Background: Mortality prediction models are applied in the intensive care unit (ICU) to stratify patients into different risk categories and to facilitate benchmarking. To ensure that the correct prediction models are applied for these purposes, the best performing models must be identified. As a first step, we aimed to establish a systematic review of mortality prediction models in critically ill patients.Methods: Mortality prediction models were searched in four databases using the following criteria: developed for use in adult ICU patients in high‐income countries, with mortality as primary or secondary outcome. Characteristics and performance measures of the models were summarized. Performance was presented in terms of discrimination, calibration and overall performance measures presented in the original publication.Results: In total, 43 mortality prediction models were included in the final analysis. In all, 15 models were only internally validated (35%), 13 externally (30%) and 10 (23%) were both internally and externally validated by the original researchers. Discrimination was assessed in 42 models (98%). Commonly used calibration measures were the Hosmer‐Lemeshow test (60%) and the calibration plot (28%). Calibration was not assessed in 11 models (26%). Overall performance was assessed in the Brier score (19%) and the Nagelkerke's R² (4.7%).Conclusions: Mortality prediction models have varying methodology, and validation and performance of individual models differ. External validation by the original researchers is often lacking and head‐to‐head comparisons are urgently needed to identify the best performing mortality prediction models for guiding clinical care and research in different settings and populations.

Critically ill patients constitute a highly heterogeneous population, with seemingly distinct patients having similar outcomes, and patients with the same admission diagnosis having opposite clinical trajectories. We aimed to develop a machine learning methodology that identifies and provides better characterization of patient clusters at high risk of mortality and kidney injury. We analysed prospectively collected data including co-morbidities, clinical examination, and laboratory parameters from a minimally-selected population of 743 patients admitted to the ICU of a Dutch hospital between 2015 and 2017. We compared four clustering methodologies and trained a classifier to predict and validate cluster membership. The contribution of different variables to the predicted cluster membership was assessed using SHapley Additive exPlanations values. We found that deep embedded clustering yielded better results compared to the traditional clustering algorithms. The best cluster configuration was achieved for 6 clusters. All clusters were clinically recognizable, and differed in in-ICU, 30-day, and 90-day mortality, as well as incidence of acute kidney injury. We identified two high mortality risk clusters with at least 60%, 40%, and 30% increased. ICU, 30-day and 90-day mortality, and a low risk cluster with 25–56% lower mortality risk. This machine learning methodology combining deep embedded clustering and variable importance analysis, which we made publicly available, is a possible solution to challenges previously encountered by clustering analyses in heterogeneous patient populations and may help improve the characterization of risk groups in critical care.

Postoperative mortality risk is associated with several preoperative, intraoperative, and postoperative factors, and age, comorbidities, and decreased preoperative kidney function are known factors associated with increased risk.^1-6 Current preoperative risk stratification methods, such as the European System for Cardiac Operative Risk Evaluation (EuroSCORE) II⁷ and American Society of Thoracic Surgeons score,^8,9 therefore use these preoperative variables to predict short-term risk. Among postoperative factors, acute kidney injury (AKI) is 1 of the best characterized factors associated with risk for mortality in coronary artery bypass grafting (CABG) and valve operations.^1,10 However, the usefulness of routinely collected and more complex intraoperative and postoperative clinical parameters other than kidney function to predict short-term and especially long-term mortality remains understudied.

Longitudinal evaluations of critically ill patients by combinations of clinical examination, biochemical analysis and critical care ultrasonography (CCUS) may detect adverse events of interventions such as fluid overload at an early stage. The Simple Intensive Care Studies (SICS) is a research line that focuses on the prognostic and diagnostic value of combinations of clinical variables. The SICS-I specifically focused on the use of clinical variables obtained within 24 h of acute admission for prediction of cardiac output (CO) and mortality. Its sequel, SICS-II, focuses on repeated evaluations during ICU admission. The first clinical examination by trained researchers is performed within 3 h after admission consisting of physical examination and educated guessing. The second clinical examination is performed within 24 h after admission and includes physical examination and educated guessing, biochemical analysis and CCUS assessments of heart, lungs, inferior vena cava (IVC) and kidney. This evaluation is repeated at days 3 and 5 after admission. CCUS images are validated by an independent expert, and all data is registered in an online secured database. Follow-up at 90 days includes registration of complications and survival status according to patient's medical charts and the municipal person registry. The primary focus of SICS-II is the association between venous congestion and organ dysfunction. The purpose of publishing this protocol is to provide details on the structure and methods of this on-going prospective observational cohort study allowing answering multiple research questions. The design of the data collection of combined clinical examination and CCUS assessments in critically ill patients are explicated. The SICS-II is open for other centers to participate and is open for other research questions that can be answered with our data.