A Hierarchical Bayesian approach to optimising hypertension management strategies

Lead Research Organisation: University of Southampton
Department Name: School of Physics and Astronomy

Abstract

Many decisions in medicine are subject to measurement uncertainties and physiological variations which mean that treatment decisions may be made erroneously. These uncertainties are rarely explicitly considered in clinical management algorithms, limiting the efficacy and efficiency of clinical care. Management of raised blood pressure (hypertension) is a particularly important example, as hypertension is the single greatest cause of death and disability worldwide. In the UK approximately 1 in 3 adults require drug treatment for hypertension, imposing a huge burden on health care delivery. In an emergent collaboration between the Southampton Astronomy group and the Department of Clinical Pharmacology of St Thomas' Hospital at King's College London, we have adapted Monte Carlo simulations used in extra-galactic Astronomy to model the random effects of measurement uncertainty in a virtual population of hypertensive individuals. Our work showed that current treatment strategies for medication are too inefficient, with typically 40% of the population not optimally controlled, and thus at risk of adverse events. Our work obtained a Silver Award at the STEM for Britain competition 2019 at the House of Commons, which prizes "ground-breaking, frontier" projects in R&D. Building on the recent success of our collaboration, in this research proposal we aim to produce a tailored Hierarchical Bayesian Monte Carlo algorithm to develop the first smart blood pressure management algorithm. This algorithm will aim to combine patient-specific factors (for example starting blood pressure, sex, age and weight) with drug efficacy and measurement error, to predict the probability of an individual achieving blood pressure control for a given approach. The model will be validated using published data (from both clinical trials and observational cohorts) and real-world patient journeys from the St Thomas' Hospital Hypertension Clinic. More specifically, making use of anonymised data in the public domain, we will adopt the smart algorithm to conduct in silico clinical trials which aim to improve the proportion of hypertensive individuals achieving the desired blood pressure target with the minimal burden on both patient and healthcare system. This series of virtual clinical trials will aim to identify the most promising management approach(s) to take forward into real-world studies. Cardiovascular diseases have a huge cost of tens of millions pounds in the UK. Whilst the final evaluation of this work would require validation by means of a clinical trial comparing a final personalised treatment plan to standard care, the present approach has the potential to rapidly perform a large number of "in-silico" (i.e, virtual/simulated) comparisons to select a near-optimal treatment plan that can be tested in a clinical trial. Furthermore, it will provide a quantitative prediction of the degree of improvement expected, with the improved plan providing the necessary information to set up the clinical trial adequately. Our project has the potential to reduce cardiovascular events, improve efficiency of healthcare delivery, thus providing a substantial saving opportunity for the NHS. We will disseminate our work through the publication of peer-reviewed manuscripts and presentations at national/international conferences. We then envision a comprehensive research dissemination programme supported by in-house dissemination officers at the University of Southampton and at King's College London.

Planned Impact

Academic dissemination and impact:
To maximise the impact of our research project within both the astronomical and medical academic communities, we aim to provide an extensive and efficient dissemination in two distinct ways: We will promote our work through the publication of peer-reviewed manuscripts and presentations at national/international conferences.
2. We then envision a comprehensive research dissemination programme supported by in-house dissemination officers at the University of Southampton and at King's College London.

Societal dissemination and impact:
Public engagement will involve face-to-face activities with three main aims:
i) Increase the interest/awareness and knowledge/understanding about the Astronomy, medical science and the interconnection between them and the impact they have on our lives; ii) Attract new audiences; iii) Communicate results to the general public via mainly online material.
We will assess the impact of the outreach and public engagement activities via quantitative and qualitative evaluations. The feedback results from the evaluations will allow us to improve activities for the next round.
We will also produce online blogs, organise at least one general public talk, prepare talks/posters for large public outreach events, organise visits to local schools and participate in open-door days.

