Can AI alleviate the challenges faced by broken healthcare systems ?
THE CHALLENGES OF MODERN HEALTHCARE SYSTEMS
Healthcare systems all over the world are experiencing tremendous challenges.
By 2030, 1 in 5 Americans is projected to be in the age of 65 years or older. Despite huge healthcare spendings (18% of the GDP) , almost 28 million are still uninsured.
In developing countries, access to healthcare is inequitably distributed. Medical professionals do not want to work in villages, and absenteeism is high. Specialised care is available only in cities.
Let's take a look at how artificial intelligence can make a difference in the field of healthcare and the implications of an AI enabled healthcare infrastructure.
Article in the guardian newspaper about the essay written by OpenAI's GPT-3. Link: https://www.theguardian.com/commentisfree/2020/sep/08/robot-wrote-this-article-gpt-3
In 1997, IBM’s Deep Blue machine defeated Gary Kasparov in chess. In recent times, artificially intelligent machines have evolved to process natural languages, interpret human speech, visual images, drive cars and much more. Here’s an article that talks about key events during the evolution of AI.
HUMANS & AI : A COMPARISON
Strengths
Broad, contextual and abstract thinking, self-awareness, sense of ethics
Weaknesses
Maintaining objectivity, intense focus on repetitive tasks.
Strengths
Objective, Accurate, and Consistent decision making
Weaknesses
Closed-boundary ‘thinking’, No self-awareness, No emotions
As we can see, humans and AI are not competing, but have entirely different skill sets. This makes AI ripe for complementing our skills and compensating for our blind spots at the workplace. Here’s a short article from
HOW CAN AI HELP HEALTHCARE ?
AI has great potential to deliver equitable healthcare around the world. AI enabled telemedicine can make healthcare accessible in places where medical infrastructure is lacking. Machine Learning algorithms are being used to aid in disease diagnosis and pioneer new treatment methods, reduce preventable deaths, assist in drug discovery, and much more.
Medical Image Analysis
Medical image analysis is one of the most flourishing areas of AI applications.
These include:
Detect disease in medical image
Segment organs, disease lesions, etc.
Electronic Health Records
Electronic Healthcare Record is a rich source of patients’ health status. AI can help doctors with summarising EHRs, predicting risk for certain disease, or suggesting diagnosis and treatments.
Drug Discovery
Developing a new drug is a long and costly process, but AI could change the landscape profoundly. One successful example is AlphaFold, which is an application of deep learning to identify the 3D structure of drug molecules.
Pathology analysis
AI can assist pathologists in analysing human biological samples through image analysis using machine learning algorithms. Pathologists can upload high-res digital slides of the bio samples to the AI based disease diagnostic tool.
AI models can use historical data of past epidemics and local geographical and demographical data to estimate the probability of occurrence of the next epidemics or pandemics in a certain region.
Some diseases require customised treatment for each patient as a one-size-fits-all model treatment does not work. Doctors use trial and error method currently to find the appropriate treatment for the patient. AI models can make this process much faster with the best treatment recommendation.
Millions of people around the globe have poor in-person access to medical care. Even in developed countries, lab tests and in person doctor visits are very expensive. Telemedicine using real or AI proxy doctors can help in reducing these costs substantially.
Ambient intelligence refers to using sensors and AI enabled systems in hospitals and other care giving spaces to assist human care providers. AI can ensure that shortage of human staff or negligence on their part do not end up costing the health of the patient.
It is important to ensure that while optimizing algorithms to make the healthcare system more efficient , we do not lose empathy for the patient. With the principles of Human centered design, we can design AI applications while ensuring that patient well-being is the top priority of the healthcare system.
Human-centered design (HCD) helps us create positive and enduring change in this rapidly altering landscape of healthcare. HCD puts the human first by
Generating solutions that are desirable to people by meeting their needs first and only as a second step look if it's technical feasible and economic viable
Guaranteeing privacy and security of healthcare data
Make sure no segment of the population is discriminated/excluded by algorithms
Read more about human-centered design in healthcare AI
Unlike other industries healthcare tech faces higher scrutiny and skepticism from governments and the public. It is very important to ensure that bad actors do not have access to patient data. These are some of the big challenges in the healthcare AI domain. The cost of failure is higher and it becomes crucial to ensure that in an effort to use technology to deliver better healthcare we do not exclude anyone.
Algorithms perform well in academic settings but often poorly with live patient data. This can be due to a lot of reasons. Often datasets used to train models are outdated as patient characteristics shift over time. It is also difficult to compare models as they’re trained on different populations with different target metrics. Sometimes models fit confounders (non relevant signals) than true relevant signals.
Ensuring patient privacy and data security is of utmost importance in this domain. Currently different methods employed to protect patient privacy include facial blurring and body masking (for images), differential privacy, federated learning, and homomorphic encryption. Despite these precautions, patient data is still not 100% fool-proof against illegal usage or selling by unauthorised people.
Federal and State legislation and regulations lag far behind the recent technological advances made in artificial intelligence in healthcare, particularly when it involves privacy issues and health data. Of the 50 states, only 4 states passed any legislation related to AI in 2021. At the federal level there is no definitive passed legislation by Congress on AI.
Patient data is often siloed in different systems such as archival systems, pathology systems, EHRs, and insurance databases. Many of them also have independent coding formats which makes it difficult for data engineers to convert all the useful data into a uniform format to feed into the machine learning model.
AI based healthcare innovations have flourished in the past few years. Yet, we are still in uncharted waters with respect to applying AI based healthcare application for the masses. The pace of policy making lags behind the pace of AI innovation. AI systems are only as good as the quality of input data fed to them, and many a times, their outputs end up reflecting the real world inequalities back to us. These factors make it all the more important for us to tread cautiously while deploying large scale AI tools.
Ref: A brief article on the state of AI regulations passed in the US.
Who is responsible when an AI system makes a fatal decision?
Is it fair to track patients longitudinally for the sake of building more accurate ML models?
Should AI be trusted to make medical decisions for humans?
The road ahead for AI in healthcare is exciting. Venture capital (VC) funding for the top 50 firms in healthcare-related AI approached $8.5 billion in 2020. Big challenges lay ahead for policy makers to catch up with the rapid technological developments in this space. Interdisciplinary collaboration between technologists, care providers, policy makers, DEI professionals, and most importantly patients is the way forward to ensure that the hi-tech future of healthcare is fair for all.