The intensive care unit (ICU) is a hospital department where critically ill patients requiring organ support or intensive monitoring are admitted. Nowadays, the care provided in an intensive care unit has advanced so that more patients are being discharged alive. Advances in ICU care have increased patient survival rates, yet ICU admission is still associated with high morbidity and mortality both during and after the stay.

Identifying patients at high risk of complications during ICU admission is crucial. Even though there has been an increasing focus on predictive models, making accurate predictions with the data currently available remains challenging. A non-conventional parameter that appears to have promising value is heart rate variability (HRV), which reflects the fluctuations in time intervals between consecutive heartbeats and can be derived from the electrocardiogram. We investigated whether heart rate variability was able to predict ICU mortality and ICU length of stay. We employed a machine learning approach, assessing three models for each outcome. We used nested cross validation to estimate the performance and optimize and select the final models. Data from 468 adult patients admitted to the ICU for 48 hours or longer were analyzed and nine HRV measures were calculated. Two HRV measures, the power in the high frequency band and the standard deviation of 5-minute average RR intervals, showed a significant difference between the ICU survivors (n=398) and ICU non-survivors (n=71). While individual HRV measures had limited predictive power for ICU mortality, combining HRV with clinical features improved performance. The best performing model was an eXtreme Boosting Gradient classifier that used clinical features in combination with three HRV measures (power in the high frequency band, power in the low frequency band and the ratio between the power in the low frequency and high frequency band) achieving an AUC of 0.76. The models predicting ICU length of stay performed poorly, with the best model achieving a mean absolute error of 5.07 days. These findings suggest that HRV, when combined with clinical features, has potential in predicting ICU mortality, though further research is necessary before clinical implementation.

Monitoring ICU survivors after discharge is equally important, as half of these survivors suffer from Post Intensive Care Syndrome, which negatively impacts their quality of life and increases their healthcare needs. A team of researchers from the ICU conducted a pilot study establishing the feasibility of monitoring fifteen ICU survivors using eHealth, but a larger clinical trial was needed to assess feasibility on a larger scale. Therefore, we aimed to prepare for the ICU Recover Box 2.0 study. We evaluated the results and challenges of the first pilot. The study protocol was revised to include more participants and the use of new smart technology, specifically the Corsano CardioWatch, replacing the non-CE marked devices from the first pilot, which required 24/7 researcher availability. Preparations included finalizing the application to the Medical Ethical Review Committee and thinking out the study logistics, resulting in a comprehensive plan for the conduct of the study. Key takeaways from this process include the recognition that research is an iterative process of
continuous learning, that the purpose of the study must be carefully considered with ethical concerns in mind, and that the increasing importance of data requires careful planning for its security, processing and storage. The METC application has been submitted. Once approved, the study can proceed on a well-prepared basis.

Patients who are admitted to the Intensive Care Unit (ICU) are extremely ill and at high risk of organ failure and death. Being admitted to the ICU is known to cause long lasting physical, cognitive, and physiological symptoms, which is called Post-Intensive Care syndrome (PICS). To provide better management of PICS, early recognition and intervention are required. Two LUMC intensivists are exploring methods to increase ICU involvement in follow-up care for ICU patients, and thereby improve outcome related to PICS. Home monitoring may provide a method feasible for the ICU. To better assess feasibility, a pilot study, called "The ICU-Recover Box", was conducted. Additionally, since patients admitted to the ICU vary widely, it is essential to determine which patients are most likely to benefit from increased ICU involvement in follow-up care. An attempt was made to determine whether measurements recorded during ICU admission contain valuable information to predict ICU mortality. When this data contains valuable information, it may also be used to predict which patients should receive the ICU-Recover Box.

Fifteen patients who were admitted to the ICU received a weight scale, blood pressure monitor and smartwatch at hospital discharge. During the twelve week follow-up period, these patients would monitor their health status, which could be monitored remotely. Based on the preliminary results of six patients, a list of recommendations for large-scale clinical research and eventual implementation of the ICU-Recover Box into standard ICU practices, was formulated. The most important recommendations were to reduce the dependency on third parties and to substitute the home monitoring devices for devices that are medically CE-marked and have better ergonomics.

