Risk prediction modelling and biomarker driven insights in heart failure

McDowell, Kirsty Marie (2026) Risk prediction modelling and biomarker driven insights in heart failure. PhD thesis, University of Glasgow.

Abstract

Heart Failure is a common problem associated with significant mortality and morbidity burden. It contributes significantly to hospital admissions and overall healthcare expenditure. The clinical trajectory of patients with heart failure can vary considerably with some being more at risk of adverse outcome than others. Advances in heart failure treatment have led to an improvement in prognosis and there is a better understanding of how to classify heart failure into phenotypes of ejection fraction. As a result, many of the risk models designed to predict risk in patients with HF are outdated, do not reflect specific phenotypes and no longer accurately discriminate risk.

Determining risk in patients with heart failure is important to help clinicians to have informed discussions with patients, their families and other healthcare providers. Using risk models integrated into clinical care pathways could help to streamline waiting lists. Risk models could be used to inform clinical trial design by enriching study cohorts with those most likely to experience the outcome. This could lead to shorter, less expensive trials, with reduced reliance on complex composite endpoints.

The role of biomarkers in multimarker risk prediction models is limited beyond the inclusion of natriuretic peptides, despite the growing interest in novel biomarkers that reflect different pathophysiological pathways. Given that natriuretic peptides are associated with outcome, strongly predict adverse outcomes and have become a target for treatment with angiotensin blocker neprilysin inhibitors it is logical to explore other potential biomarkers for similar prognostic and therapeutic value.

This thesis has two principal aims. Firstly to develop and validate a prediction model for clinical outcomes in heart failure with preserved ejection fraction (HFpEF) in a contemporary cohort, using routinely collected variables, to help with reproducibility of the model in clinical care. The model was used to assess the treatment effect of a contemporary treatment (finerenone) for HFpEF across the spectrum of defined risk.

Secondly I aimed to explore the incremental predictive value of circulating biomarkers to a robust prediction model developed in heart failure with reduced ejection fraction (HFrEF) and to explore 4 novel biomarkers and their association with outcome in HFrEF in greater detail, including how treatment with sodiumglucose cotransporter-2 inhibitors (SGLT2i) affected the measured level. This could potentially gain insight into the mechanism of action of these drugs.

Using large scale datasets from major clinical trials, including DELIVER, PARAGON-HF, I-PRESERVE AND FINEARTS-HF, novel risk prediction models for the outcome of cardiovascular death or heart failure hospitalization, cardiovascular death and all cause death were derived and externally validated for patients with HFpEF phenotypes. All models performed well with good discrimination and calibration in derivation and validation cohorts. Secondary analysis demonstrated that the treatment effects of finerenone, a non-steroidal mineralocorticoid receptor antagonist, was greatest in patients with heart failure and preserved ejection fraction who had the highest risk as defined by the models.

The addition of 11 biomarkers, measured in the PARADIGM-HF trial, to the Risk of Events and Deaths in the Contemporary Treatment of Heart Failure (PREDICTHF risk model) demonstrated that beyond high sensitivity troponin-T no other biomarker remained an independent predictor of all 3 endpoints explored (cardiovascular death or heart failure hospitalization, cardiovascular death and all-cause death). No biomarker improved the model reclassification indices. Three biomarkers (uric acid, vascular cell adhesion molecule 1 (VCAM-1) and interleukin 6 (IL6) were measured in DAPA-HF and found to be elevated in a contemporary cohort of patients with HFrEF. All biomarkers were associated with adverse outcome even after extensive adjustment for other variables. Uric acid was significantly reduced in those receiving dapagliflozin vs placebo. There was no significant difference in VCAM-1 or IL6 levels between baseline and follow up between those receiving dapagliflozin vs placebo. The addition of IL6 to PREDICT-HF improved model performance.

Together, these analyses demonstrate that phenotyping patients by ejection fraction helps to improve the performance of risk prediction model as compared to those derived in mixed cohorts. Accurate risk prediction is possible in HFpEF and multimarker models can help to assess treatment effects across strata of risk, supporting the evolution of personalized care in heart failure. Novel biomarkers are associated with outcome but do not always help with risk prediction beyond what is provided by routinely collected variables. IL-6 is elevated in heart failure, is associated with adverse outcome and improves a robustly validated prediction model in HFrEF. Given the availability of treatment to alter this pathway an outcome trial assessing anti-IL-6 therapy in HFrEF patients may be worth considering.

• Item Type: Thesis (PhD)
• Qualification Level: Doctoral
• Additional Information: Supported by funding from the National Institutes of Health and Dr Elke Platz.
• Subjects: R Medicine > RC Internal medicine
• Colleges/Schools: College of Medical Veterinary and Life Sciences > School of Cardiovascular & Metabolic Health
• Funder's Name: National Institutes of Health (NIH)
• Supervisor's Name: McMurray, Professor John, Campbell, Dr. Ross and Jhund, Professor Pardeep
• Date of Award: 2026
• Depositing User: Theses Team
• Unique ID: glathesis:2026-85870
• Copyright: Copyright of this thesis is held by the author.
• Date Deposited: 28 Apr 2026 13:25
• Last Modified: 28 Apr 2026 13:26
• Thesis DOI: 10.5525/gla.thesis.85870
• URI: https://theses.gla.ac.uk/id/eprint/85870

Related URLs

• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI
• Article DOI