revolutionize ICU care. However, the routine clinical data collected in these settings are often criticized for their inherent imperfections, including biases and errors, which some argue reduce model quality and reliability. Critics contend that these imperfections result from the data's primary purpose being clinical documentation, not research. Variations across institutions in how data are collected, processed, and maintained exacerbate these issues, raising concerns about model generalizability. Additionally, models derived from routine data, while showing promise in development phases, are rarely implemented at the bedside [1, 2]. Despite these challenges, routinely collected clinical data offer unparalleled opportunities for advancing ICU research and improving patient outcomes [3]. Routine data imperfections are not merely a challenge; they are an asset. These datasets mirror the complexities and messiness of real-world healthcare settings. AI models trained on routine data are inherently better equipped to navigate real-world complexities, making them more suitable to clinical deployment. If models were trained exclusively on highly curated and sanitized datasets, they risk being

Background/Objectives: Infective endocarditis (IE) often requires surgical intervention, with postoperative acute kidney injury (AKI), posing a significant concern. This retrospective study aimed to investigate AKI incidence, its impact on short-term mortality, and identify modifiable factors in patients with IE scheduled for valve surgery. Methods: This single-center study enrolled 130 consecutive IE patients from 2013 to 2021 undergoing valve surgery. The creatinine levels were monitored pre- and postoperatively, and AKI was defined by Kidney Disease: Improving Global Outcomes (KDIGO) criteria. Patient demographics, comorbidities, procedural details, and complications were recorded. Primary outcomes included AKI incidence; the relevance of creatinine levels for AKI detection; and the association of AKI with 30-, 60-, and 180-day mortality. Modifiable factors contributing to AKI were explored as secondary outcomes. Results: Postoperatively, 35.4% developed AKI. The highest creatinine elevation occurred on the second postoperative day. Best predictive value for AKI was a creatinine level of 1.35 mg/dL on the second day (AUC: 0.901; sensitivity: 0.89, specificity: 0.79). Elevated creatinine levels on the second day were robust predictors for short-term mortality at 30, 60, and 180 days postoperatively (AUC ranging from 0.708 to 0.789). CK-MB levels at 24 h postoperatively and minimum hemoglobin during surgery were identified as independent predictors for AKI in logistic regression. Conclusion s: This study highlights the crucial role of creatinine levels in predicting short-term mortality in surgical IE patients. A specific threshold (1.35 mg/dL) provides a practical marker for risk stratification, offering insights for refining perioperative strategies and optimizing outcomes in this challenging patient population.

Background/Objectives: Infectious Endocarditis often requires surgical intervention, with postoperative acute kidney injury (AKI) posing a significant concern. This retrospective study aimed to investigate AKI incidence, its impact on short-term mortality, and identify modifiable factors in patients with endocarditis scheduled for valve surgery. Methods: This single center study enrolled 130 consecutive endocarditis patients from 2013 to 2021 undergoing valve surgery. Creatinine levels were monitored pre- and postoperatively, and AKI was defined by Kidney Disease: Improving Global Outcomes (KDIGO) criteria. Patient demographics, comorbidities, procedural details, and complications were recorded. Primary outcomes included AKI incidence, the relevance of creatinine levels for AKI detection, and the association of AKI with 30-, 60- and 180-day mortality. Modifiable factors contributing to AKI were explored as secondary outcomes. Results: Postoperatively, 35.4% developed AKI. The highest creatinine elevation occurred on the 2nd postoperative day. Best predictive value for AKI was a creatinine level of 1.35 mg/dl on the 2nd day (AUC: 0.901; sensitivity: 0.89, specificity: 0.79). Elevated creatinine levels on the 2nd day were robust predictors for short-term mortality at 30, 60 and 180 days postoperatively (AUC ranging from 0.708 to 0.789). CK-MB levels at 24 hours postoperatively and minimum hemoglobin during surgery were identified as independent predictors for AKI in logistic regression. Conclusions: This study highlights the crucial role of creatinine levels in predicting short-term mortality in surgical endocarditis patients. A specific threshold (1.35 mg/dl) provides a practical marker for risk stratification, offering insights for refining perioperative strategies and optimizing outcomes in this challenging patient population.

Objectives: EuroSCORE II is widely used to predict perioperative and 30-day mortality in cardiac surgery, yet data on its ability to predict long-term outcomes remain limited. This study investigates whether EuroSCORE II is associated with one-year and long-term mortality in a heterogeneous population undergoing major cardiac surgery with cardiopulmonary bypass. Methods: A retrospective cohort study was conducted including 2179 patients who underwent elective or urgent cardiac surgery with cardiopulmonary bypass between 2017 and 2021 at the University Hospital Salzburg. Data were extracted from the Salzburg Intensive Care database (SICdb) and supplemented with mortality information from Statistik Austria. EuroSCORE II values were compared between survivors and non-survivors. Kaplan–Meier analyses, Cox regression and logistic regression with ROC analysis were performed to evaluate the predictive association of EuroSCORE II with mortality. Results: EuroSCORE II was significantly higher in patients who died within one year and in those who died during a mean follow-up period of 1152.67 ± 521.39 days. Patients who survived at least one year had a median EuroSCORE II of 2.2, whereas those who died within one year had a median of 7.0. Cox regression demonstrated a hazard ratio of 1.062 for one-year mortality and 1.058 for long-term mortality. Kaplan–Meier curves showed significantly reduced survival with increasing EuroSCORE II quartiles. Logistic regression for one-year mortality yielded an AUC of 0.773, indicating good discriminative ability. Conclusions: EuroSCORE II is significantly associated with long-term mortality after major cardiac surgery, demonstrating good discriminatory performance. These findings support its potential utility not only as a short-term but also as a long-term prognostic indicator in cardiac surgery populations.