May 6, 2022

Are AI Tools Changing the Face of Healthcare Delivery?

By Swetha Raman Chakravarthy, MD

Recently, a UK-based health tech firm announced a partnership to work on AI-enabled Cardiac Ultrasound tools that can be used in accessible care settings. Caption Health and Ultromics jointly launched the EchoGo, that uses Machine Learning (ML) and Artificial Intelligence (AI) to provide automated analysis of left ventricular volumes, ejection fraction, cardiac strain, and diagnostic support. This cloud-based tool could  potentially  be  used  by  primary  care  physicians,  located anywhere, to diagnose cardiac failure and coronary artery disease faster and earlier in the course of the disease. Hence, avoiding a traditional on-site solution.

Similar to diagnostic tools, we have risk assessment tools like the Retina Risk which was developed by a couple of doctors from Iceland. This app screens patients for Diabetic Retinopathy in Primary Care, and helps people determine their individualized risk of developing diabetic retinopathy or vision loss using an algorithm. It has proved to be handy for a family physician, who is aiming for preventive care for their diabetic patients.

Journey of AI began in the 1950s when it was called Expert Systems and was incorporated into clinical settings only in the past few decades. The recent pandemic saw a sudden surge in the use of AI and its applications in healthcare, ranging from early intervention analytics, clinical decision support, specialist collaboration, to process improvements for providers and patients.


Currently there is a parallel growth in population and diseases, which is an enormous challenge for healthcare systems. Health inequity is increasing due to decreased access to specialist care for all populations and communities, especially the rural areas. There is a rise in healthcare costs and workforce is struggling to meet all needs of their patients. Also, more complex health data is being generated from different sources like Electronic Medical Records, wearable technologies, mobile health apps, and medical devices which is underutilized. In 2020, this volume hit 2000 exabytes and is expected to increase by 48% every year. This big data is chaotic, distributed, and voluminous but has the potential to yield valuable insights into an individual’s future health status.

Predictive and prescriptive analytics, modeling, and acquiring pattern- based insights can be utilized by all ‘players’ in primary care, which includes patients, healthcare providers, researchers, administrative staff, and medical education.

  1. Diagnostic and treatment decision support: Like EchoGo and Retina Risk there are more tools available in the health IT landscape.
  2. Population health evaluation: Secondary use of data includes information extraction and description that could provide better insights about a community’s population health. For example: Based on available EMR data, we could use an AI algorithm to detect dementia and its overall prevalence in the community.
  3. Prediction of diseases: Individual patients’ health data can be analyzed to predict their future health outcomes. For example: During EMR entry of a patient’s health information, an algorithm could run in the background to detect their risk for obesity, and system can alert both the provider and patient about their impending risk.
  4. Research: EMR data contains structured and unstructured components which can be used in a de-identified form for further research studies. For example: Clinics could de-identify data and study the effectiveness of cholesterol lowering drugs.
  5. Knowledge base and ontology construction: Ontologies are key enabling technologies for the semantic web. The provision of customizable, computable health information through EMR interfaces and semantic web technologies can transform the way healthcare providers will leverage technology.
  6. Practice optimization and improvements: Automating processes and improving efficiencies can also be driven by AI. Few components include improvement of patient engagement, patient satisfaction, appointment scheduling.
  7. Policy decisions and evaluation: A downstream evaluation of data can be conducted to reassess different policy decisions made for primary care clinics.


AI and ML terms are now widely used in healthcare, and it seems to be at the peak of a technology’s hype curve. However, the implementation of AI in healthcare system requires more forethought, as it is a highly sensitive and complex system. Primary challenges with AI in healthcare include:

  • Patient privacy: With big data coming from multiple sources, patients might not be aware if their information is in the AI’s dataset. So, robust frameworks should be in place at multi- stakeholder levels.
  • Standardized data: For the machine to learn and adapt from patterns of data, there must be a common standard to capture data We could have a standard EMR system across all primary care clinics, containing critical information in structured formats.
  • Data bias: AI-ML depend upon content in the datasets. For a fair prediction it must represent the real scenario of a large population. Otherwise, there could be a minority bias or socioeconomic bias where AI is skewed towards one perspective.
  • Reliability and accountability: If a patient’s treatment went downslope due to an AI’s decision, who should be held accountable- the doctor or a machine?
  • Data security: Transforming technologies and interoperability has made it easy to share, but is it always secure? Personal data including one’s genetic data can be shared over emails without knowing who the end-user is!
  • Data scientists: We do not have enough workforce who ace at data science and understand the contextual aspect of clinical medicine and its operations for an effective application.
  • Varied data: Primary care data is very varied and contextual. Extrapolating an algorithm that works in a specialist office to a primary care might not yield the best prediction.
  • Black box: AI tools that give very little rationale for their decisions will affect transparency. AI-ML uses computation that goes through hidden layers before providing an outcome. If providers are unable to explain this process, things could go awry.

Hence, striking a safe balance between risks and rewards for a holistic approach to AI-assisted care is vital and it requires brainstorming from healthcare providers, technology developers, regulators, patients/consumers, and policymakers. We need AI guiding principles from the Government of Canada for Healthcare for a safer implementation in primary care.