Artificial Intelligence in Predicting Early and Long Term Outcomes in Cardiovascular Disease

Heart attacks remains the most common cause of premature death in the world. Despite modern medical treatment, there is increased risk of these patients having further heart attacks or indeed sudden death. This is often very difficult to predict in short or long term basis. Several risk prediction models have been developed to study the risk of recurrent events. However, these risks scores are not always accurate and do not perform very well in all patients. This is due to a number of factors including the fact that the diagnostic criteria and definitions of heart attack have evolved over time. Furthermore, the treatment strategies following a heart attack has progressed significantly over last many years.

Newer data analysis techniques based on artificial intelligence programme have been developed in the last few years that are superior to traditional utilised statistical analysis, and are currently utilized in several scientific fields. We proposed to use the machine learning techniques to analyse the data gathered from patients with heart attacks to see if these provide superior risk prediction of cardiac events following a heart attack.

Using state of the art technology, we will analyse the routinely collected data from the electronic records and pre-existing database in patients with heart attacks at the Bristol Heart Institute. Using machine learning techniques, we will create algorithms that may better predict future cardiovascular events in the first phase in about 10000 patients. The will be undertaken after testing several techniques that are well established within the field of machine learning. We will use then use the model that best predicts the future cardiac events in patient data in patients with heart attack a national level in a big data analysis.

In addition, we will also study the utility of this model to predict long term outcomes following a heart attack. Furthermore, pathways in patient management in patients with heart attack will be studied, using a technique called as process mining. This will allow us to identify the bottleneck in admission to treatment of these patients.

If the study confirms the performance of the new model better than pre-existing, this is likely to have significant impact on patient care straightaway. It will help us identify patients that are at the highest risk of future cardiac events. Ultimately, this the transform the care in these patient groups by more intensive monitoring and treatment.

Myocardial infarction remains the leading cause of death worldwide. Prediction of short and long term mortality following acute coronary syndromes (ACS) remains a challenge. Currently, there are only few prediction models which have been developed and calibrated for ACS patients that include GRACE and TIMI risk scores. These scores have demonstrated lower calibration and discrimination for high risk cases and significant calibration drift linked to changes in patient's characteristics and the introduction of new treatment over the time. This is more relevant to introduction of high sensitivity troponin assays, that significant increase the number of patients diagnosed to have ACS. These score have been developed using parametric logistic regression to calculate the risk of death.

In a proof-of-concept study, a local, big data-driven, machine learning approach is compared to existing GRACE and TIMI risk scores to predict mortality in patients admitted with ACS. Pilot data will be used to guide further research on the application of machine learning algorithms in risk prediction in cardiovascular disease at national level using national NICOR database. These data will guide future projects aiming to build an automated predictive analytics framework for machine-learning algorithm with high discriminatory ability for assessing patients admitted with acute coronary syndrome. We will also explore the prediction of long term outcomes following ACS following initial index events and will assess the ACS pathway in these patients to identify factors responsible for potential delay, and assess if certain subgroups of patients benefit for early intervention.

This important work will potentially refine the risk prediction in patients with ACS, and ultimately help us identify patients who will be benefit from aggressive monitoring and treatment. This is essential to develop an effective strategy to reduce mortality and recurrent events following myocardial infarction.

The study has potential to have a significant impact on the overall risk stratification of patients following myocardial infarction (MI). The current models for risk stratification based on TIMI and GRACE scores have been developed in the era when majority of ST segment elevation myocardial infarctions (STEMI) were thrombolysed. Since then, there has been a step change in the treatment of MI including primary percutaneous coronary intervention (PCI), early invasive treatment for non-STEMI, and other pharmacological interventions.

As such, although the GRACE and TIMI scores still remain relevant, they need to be evaluated in the current practice. This is even more relevant since the current introduction of high sensitivity troponin assays have made a step change in the diagnosis of MI. The overall prognosis and outcomes of these patients with in small troponin rise is likely to be different to traditional patients presenting as STEMIs and indeed very unstable NSTEMIs.

If the study achieves its intended goals of refining the risk stratification of patients with myocardial infarction, this will have a wide ranging impact on the patients with myocardial infarction is likely to be widely adapted in the clinical practice straightaway.

Adaption or indeed changing clinical practice especially the treatment pathways post MI (such as time from admission to cath-lab, time to discharge etc), can be robustly analysed using process mining using machine learning, and this meaning this project will be of great interest of the general medical and cardiology community.

As such the likely beneficiaries of the project are the policy makers at the national and international levels such as ESC, ACC, and NICE. If this study suggest benefits over existing risk prediction models, this is straightaway likely to be included in the updated guidelines for management of ACS: Emergency departments, frontline clinical staff treating patients with chest pain, cardiac units managing patients with MI, and finally the primary care physicians since the aims of the project including short and long term risk prediction following myocardial infarction.

The ultimate beneficiaries are the wider public because this project is likely to have an impact. The overall impact of the proposed studies is likely to be widespread across multiple stakeholder groups and various regions including the inclusion of low- and middle-income nations.