Machine learning, a high-trending topic in popular literature and news of late, is actually not a new phenomenon in computer science. Academic research around neural networks actually began as early as 1943, when McCulloch and Pitts wrote a paper on how neurons might work, based on a simple neural network based on electric circuits. The phrase ‘Artificial Intelligence or AI’ was first coined at a famous Dartmouth College conference in 1956!
AI has a long, chequered history in medicine. Its early applications started in the 1970s with biomedical AI-based systems like Internist-1, CASNET, and MYCIN. From being labelled a failure in 1973 to the success of the ‘expert systems’ market in the 1980s and the subsequent collapse of the specialized AI market, the interest in the application of AI in the medical field has waxed and waned.
What has changed now?
Over the past decades, computers have broadly automated tasks that programmers could describe with clear rules and algorithms. Recent advances in machine learning, as Jeff Bezos elegantly explained, now allow us to automate tasks where describing the precise rules is harder. In addition to faster and more powerful computers and data communication, another key contributor to the rapid recent progress in AI has simply been the availability of sufficient data to ‘train’ AI models. As a recent study pointed out [4], availability of key dataset capabilities has been the key accelerator in advances around AI.
For example, availability of key datasets caused ‘hockey-stick’ growth in drug discovery. In the 1990’s, application of AI in high-throughput screening (HTS) techniques were necessitated due to advances in combinatorial chemistry. Millions of novel chemical compounds were developed and needed to be tested in a relatively short timeline for viability as possible drugs, resulting in the both the availability of large databases and the need for AI-based techniques to analyse them. Simultaneously, advances in gene sequencing studies led to the collection of billions of data points. Availability of data and advances in computing, together, led to a rapid acceleration in AI-based drug discovery. For example, in 1994, the US National Cancer Institute developed a drug discovery programme for cancer and AIDS using predictive analytic and AI.
The hype
The combination of big data, predictive analytics, and machine learning is revolutionizing areas as diverse as stock and derivatives trading, marketing, retail, e-commerce, and mobile games. Healthcare is not far behind as an exciting domain for such solutions what with predictive analytics and machine learning becoming some of the most-discussed, and perhaps most-hyped, topics in the context of ‘digital health’.
According to IDC, a well-regarded analyst firm, about 30 per cent of healthcare providers will start using cognitive analytics, AI, and machine learning technology for predictive analysis by 2018. Cutting through the hype, such solutions do portend a revolution in specific healthcare realms like chronic disease prediction, prevention and management, drug discovery, improving patient care, and hospital administration. Specifically, let us try to sift through the hype in two healthcare areas where AI will likely have an imminent impact.
Sifting through the hype in chronic disease prediction and management
Following are some of the glaring gaps in preventive healthcare today.
n For most people, preventive health ends with just a regular health check
n The health check report, which often contains readings for 20+ parameters, is not easy to understand for most users
n The readings and results in a health check report primarily indicate prevalence or imminent onset of a condition (e.g. you are diabetic/pre-diabetic) and nothing about the future risks (e.g. your family history and lifestyle put you at increased risk of hypertension over the next two years)
n The inability to understand results and foresee risks, many who need help don’t visit a physician. A recent study carried out by healthi indicates that as many as 26 per cent to 28 per cent of users who went through an annual preventive health check and needed help, do not visit physicians.
Predictive analytics and personalized user engagement, based on machine learning, can help address many of these gaps to make preventive healthcare truly effective.
Addressing the gaps healthi believes that the applicability of such technological solutions does not entail replacing the opinions of qualified physicians or clinical practitioners, but rather assisting practitioners with analysis or engaging and empowering users to avoid chronic illnesses. Predictive analytics, building upon scientifically validated chronic disease risk prediction models, can help users see a complete picture of their current and future health.
Machine learning can help improve the efficacy of chronic risk predictions by constantly fine-tuning them and further honing their accuracy with each additional user. It can also help craft personalized care regimen for each user, taking into account their unique history, lifestyle, and preferences. Additionally, it will also help put them in touch with practitioners who can help them adapt to their evolving health status.
The ability to rapidly analyse very large and often sparse data sets is the foundation for the predictive learning and personalization pillars. A combination of these can make the preventive health journey personalized and ‘one-size-fit-one’, and thereby insightful, engaging, and effective.
AI in drug discovery
As with preventive health, one must take a balanced view on how AI can aid drug discovery. AI-based data analytics can revolutionize the various stages in the drug discovery and development process. However, mere data analytics cannot be a substitute for chemical synthesis, laboratory experiments, trials, regulatory approvals, and production stages.
AI can accelerate R&D efforts, minimize the time and cost of early drug discovery, and help anticipate possible toxicity risks or side effects at late-stage trials to hopefully avoid tragic incidents in human trials. It can help consider genomics and other research to generate ideas around therapies. AI-big pharma collaborations announced recently (e.g. Pfizer and IBM Watson, Sanofi and Recursion etc.) are likely looking to tap into such efficiencies.
Conclusion
AI had promised much and failed to deliver in its earlier incarnations. However, we currently appear to have a perfect collusion of factors, namely the availability of computing, networking resources and large datasets, which enable AI to live up to its potential for driving impactful outcomes in healthcare.
Organizations would do well to integrate an aptitude for AI into their growth plans, organizational design and product roadmap.
(The author is co-founder and COO, healthi.in)