Challenges and Opportunities of a Cardiovascular Learning Healthcare System: Informed Consent, Trust, and a Pandemic

Abstract

Cardiovascular disease (CVD) remains a major health burden, requiring lifelong cardiovascular risk management (CVRM) for millions of patients. This is complex because patients often have multiple health conditions requiring a multidisciplinary approach. Guidelines exist to assist healthcare professionals in managing cardiovascular risk, but they are not always followed in clinical care and can quickly become outdated. A learning healthcare system (LHS) has the potential to improve cardiovascular care by continuously analysing routine care data, identifying areas of improvement and implementing changes in clinical practice. However, few LHSs are operational, and little is known about their challenges in real-world settings. This thesis explores these challenges by studying the cardiovascular LHS implemented at the University Medical Center Utrecht, focusing on the reuse of routine care data, obtaining informed consent, and the implementation of innovative clinical decision support systems (CDSSs). Chapter 2 examines differences in patient characteristics between consenting, non-consenting and non-responding patients. Compared with consenting patients, non-consenting patients had a higher CVD burden and fewer extractable CVRM indicators in structured electronic health record (EHR) fields. Despite these differences, Chapter 3 shows that established relations between cardiovascular risk factors were similar in consenting patients and the overall target population, suggesting that the selection unlikely led to biased relations. Chapter 4 suggests that an electronic informed consent (eIC) may result in less pronounced selection (i.e. in a study population more representative of the target population) than the traditional paper-based IC originally used in the LHS, potentially improving the generalisability of results. An eIC may therefore be a good alternative to a traditional informed consent for learning activities within the LHS that require consent. Chapter 5 compares outpatient attendance, CVRM and cardiovascular health before, during and after the COVID-19 pandemic and finds a significant decrease in first outpatient appointments since the implementation of the first lockdown. Furthermore, compared with the pre-pandemic reference period, patients who attended their first appointment at certain outpatient departments during the first lockdown had better cardiovascular health, suggesting that those at higher CVD risk may have avoided or been unable to attend the hospital during the lockdown. However, the extractability of CVRM indicators from structured EHR fields did not change during the COVID-19 pandemic. Chapter 6 describes how providing feedback to healthcare providers on the registration of CVRM indicators in structured EHR fields through dashboards alone did not contribute to more structured registration and identifies the perceived barriers to structured registration by healthcare providers, including time constraints, unclear responsibilities and unfamiliarity with the EHR system. Chapter 7 investigates whether the use of an artificial intelligence (AI) based CDSS affects patient-physician trust and highlights the need for transparency and patient involvement in the development and implementation of such AI tools in the patient room. Finally, Chapter 8 summarises the key findings of this thesis, discusses their implications and provides recommendations for future research and practice. The lessons learned from this thesis can support the further development of our LHS, as well as guide other hospitals in building their own LHS.