Economic impact:
i) Current success. Through the promising results of our initial work with St Thomas' Hospital,
PhD Astronomy student Lorenzo Zanisi obtained a Silver Award at the STEM for Britain competition 2019 at the House of Commons, which prizes "ground-breaking, frontier" projects in R&D. This project is in fact designed to fit in very well with the University of Southampton's, and more generally STFC's, distinctive interdisciplinary research and enterprise primary goal to "generate research and technologies that give real economic and social benefits".

ii) Savings for healthcare delivery. The patient-centric outcomes of blood pressure misclassification are a failure to reduce preventable cardiovascular events and/or side-effects from excessive medication or blood pressure-lowering. There is a subsequent significant economic burden on the NHS as a result of misclassification, and therefore an equal opportunity for savings. The potential societal impact of this project is thus enormous given that approximately one third of adults are hypertensive. Improvement to the efficiency and efficacy of hypertensive management would lead to improved patient outcomes through reduced cardiovascular events, reduced overtreatment and economic savings.

iii) Future technological developments. After the end of the grant we expect, partly supported by additional funding, to incorporate the smart algorithm into an App which will be piloted within the Hypertension Clinic at St Thomas' Hospital. The latter will incorporate the simulations' outputs and will aim to help practitioners decide the best way to proceed with treatment for a given clinical encounter.

ii) Research Excellence Framework. Last but not least, with the close collaboration of St Thomas' Hospital and the additional support of the current co-Is, we believe this research project can lay the grounds for more substantial interdisciplinary grant proposals and a competitive impact case for the next Research Excellence Framework assessments (2021, 2028). The project in fact directly transposes statistical and numerical algorithms, developed in the context of extra-galactic Astronomy, to vital medical applications with a clear societal impact, and accompanied by a detailed programme for testing and validation.

Publications

10 25 50
 
Description STFC Impact Acceleration grants
Amount £50,000 (GBP)
Organisation Science and Technologies Facilities Council (STFC) 
Sector Public
Country United Kingdom
Start 11/2019 
End 06/2021
 
Description Strategic Development Fund Award
Amount £57,171 (GBP)
Organisation University of Southampton 
Sector Academic/University
Country United Kingdom
Start 03/2021 
End 07/2021
 
Description The Alan Turing Pilot Projects
Amount £77,000 (GBP)
Organisation Alan Turing Institute 
Sector Academic/University
Country United Kingdom
Start 02/2020 
End 07/2021
 
Title Numerical convolution of blood pressure distributions with any type of filter 
Description Although simple and accurate, the relatively slow nature of Monte Carlo simulations applied to an entire population of individuals makes it challenging to use in everyday practice. As such we further built upon an existing analytic convolutional formulation, originally designed in the context of Astrophysics [Shankar et al., 2004; Marconi et al., 2004], which we solve using numerical integration methods for an entire population of individuals in a single pass. Besides its speed, a key aspect of this new formalism is that it can be applied to any arbitrary set of constraints that can be expressed via probability distributions. 
Type Of Material Data analysis technique 
Year Produced 2021 
Provided To Others? No  
Impact This numerical convolution algorithm will be used in the development of the App laying the foundations for personalised treatment for each individual by leveraging the history of measured blood pressure distributions obtained during previous visits, and using this history to: (i) individually estimate and track true (systolic) blood preddure over time for each patient, (ii) estimate the variability in the measurement device, (iii) estimate the individual's unique physiological variability in the medication response (i.e. drug effectiveness). We envision that this flexible and fast numerical convolution method will be published in the App paper which is now under writing and will be submitted soon. 
 
Description Collaboration with Clinical Pharmacologists 
Organisation King's College London
Country United Kingdom 
Sector Academic/University 
PI Contribution In an ongoing successful collaboration, already partly funded by STFC, between the Southampton Astronomy group and Clinical Pharmacologists at King's College London, we have transferred state-of-the-art techniques used in extra-galactic astronomy to medical science.
Collaborator Contribution Dr Christopher Floyd brings a strong track record of relevant cardiovascular research and clinical experience, providing expertise in the vast array of parameters involved that describe real-world treatment strategies and their feasibility for clinical implementation. Furthermore, Dr Floyd, CEO of Cranworth, is playing a vital role in the further development of the App. Dr Floyd will also provide us with direct access to data from Nigerian clinics, readily available from his close collaborators.
Impact *Two published research papers *One paper on the cardiovascular disease risk model to be submitted *One paper on the mobile App, in preparation
Start Year 2019
 
