Automating the fight against heart disease with artificial intelligence

Carolyn Lam, PhD

Professor in Cardiology, cofounder of Us2.ai

James Hare, MBA

Cofounder and Chief Executive Officer of Us2.ai

Automating the fight against heart disease with artificial intelligence

29 September 2020

Quick Takes

  • Timely and accurate diagnosis is crucial to enable early initiation of key life saving therapies and reduce hospitalizations

  • Artificial intelligence (AI) software can simplify and democratize diagnostic tools that are traditionally resource intensive, such as echocardiography for heart failure

  • Cautious skepticism in the reliability of AI tools can be overcome with proven value to health systems and patients

Timely and accurate diagnosis is crucial to enable early initiation of key life saving therapies and to reduce hospitalizations.

Us2.ai is automating the fight against heart disease by using artificial intelligence (AI) software to simplify and democratize ultrasound of the heart – the most commonly used tool for the detection of cardiovascular risk. The founders, Carolyn Lam and James Hare, share their insights on why delivering high-value solutions will transform healthcare.

Turning on the lights: A better understanding of our own heart health 

HT: What unmet medical need did you set out to address with your start-up?

Us2.ai: Heart failure (HF) is a global public health problem, affecting over 64 million people worldwide.1 The cost worldwide is estimated at $108 billion per year with a large bulk of the costs due to hospitalization.2 Despite advances in management, HF remains as malignant as some common cancers3 and survival post-diagnosis is only around 50% at 5 years.4 

As the overall awareness is quite low – very often, this condition remains underdiagnosed and undertreated. In addition, two thirds of the HF patients are readmitted to hospital within one year. Due to high and increasing prevalence rates, and frequent rehospitalizations, HF constitutes an enormous economic burden for the healthcare institutions and healthcare systems in industrialized countries.

The economic impact of a disease is considered in terms of direct and indirect costs. Direct costs include healthcare expenditure on hospital services, medications, physician costs, primary healthcare costs and follow-up. Indirect costs include healthcare expenditure in terms of lost productivity resulting from morbidity and mortality, sickness benefit and welfare support.

While kitesurfing around the world as a retired tech entrepreneur, James Hare, one of the founders of Us2.ai, underwent his first ever ultrasound exam and was surprised to discover just how little we know about our own hearts, even though cardiovascular disease is the number one killer worldwide. Us2.ai was created to solve this problem by using AI to democratize access to one’s heart health.

It is well known that when appropriate HF therapy is initiated early, it leads to better outcomes for patients and reduced hospitalizations.5 Therefore, timely and accurate diagnosis is crucial to enable early initiation of key life saving therapies, reduction in hospitalizations and the direct impact on the reduction of costs.

HT: If better HF diagnosis can impact cost reduction and hospitalization, then what are the main reasons for late- or under diagnosis? 

Us2.ai: Echocardiography, or an ultrasound of the heart, is a doctor’s first tool of choice for the diagnosis of heart disease. But even though echocardiography is the safest, most common and most patient friendly imaging modality available, it’s still a slow, complicated process that is rate limited by the availability of cardiologists in larger hospital settings.  

First and third world countries alike face backlogs and are unable to meet current demand for echocardiography, much less provide increased access and widespread longitudinal tracking for early detection. The direct problem for the hospital management is the limitation of resources (e.g. cardiologists to see the patients and read the echocardiography), that result in long waiting lists adding more pressure to the healthcare system.

Moreover, a key reason for under-diagnosis of heart failure is the non-specificity of presenting symptoms and signs, necessitating objective diagnostic tests. The measurement of plasma natriuretic peptide (NP) concentration is recommended by international guidelines for the initial diagnosis of HF. However, NP levels are influenced by common comorbidities of HF, as well as a number of other cardiovascular and noncardiovascular diseases such as atrial fibrillation, advanced age, renal failure and obesity.6,7 

This makes echocardiography a necessary tool to confirm HF as well as to distinguish between types of HF. This is critical to avoid misdiagnosis and select appropriate therapies.1 The drawbacks of traditional echocardiography are that it is highly manual, time consuming, error-prone, limited to specialists, and involves long waiting times (e.g. up to 9 months in some areas of NHS Scotland). 

Artificial intelligence can help early initiation of key life saving therapies and lead to reduced hospitalizations

HT: How can the use of AI in cardiology allow for timely diagnosis of HF, as well as other diseases?

Us2.ai: Recent advancements in AI can allow fully automated, fast and reproducible echocardiographic image analysis; turning a manual process of 30 minutes, 250 clicks, with up to 21% variability among fully trained sonographers analyzing the same exam, into an AI-automated process taking 2 minutes, 1 click, with 0% variability.8 Such AI-enabled echocardiographic interpretation therefore not only increases efficiency and accuracy, but also opens the door to decision support for non-specialists.

