Transforming clinical decisions with artificial intelligence: the HYPERMARKER project

European Heart Journal, vol. 47, pp. 3232–3234

Abstract

Modern clinical practice benefits from a broad range of therapies that have been tested within clinical trials to effectively treat and manage cardiovascular disease (CVD). Therein also lies a major problem, with the diversity of patients, their clinical presentation, physician education, and the wide choice of medication options contributing to suboptimal CVD treatment. A notable example is hypertension, where despite clear guidelines,1 only a minority of patients meet blood pressure targets.2 Personalized therapeutic approaches across different medical specialities have long been imagined, but have rarely become part of routine clinical practice. However, with parallel advances in electronic healthcare records, patient-centred care and now artificial intelligence (AI),3 there is an opportunity to embed a more individualized approach to treatment selection (Figure 1). This has the potential to take advantage of developments in the ‘omic’ space, including the use of metabolomics which encompass an array of individual factors from the genome to a person’s environment.4 The value of personalizing hypertension therapy is clear,5 but evidence for how this should be deployed is lacking. Challenges and opportunities for treatment selection and the HYPERMARKER approach to change the hypertension paradigm. Top left panel: Factors that can affect treatment selection and choice in usual clinical practice, which often lead to variable treatment response. Top right panel: Where artificial intelligence can integrate data sources to help personalize therapy, leading to more predictable treatment response. Lower panel: The HYPERMARKER approach to improve treatment response for patients with elevated systolic blood pressure; for further details, see https://www.hypermarker.eu/ or the published design paper6 The HYPERMARKER consortium was designed to test the hypothesis that mass spectrometry-derived metabolomic profiles and clinical factors, curated by AI, could help to personalize the choice of medication class to treat elevated systolic blood pressure (SBP).6 Co-creation with a patient and public involvement team, use of home devices, and an online patient portal support the empowerment of people to take an active role in making shared treatment decisions. Funded by EU Horizon and UK Research & Innovation, HYPERMARKER includes 13 public and private partners across 5 countries. Metabolomic profiles have been generated from over 4000 patient blood samples stored from European trials and cohort studies that have assessed change in SBP from different classes and combinations of hypertension medication. Integrating both clinical and metabolomic features, an AI tool was created with explainability at its core to synthesize these data into an estimated probability of treatment success. The HYPERMARKER trial has recently commenced as a proof-of-concept pragmatic randomized trial, utilizing this tool to achieve more informed choices in SBP treatment selection (Figure 1). The trial will randomize 400 participants across multiple countries to initial use of this smart approach vs traditional standard-of-care drug prescription, with the primary outcome of change in patient-recorded, at-home SBP during a 4-week period after treatment change. To ensure equity, every participant and their clinical team, regardless of the original randomized group, will subsequently be offered the results from the smart approach, updated with knowledge gained during the first phase of the trial. The trial will also demonstrate a practical approach than can support the real-world deployment of guideline recommendations, in particular the use of early low-dose combination therapy to achieve better SBP control whilst minimizing side effects.1 Although genomics can be a powerful tool for treatment selection in relation to specific gene disorders, most common diseases have complex underlying genetic predisposition, variable penetrance, and diverse phenotypes. In contrast, metabolomics offers a current physiological snapshot of the individual’s health status accounting for other influences such as diet, medication, environmental exposure, and concomitant disease. Pharmaco-metabolomics allows for the monitoring of treatment efficacy and toxicity in real time, enabling personalized therapeutics and identification of non-responders that genetic markers might miss. Humans can accurately handle at best three-way interactions,7 and yet in clinical practice we are faced with countless interacting individual patient variables. This further multiplies once we consider blood tests, -omics and patient-reported measures, as well as their inherent variance within and across each person. Like any other tool, the value of AI is dependent its design, data source, and delivery3; however, there are numerous examples from clinical trial work demonstrating that AI can enhance CVD care,8 stratify therapy,9 and take account of complex and even missing data.10 In the context of decision-making for treatment selection, AI is a tool that can potentially synthesize the complex array of