One recommendation was to assess which patients are most likely to benefit from home monitoring before implementing the ICU-Recover Box into standard ICU procedures. An attempt was made to predict ICU outcome using a machine learning approach. Data of 1364 patients, admitted to the LUMC ICU in 2017 and 2018, was collected and labeled "Survivor" or "ICU-Death", based on the ICU mortality. Features consisted of patient characteristics, hemodynamic status, and medication administrations. An Extreme Gradient Boosting (XGB) classifier was selected and optimized. Training and testing of this classifier was iterated 100 times. Mean AUC over all iterations was 0.809 with a standard deviation of 0.036. Mean sensitivity was 0.298 and mean specificity was 0.964, with standard deviations of 0.067 and 0.011 respectively. The mean corresponding F1 score was 0.851 with a standard deviation of 0.020. These results indicated that data recorded on the ICU seems to contain valuable information to predict ICU mortality. To identify whether this data could be used to determine which patients should receive the ICU-Recover Box, further research is required.

Cardiothoracic surgery is a common treatment for cardiovascular diseases. Patients are admitted to the ICU after cardiothoracic surgery for continuous monitoring to prevent or treat postoperative complications as much and as soon as possible. One of these complications is the development of circulatory shock. It is likely caused by one or a combination of several factors, leading to increased morbidity and mortality in the ICU. The main purpose of the circulation is to transport O2 and nutrients to the tissues and remove waste products of the tissues via the tissue’s microcirculation. Under normal conditions, O2 supply exceeds O2 demand. However, during circulatory shock, the circulation cannot meet the perfusion demands of the organs, leading to organ dysfunction and organ failure. Resuscitation procedures for patients with circulatory shock focus on normalizing macrocirculatory parameters, such as CO and SvO2, by administering fluids and vasopressors to support tissue perfusion. Improvement in macrocirculatory parameters is expected to be paralleled by improvement in microcirculatory perfusion and tissue oxygenation (i.e., hemodynamic coherence), but it appears that these do not always improve simultaneously. Loss of this coherence has been associated with adverse outcomes. 

The microcirculation can be imaged sublingually with an HVM. Studies in patients with septic shock have shown that hemodynamic coherence is often lacking. Therefore, it could be valuable to monitor the microcirculation of cardiothoracic surgery patients. 

This thesis aimed to investigate the postoperative time course of microcirculatory parameters in patients admitted to the ICU after cardiothoracic surgery with and without circulatory shock,  the relationship between macro- and microcirculation, and the usage of leukocyte detection in understanding patient’s systemic inflammation.  

Chapter 2 provides general background information on cardiothoracic surgery, CPB, the physiology of the microcirculation, the latest generation of HVM, pathophysiological changes in the microcirculation after cardiothoracic surgery, leukocyte-endothelium interactions, the macrocirculation, and hemodynamic coherence. A retrospective study of cardiothoracic surgical patients with shock is described in Chapter 3 of this thesis. The results showed that the microcirculation might adapt to compensate for the circulatory shock state by decreasing RBCv and increasing FCD, TVD, and cHct compared with normal values of healthy volunteers while maintaining tRBCp. Chapter 4 describes a prospective study comparing cardiothoracic surgical patients with shock from Chapter 3 with cardiothoracic surgical patients without shock. The comparison between these two groups showed that both groups exhibited different behavior of the microcirculation. However, the underlying mechanism is not understood and requires further research. Chapter 5 contains an explanatory review of the use of STDs for leukocyte detection. Chapter 6 provides a general discussion and reviews the future prospects of microcirculation measurements as a tool in the management of critically ill patients. Our findings should be examined in more extensive clinical trials to determine whether microcirculatory changes contribute to the development of shock.