A combination of circulating and echocardiographic biomarkers would represent the ideal diagnostic panel. Such combined interpretation of multimodal data was not possible in the past since blood-based and imaging-based labs largely functioned independent of each other. In the current era of linked electronic health records and picture archiving and communication systems (PACS) in many hospitals, the development of true “companion diagnostics” with combined interpretation of both blood and imaging biomarkers is possible. 

Moreover, advancements in medical AI enable deep learning models to be developed for greater diagnostic/predictive precision than ever achieved before. Automation of these algorithms, built into decision support tools for clinical application, has the potential to transform the diagnosis of HF and ensure quality and efficiency in the hospital’s HF patient care. 

Bringing heart care closer to home 

Finally, the availability of point of care (POC) testing for both NT-proBNP and echocardiography (using mobile echocardiography probes connected to handheld smart devices) can foreseeably bring these novel AI-enabled tools to the primary care or community setting. Indeed, the current COVID-19 pandemic has highlighted the urgent need for such point-of-care community-based testing, with the NHS Scotland Heart Failure Transition and Recovery Plan in response to COVID-19 stating that:

The COVID-19 pandemic has placed enormous pressure on NHS Scotland and the impact on heart failure care has been devastating… This has resulted in a large backlog of people awaiting tests (NTproBNP, Echocardiography) to confirm the diagnosis of heart failure and commence disease-modifying treatment… In order to facilitate safe physical distancing for patients and for staff and to minimise hospital attendances for patients… provision of community point-of-care diagnostics (NTproBNP, echocardiography) will be necessary… NHS Scotland Boards should give priority to developing hospital-based point of care diagnostic clinics… with a commitment to progressing to community delivered services at the earliest opportunity.

HT: How will the use of AI transform healthcare? 

Us2.ai: AI will support the: 

1. Reduction of hospital costs due to frequent re-hospitalizations.

2. Reduction of long waiting lists for timely, complex procedures such as echography

3. More efficient use of hospital staff and hospital resources (people management)

4. Creation of decision support tools to improve the patient and physician experience (since satisfied personnel is of the highest importance to a long term solution in any industry)

Where there’s a need, there’s a way

HT: Do you experience resistance from health systems towards adopting AI-based tools? 

Us2.ai: No. So far we have experienced strong interest in and acceptance of our AI-based tools.  The simple reality is that there is so much pent up demand for greater access to echocardiography that health systems have been eager to explore and invest in tools like AI that can boost productivity.  However, there is a great deal of cautious skepticism in the reliability of AI tools, and rightly so.  The burden is appropriately on us software providers to prove that our algorithms meet the accuracy and reliability necessary for patient care.

We overcome resistance through the same mechanisms that non AI-based tools must go through: clinical trials, peer reviewed publications, regulatory approval and case studies of successful customer adoption. It’s very trendy to want to disrupt these procedures and to want to change the industry. 

HT: What advice would you give to other companies wishing to transform traditional practice or methods? 

Us2.ai: Ignore being trendy. Focus on providing value to your customers and their patients, believe passionately in your mission and work incessantly.  The rest will work itself out.

Us2.ai successfully participated in Startup Creasphere, a leading digital health accelerator that strives to transform healthcare together with startups.

Carolyn Lam, PhD Director of the Clinical & Translational Research Office, National Heart Centre of Singapore. Professor of Cardiology at Duke-NUS Graduate Medical School. Leads multiple global clinical trials. Received the L’Oreal Women in Science award. Currently on the Editorial Boards of Circulation, and the European Journal of Heart Failure, with prior editorial posts at 10 other journals. Has published over 175 peer reviewed articles and books. Host of the weekly medical podcast Circulation on the Run, and co-host of the television show Body & Soul. PhD Univ. Groningen. Masters Mayo Clinic. Stanford GSB exec. program.

James Hare, MBA Co-founder and President of eDreams (2014 IPO), one of Europe’s largest e-commerce sites with a global presence in 44 countries and over €4.5 billion in annual bookings. Biz dev and marketing roles throughout US, Europe and Asia, with EFI, UbiSoft, Netscape and the World Economic Forum. MBA Stanford. BA Harvard.

References

  1. GBD 2017 Disease and Injury Incidence and Prevalence Collaborators. (2018). Lancet 392, 1859-1922
  2. Cook et al. (2014). J Cardiol 171, 368-376  
  3. Stewart et al. (2001). Eur J Heart Fail 3, 315-322  
  4. Roger et al. (2004). JAMA 292, 344-350
  5. Yancy et al. (2017). JACC 70, 776-803
  6. Maisel et al. (2008). Eur J Heart Fail 10, 824-839
  7. Madamanchi et al. (2014). Int J Cardiol 176, 611-617
  8. Us2.ai. (2020). Unpublished data on file