Machine Learning: The Heart of Precision Medicine

Atrial fibrillation is a common form of heart arrhythmia that affects tens of thousands of New Zealanders around the country.
cute girl sitting in between her grandparents

Summer of Research

Project by Nicholas James, University of Auckland, supervised by Dr Patrick Gladding.

You’ve seen it happen in movies I’m sure. A main character is flat lining and the doctor yells ‘CLEAR!’ before bringing down the paddles and zapping them back to life. Yeah, that’s not actually what those machines do. In real life, defibrillators are used to shock an abnormally beating heart back into a normal rhythm. It doesn’t start a stopped heart, but sort of ‘resets’ a malfunctioning one.

Atrial fibrillation is a common form of heart arrhythmia that affects tens of thousands of New Zealanders around the country. It’s a disease that causes an abnormal heart rhythm, leading to symptoms such as heart palpitations, fainting, lightheadedness, shortness of breath or chest pain. It is also related to higher risks of heart failure, dementia and stroke.

Atrial fibrillation can be treated in a number of ways. There are several medications that can be used to treat both atrial fibrillation and its symptoms, such as warfarin to prevent strokes or metoprolol to control heartrate. Apart from pharmaceuticals however, the main alternative for rhythm control is Direct Current Cardioversion (DCCV). DCCV applies an electrical current across the heart to revert it back into a normal rhythm.

About 80% of patients who undergo DCCV return to a normal heart rhythm [1]. Unfortunately, after only 6 months, half will have gone back to an abnormal rhythm. After 12 months, only 1/3rd retain normal rhythm [2,3]. Only a few factors have been found to determine the likelihood of retaining a normal heart rhythm after DCCV. These include: duration of current atrial fibrillation, physical left atrial dimensions, and genetics [4,5].

Nicholas James felt that it would be helpful to identify more factors that make a patient likely to retain a normal heart rhythm after DCCV. This would both validate existing research, as well as attempt to discover new information that could lead to improved treatment for atrial fibrillation sufferers. As part of his summer of research project, Nicholas would collect anonymous data on atrial fibrillation patients and their treatment, and use machine learning to develop a predicted rate of DCCV success for different patient profiles.

Nicholas studied the records of 146 patients who underwent DCCV during a 22-month period. These patients were also sent mouth swabs for genetic analysis, to see whether certain DNA variations were linked to the success of DCCV.

Patient data was separated into two groups:

  • Patients who reverted to atrial fibrilliation within 3 months of DCCV (Low success).
  • Patients who retained normal rhythm for one year or longer after DCCV (High success).

As shown in the graph below, approximately one third of patients reverted into atrial fibrillation at 3 months after DCCV, and over half reverted by 12 months.

Machine learning outputs showed the relative importance of different factors associated with low success rates of DCCV. The most important factor was the length of time between diagnosis and DCCV treatment. The sooner a patient was treated with DCCV after diagnosis, the more likely they were to retain a normal heart rhythm for longer. Nicholas hypothesised that this may be because the heart undergoes electrical remodeling if atrial fibrillation continues for long periods of time.

Nicholas’ research examined other factors such as medication used, age, ethnicity, and genetics. Each of these areas showed potential to influence the effects of DCCV, but would have benefitted from a larger sample size. Regardless, the study effectively demonstrates the benefit of using machine learning to analyse health data. Being able to predict the outcomes of certain treatments for individuals is a huge focus for the Precision Driven Health initiative, improving the healthcare experience for patients and clinicians alike.

Nicholas James is among a group of students who took part in the summer of research programme funded by Precision Driven Health. This month we are featuring a blog series examining these projects. While at an elementary stage and considered to be a ‘proof of concept’, these projects offer fresh insights into what the world of healthcare will look like when precision medicine is fully implemented.

1. Boodhoo L, Mitchell AR, Bordoli G, Lloyd G, Patel N, Sulke N. DC cardioversion of persistent atrial fibrillation: a comparison of two protocols. International journal of cardiology. 2007 Jan 2;114(1):16-21.

2. Van Gelder IC, Crijns HJ, Van Gilst WH, Verwer R, Lie KI. Prediction of uneventful cardioversion and maintenance of sinus rhythm from direct-current electrical cardioversion of chronic atrial fibrillation and flutter. The American journal of cardiology. 1991 Jul 1;68(1):41-6.

3. Luong C, Gin K, Bennett M, Jue J, Ramanathan K, Barnes M, Colley P, Thompson D, Tsang TS. Prediction of Atrial Fibrillation Recurrence at Six Months Post Direct Current Cardioversion: Right Atrial Volume, Left Atrial Volume or Both?. Journal of the American College of Cardiology. 2013 Mar 12;61(10):E369.

4. Fornengo C, Antolini M, Frea S, Gallo C, Marra WG, Morello M, Gaita F. Prediction of atrial fibrillation recurrence after cardioversion in patients with left-atrial dilation. Eur Heart J Cardiovasc Imaging. 2014 Oct 1:jeu193.

5. Parvez B, Shoemaker MB, Muhammad R, Richardson R, Jiang L, Blair MA, Roden DM, Darbar D. Common genetic polymorphism at 4q25 locus predicts atrial fibrillation recurrence after successful cardioversion. Heart Rhythm. 2013 Jun 30;10(6):849-55.