Academic dissertations

Patients acutely admitted to medical wards are at significant risk of clinical deterioration or adverse events. To address this, most hospitals have implemented protocols to monitor vital signs effectively, aiming to detect early signs of deterioration and enable timely interventions before the condition worsens. These protocols rely heavily on monitoring vital signs such as heart rate (HR), peripheral oxygen saturation (SpO2), blood pressure (BP), temperature (TP), and respiratory frequency (RF). This approach is based on evidence showing that abnormalities in vital signs often precede events like in-hospital cardiac arrest, unplanned intensive care unit (ICU) admissions, and mortality. Currently, the standard of care for monitoring vital signs in Denmark and many other regions involves manual, intermittent measurements within the UK-developed National Early Warning Score (NEWS) framework.

While technological advances in recent years have increased the use of wireless biosensors for monitoring biological signals for a range of different uses, the majority of hospitalized patients still rely on manual measurements for medical staff to track their physiological condition. How ever, systems for vital signs monitoring are now entering the market with the aim to improve patient safety and care at hospitals. These systems will change the paradigm of patient monitoring by introducing temporal and timely data in a resolution not previously adapted at general wards. With the increasing volume of biological signals, the one question to answer is: how are these to be analyzed and used to identify early signs of deterioration? By using unambiguous biosensors on patients during their hospitalization, we aim to explore methods for analyzing the signals ob tained given the challenges they entail, i.e. obtaining proper labels, working with missing data, and identifying early markers of physiological deterioration. As much of the data recorded during vital signs monitoring is not easily labeled for supervised learning, we explore the use of a Deep Generative Model to utilize unlabeled and labeled data in a semi-supervised manner.

A significant proportion of patients experience cognitive impairment after anesthesia and surgery1. Regardless of a proven relation between PCI and increased postoperative morbidity and mortality, most cases go unrecognized in clinical settings, partly due to the subtle symptomatology of the conditions2. PCI can be sub-classified into a time-dependent range of overlapping conditions with Emergence delirium (ED) occurring during the immediate postoperative period in the post-anesthesia care unit (PACU), postoperative delirium (POD) during the first postoperative days, and postoperative cognitive dysfunction (POCD) in days to months postoperatively. Furthermore, delirium exists in three different motoric subtypes, which are hyperactive, hypoactive, or mixed-type delirium3.

After major surgery, patients are at a high risk of various complications. The current standard of care for postoperative patients with regard to vital sign measurements is the intermittent monitoring of vital sign values such as heart rate (HR), blood pressure (BP), peripheral oxygen saturation (SpO2), and respiration frequency (RF). The model used in many health systems around the world is an adaptation of the British National Early Warning Score (NEWS) system. There is evidence to suggest that NEWS is able to identify patients with impending complications hours before they are recognized clinically, however, this has paradoxically not been shown to reduce ward morbidity and mortality. Continuous wireless vital sign monitoring and alerting systems (CVSMA) is a cutting-edge technology that has the potential to improve surgical outcomes, but so far, very little knowledge exists on how to interpret continuous vital signs, especially so outside of the intensive care unit. Several problems need to be solved before CVSMA systems can be expected to be a valuable tool in the care of postoperative patients.

Based on my online grocery shopping, Google knows my favorite dish before I do; my phone tells me when I’ve had enough sleep; I can track my kids’ activity while I’m at work; and it is possible to perform surgery on a patient in Denmark while the surgeon is located in another country [2]. For decades, technology has advanced rapidly, making our lives easier – and more complex in some ways. But despite technological advances and ongoing research, it is still challenging to detect patient deterioration in general hospital wards, avoiding severe complications. It is estimated that more than 313 million surgeries are performed each year [3], and both relatives and patients are relieved and assume that the most dangerous part is over when surgery is completed. Unfortunately, this belief is frequently incorrect as mortality 30 days after surgery is 100 times more common than intraoperative mortality, maybe because up to 30% of all patients undergoing major surgery will develop a complication within 30 days [4–6]. While surgical techniques and intraoperative treatment and care are constantly evolving, evaluated, and improved to reduce intraoperative mortality, much less emphasis has been placed on the postoperative period.

Patients who are hospitalized trust they are entering a safe place. They believe they are in the optimal place for prompt action, should their condition deteriorate further. However, there are data suggesting that this is often not the case. Unrecognized physiological instability can progress to severe complications, making it essential that impending deterioration is detected early. Current standard monitoring relies on intermittent observations and concern exists that these systems do not perform as well as expected, in particular in patients with chronic obstructive pulmonary disease (COPD). Patients admitted with acute exacerbation of COPD (AECOPD) have high mortality rates and may deteriorate suddenly. Wireless monitoring systems can measure vital signs continuously and this may improve patient care by permitting physiological abnormalities to be detected earlier. This thesis aimed to investigate vital sign abnormalities in the AECOPD setting (Study 1) and to evaluate the accuracy of wireless devices (Study 2). We also aimed to investigate the association between frequency and duration of vital sign abnormalities and subsequent serious adverse events (SAEs) (Study 3).