multi-domain health data—in this case, bringing together metabolomics, clinical factors, and individual treatment response to offer optimized hypertension management. Personalized approaches are already widespread, for example, genetic factors determining oncology treatment, the gut microbiome for precision nutrition, and real-time monitoring of physiology for diabetes and arrhythmia management. HYPERMARKER will demonstrate if this can extend to hypertension - a disorder affecting one in three of the global population, and the direct cause of millions of deaths annually, substantial morbidity, and unaffordable healthcare cost. By moving ‘downstream’ and using metabolomics, we seek to avoid the pitfalls that genomics poses for complex multi-morbidity conditions, providing a more functional output that can hopefully inform management. Even if there is no direct impact on blood pressure compared with usual care, HYPERMARKER will provide critical knowledge on how to implement AI tools to enhance shared decision-making between patients and clinicians, setting up for the anticipated future of technology-augmented routine healthcare. If successful, HYPERMARKER could ignite a new paradigm for treatment selection within common disorders, contributing to better patient outcomes, more robust health policy, and lower financial and societal burden from CVD. We acknowledge the members of the HYPERMARKER consortium: Animesh Acharjee, Banafsheh Arshi, Arie Baak, Birgit Beger, Iris Bos, Laura Bravo Merodio, Dominic Russ, Karina Bunting, Alex Carter, Asgher Champsi, Matthew Chapman, Annemarijn Douwes, Ana Duarte, Mai Elkhalek, Daniel Engler, Towfeeq Fairooz, Ana Ferrer-Albero, Pim Gal, Georgios V. Gkoutos, Kelly Grijsbach, Rick Grobbee, Xin Guan, Imane Guellil, Thomas Hankemeier, Ariadne Hogenboom, Abigail Johnson, Maria Jose Forner, Alida Kindt, Jonathan Knikman, Dipak Kotecha, Bart Lagerwaard, Fernando Martinez, Alistair Mcguire, Elham Memariam, Bristi Miah, Alastair Mobley, Santiago Montes, Elias Mossialos, Sam Muller, Alexandra Muñoz, Anouska Nithyanandan, Daniella Okyere, Durga Parkhi, Beatriz Peris Cuenca, Josep Redon, Gregorio Sambataro, Patricia Schlieker, Renate B. Schnabel, Erik Schulte, Diede Smith, Wilko Spiering, Carl Steinbeisser, Alicia Uijl, Merys Valdez, Yvonne T. van der Schouw, Oscar Van Dijk, Nikky Veerman, Roel Vermeulen, Jelle Vlaanderen, Xiaoxia Wang, John Williams, Yuanwei Xu, Elahe Zare Borzeshi, and Seraphine Zeitouny. All authors have completed the ICMJE uniform disclosure form and declare the following: D.K. reports receiving grants from the European Union Horizon scheme and UK Research and Innovation for this work; in addition to the National Institute for Health Research, British Heart Foundation, UK National Health Service, European Union Innovative Medicines Initiative and Innovative Health Initiative, European Society of Cardiology, UK Dept. for Business, Energy & Industrial Strategy, and Cook and Wolstenholme Charitable Trust; and personal fees from Atricure and Sandoz outside this work. T.H. and D.E.G. report receiving a grant from the European Union Horizon scheme for this work. No data were generated or analysed for or in support of this paper. The HYPERMARKER project is funded by the European Union (EU Horizon; grant number 101095480) and United Kingdom Research and Innovation (UKRI; grant number 10061996). The views and opinions expressed are those of the authors only and do not necessarily reflect those of the EU or UKRI. Neither the EU, UKRI, nor the granting authority can be held responsible for them. The work at the trial coordination centre was supported by the National Institute for Health and Care Research Birmingham Biomedical Research Centre (NIHR203326), British Heart Foundation Birmingham Accelerator, MRC Health Data Research UK (HDRUK/CFC/01), NHS Data for R&D Subnational Secure Data Environment programme (West Midlands), and the Cook & Wolstenholme Charitable Trust. UK: Approved by the North West–Greater Manchester West Research Ethics Committee (25/NW/0296) and the National Health Service Health Research Authority (IRAS project: 354889). Other countries pending. ClinicalTrials.gov: NCT07294794. ISRCTN: 29385951.

Authors 3

1. Dipak Kotecha corresponding

   University Medical Center Utrecht · University of Birmingham · University Hospitals Birmingham NHS Foundation Trust · NIHR Birmingham Biomedical Research Centre

   Affiliation as printed

   Department of Cardiovascular Sciences, University of Birmingham , Vincent Drive, Birmingham B15 2TT ,

   Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht , Universiteitsweg 100, 3584 CG Utrecht ,

   NIHR Birmingham Biomedical Research Centre, University Hospitals Birmingham NHS Foundation Trust , Mendelsohn Way, Birmingham B15 2TH ,

2. Thomas Hankemeier Aachen Metabolomics and Analytics Centre Leiden Academic Centre for Drug Research

   Leiden University

   Affiliation as printed

   Metabolomics and Analytics Centre, Leiden Academic Centre for Drug Research, Leiden University , Leiden ,

3. Diederick E. Grobbee

   University Medical Center Utrecht

   Affiliation as printed

   Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht , Universiteitsweg 100, 3584 CG Utrecht ,