Optimising hypertension management strategies

Using smart algorithms to optimise treatment decisions in blood pressure management and reduce economic burden on healthcare providers

Project status

Ongoing

Introduction

Current medical management of a number of common conditions are subject to measurement uncertainties and physiological variations which are rarely explicitly considered by clinical management algorithms, limiting the efficacy and efficiency of clinical care. This project proposes an innovative Bayesian hierarchical model to understand how blood pressure measurement uncertainties in large groups of patients, with complex behaviours, impact clinical decision-making, and to identify optimal management strategies for given measurement uncertainties. This has the potential to reduce cardiovascular events, increase the number of added quality-adjusted life-years, and improve the efficiency of healthcare delivery, thus providing a substantial saving opportunity for the NHS.

Explaining the science

This work proposes a Bayesian hierarchical model (BHM) that allows the optimisation of outcomes (e.g. the fraction of patients being treated correctly, or the predicted number of additional quality-adjusted life years), based on realistic input data (e.g. repeated blood pressure measurements, or other risk factors), and subject to real-world constraints (e.g. the cost of repeated measurements and medications, or the expected rate of patient adherence as a function treatment/measurement complexity).

The model uses Markov chain Monte Carlo methods to infer the output of the BHM, that is, the multi-dimensional posterior (output) distribution of the parameters of the model given the observations. These parameters represent the potentially adjustable aspects of any proposed treatment programme, such as the frequency and number of blood pressure measurements, the precision of each measurement, the weighting assigned to other risk factors, the cost of measurements and treatments. Finding the optimal values of these parameters – as well as the optimal set of parameters – is the ultimate goal of the modelling effort.

The pilot model investigates systolic blood pressure (BP) only, imposing a Gaussian distribution of measurement uncertainty and of drug effectiveness and assuming equal efficacy of anti-hypertensive treatments irrespective of BP or number of drugs. As a next step, the researchers plan to relax these assumptions and incorporate the following key features:

  • Examine the influence of deviations of measurement uncertainty and drug response from a Gaussian distribution
  • Wider choice of distributions including Gaussian, skewed and multi-modal to more accurately model the factors which influence response to treatment (e.g. age, ethnicity) and other factors determining cardiovascular risk
  • Option of performing two or more titration steps at once, and using fractions of ’full dose’
  • Inclusion of diastolic BP and pulse pressure
  • Variation in drug efficacy according to titration step and preceding titrations steps
  • BP variance from different measurement methods and the ability to switch between methods during a single individual’s treatment pathway
  • Include the economic modelling to investigate the healthcare burden of different management strategies

Project aims

Pilot data has demonstrated that large proportions of patients whose blood pressure levels are thought to be controlled are actually under- or over-treated. These misclassifications are a failure to reduce preventable cardiovascular events and/or side-effects from excessive medication. One key goal of this project is to develop advanced statistical models to conduct in silico clinical trials and explore the efficiency of various blood pressure management strategies; and subsequently perform clinical trials based on the most promising in silico results.

In doing so, the work aims to impact future iterations of the hypertension guideline from the National Institute for Health and Care Excellence (NICE) by being the first model to consider measurement error when making treatment decisions.

Mistreated patients resulting from misclassifications are a significant economic burden to the NHS. As such, the potential societal and economic impacts of this project are enormous given that approximately one third of adults are hypertensive. Improvements to the efficiency and efficacy of hypertensive management strategies would impact on patient outcomes by reducing both cardiovascular events and overtreatments, and in turn, lead to economic savings for the healthcare delivery.

Applications

It is important to effectively translate this research outcome to clinical practice by providing clinicians and/or patients with tools they can use. To this end, one objective of this project is to incorporate the outputs of the algorithm into a mobile app with the aim to help practitioners decide on the best way to proceed with treatment for a given clinical encounter.

Bayesian hierarchical models have long been used as a method of choice in science because of their native handling of uncertainty, their performance with sparse data, and their highly interpretable and intuitive nature. For this reason, these models are used in applications where data are heterogeneous and noisy, or when uncertainty needs to be clearly understood. This include areas such as extra-galactic astronomy, insurance, finance or healthcare.

Recent updates

November 2019

The promising results of this initial work lead to the obtention of a Silver Award at the STEM for Britain competition 2019 at the House of Commons, which prizes "ground-breaking, frontier" projects in R&D.