Introduction
In the era of growing antimicrobial resistance, early detection and immediate treatment of antibiotic-resistant infections are crucial to ensuring successful outcomes in critically ill patients. The aim of this study is to apply machine learning (ML) to create classifiers that predict antibiotic resistance in postoperative infections caused by gram-negative pathogens, based on information retrievable from the Electronic Health Record. To determine if training prediction models on specific cultures improve the predictions’ performance, eight sub-datasets have been created that only included specific culture sources or cultures containing specific bacteria.
Methods
All surgical procedures in the LUMC between January 2015 and May 2023 after which gram-negative bacteria had been cultured within 30 days after surgery were included. Logistic regression, random forest and support vector machine models were developed, evaluated using area under the receiver operating characteristics (AUROC) metric and calibrated afterwards. For each model, the most important predictors were determined using SHAP values.
Results
In total 27 models were created for the dataset and eight sub-datasets. The dataset containing all positive postoperative cultures within 30 days of surgery consisted of 5777 procedures with a resistance rate of 27.4%. The AUROC for the models developed for the whole dataset ranged between 0.65-0.68 on an unseen test set. The AUROC on unseen data for models developed on specific culture sources ranged between 0.63-0.79 and for those trained on specific bacteria between 0.63-0.75. The included features that were deemed most important are the presence of a previous resistant culture, abdominal surgery, the presence of an indwelling device after surgery, and the patient’s sex, considering all (sub-)datasets.
Conclusions
This study shows that ML holds promise for predicting antimicrobial resistance. However, the current results are insufficient to support the implementation of these models in clinical practice to assist clinicians in choosing appropriate antibiotic therapy prior to receiving antibiogram results. With the current research, it cannot be proved with certainty that training models to particular postoperative infections enhances the predictions’ performance. To further investigate the potential clinical benefit of applying ML prediction models in the context of antibiotic resistance, further research is necessary on extracting more features, with increased quality, which are available at the time the culture is taken.

Intensive Care Unit (ICU) readmission is a serious adverse event associated with high mortality rates and costs. Prediction of ICU readmission could support physicians in their decision to discharge patients from the ICU to lower care wards. Due to increasing ICU data availability, Artificial Intelligence (AI) models in the form of machine learning (ML) algorithms can be used to build high-performing decision support tools. To have impact on patient outcomes, these decision support tools should have high discriminative performance and should be explainable to the ICU physician. The goal of this thesis was to compare several types of ML models on predictive performance and explainability for the prediction of ICU readmission for discharge decision support. The scientific paper that aims to answer this question can be found in Part III of this thesis. In a broader perspective, we proposed a framework for the development and implementation of clinically valuable AI-based decision support.
First, a systematic review was conducted to examine current literature on ML prediction models for ICU readmission (Part I). We concluded that previously developed models reported inappropriate performance metrics and were not implemented in clinical practice. Furthermore, previous work did not compare explainable outcomes in terms of patient factors contributing to the risk of readmission between models. Secondly, we conducted a questionnaire among ICU physicians to investigate current discharge practices and their attitude towards the use of AI tools in their work processes (Part II). Although not all physicians agreed that the decision to discharge ICU patients is complex, most of them do believe in the clinical value of an AI-based discharge decision support tool. Thirdly, we developed several prediction models for ICU readmission and compared them on discriminative performance, calibration properties, and explainability (Part III). We concluded that advanced ML models did not outperform logistic regression in terms of discriminative performance and calibration properties. However, the explanations of XGBoost, a state-of-the-art ML algorithm, were more in line with the ICU physician’s clinical reasoning compared to logistic regression and neural networks. Lastly, we designed a study protocol to prospectively evaluate the predictive performance of Pacmed Critical, a CE-certified AI-based discharge decision support tool, and that of the ICU physician (Part IV).
This thesis contributed to making the step from developing high-performing prediction models to clinical adoption of an ICU discharge decision support system. Due to small differences in discriminative power and calibration properties between models, the model best explainable to the physician and most in line with clinical reasoning should be chosen for decision support. Before final implementation, impact on patient outcomes and costs will need to be studied in prospective trials.