Description Collaboration with The Alan Turing Institute 
Organisation Alan Turing Institute
Country United Kingdom 
Sector Academic/University 
PI Contribution The project supported by STFC has significantly contributed to a long-term partnership with the Alan Turing Institute. I have indeed been elected a fellow of the Turing since 2019 (https://www.turing.ac.uk/people/researchers/francesco-shankar). The grant awarded by the Turing allowed our team to hire a postdoc who developed a cardiovascular disease (CVD) event risk model which, in addition to incorporating current clinical measures of risk (i.e. the QRISK tool), provides a novel tool also able to dynamically adjust risk level with SBP as it evolves through treatment, providing a more realistic estimation of CVD risk than current approaches. The model was built based upon previously acquired empirical data and is also able to segregate risk by type of CVD event (e.g. myocardial infarction or stroke). The model employs Monte Carlo techniques to estimate number of CVD events across populations from individual's risk level and provides a dynamic assessment of evolving risk of CVD events, through which treatment strategies may be compared across several treatment cycles across populations.
Collaborator Contribution The Alan Turing Institute supported provided a grant of ~£80K.
Impact To-date, we have used the Monte Carlo and CVD models to investigate the effect of initial and ongoing measurement error, age and gender on blood pressure (BP) and cardiovascular risk. To realise the full impact of this work we need to expand the scope of our investigation of treatment parameters and verify our models using real-world data so that they can then be applied to targeted populations and individual patients in a robust manner. As such, our longer term objective is to further exploit our models, by conducting in silico clinical trials that mimic a vast array of different treatment strategies, such as varying timing of treatments and dose. These trials will identify which BP management treatment strategies are most likely to improve short- and long-term BP control and reduce population CVD events across the population, whilst also considering the trade-off with treatment and visit costs. Supported by external grants (e.g., STFC IAA), we are building a mobile App that starting from a few BP measurements as well as patient's historical data prior to hypertension treatment, can provide the probability distribution of the effectiveness of a given treatment, taking into account all relevant measurement uncertainties, as well as include a CDV risk probability model. We expect this tool, once ready and fully tested, to be an invaluable resource for all medical practioners.
Start Year 2019
 
Title Mobile clinical App 
Description Our current Monte Carlo simulations are powerful techniques to investigate the impact of drug response, measurement variability, and therapeutic inertia on large samples of hypertensive individuals. By including measurement uncertainties, they in fact allow to distinguish measured blood pressure (BP) from true/intrinsic BP distributions. Our technique is therefore able to estimate the proportion of inaccurate BP estimates caused by inherent variability in non-invasive BP measurements. To transition our work to a stage of real world impact, supported by Impact Acceleration grants, we have developed a first stage of a mobile App to act as a decision support tool for clinicians. Our team, composed of expert numerical modellers, statisticians, medical scientists and App developers, is indeed best placed to create new and unparalleled digital tools that for the first time will allow to achieve personalised antihypertensive therapies beyond crude cut-offs based on age and ethnicity. The App aims at predicting current and future true systolic blood pressure (tSBP) after medication(s) from a single SBP measurement (mSBP). In particular, the App estimates the current true BP and the probability of true BP to be below the recommended treatment threshold of 140 mmHg [NICE hypertension guidelines, 2019] after a single, double and no medications at 95% confidence level. We have factored in a number of clinical constraints to design a robust tool for making inference at each patient's visit. First, the average reduction in BP of a medication has been shown to be dependent on true BP [Law, 2003; Law et al., 2009] (i.e. the lower the true BP, the lower the reduction). Second, each medication intake must strictly reduce true BP. This work on the App lays the foundations for personalised treatment for each individual by leveraging the history of measure BPs obtained during previous visits, and using this history to: (i) individually estimate and track tSBP over time for each patient, (ii) estimate the variability in the measurement device, (iii) estimate the individual's unique physiological variability in the medication response (i.e. drug effectiveness). 
Type Management of Diseases and Conditions
Current Stage Of Development Refinement. Non-clinical
Year Development Stage Completed 2021
Development Status Under active development/distribution
Impact Via further development of our mobile App we aim to transition from population-level prediction to the personalisation of individual care. With sufficiently tested models, we will incorporate additional functionalities into our digital tools. For example, the risk levels (based upon evolving BP, gender age and ten-year risk), which the Monte Carlo estimation uses to evaluate whether an individual within a population suffers with a cardiovascular event at each treatment cycle, will also be extracted to provide feedback on risk of different types of event (e.g. stroke, cardiovascular related death, angina, etc.), and will be incorporated into the App to provide further useful information to clinicians. Our work will inform national/international guidelines, with the primary aim of optimisation of both cost (e.g. by fewer visits to the doctor or reduced treatment) and BP control (e.g. lower risk of cardiovascular event) across populations, and will ultimately provide a more efficient and personalised treatment, which is only heuristically implemented in current practice.