doctor healthcare

Todor Primov, life sciences product manager at Ontotext, discusses some of the ways artificial intelligence (AI) is changing the face of healthcare.

A common joke among certain circles is that AI can be defined as “whatever we haven’t done yet”. While the current popularity of AI among investors, startups and the wider media is unparalleled, we need to retain focus on what exactly AI means, and its specific sector-based applications.

In healthcare and pharmaceutical research, ‘AI-like’ technologies hold huge promise, but what are the problems these solutions are purporting to solve, and where are the business opportunities?

Sheer data volume

Data volume is a problem in the healthcare sector too. A recently published scientific paper highlights the scale of the problem that medical researchers face on a regular basis: scientists sifted through the genetic sequences of nearly 600,000 adults in order to find the genes of just 13 people who were resistant to a particular strain of disease. As the article notes, in the future 600,000 data points may not be enough to isolate a potential cure: researchers will need millions.

Given the scale of modern medical studies, it’s unsurprising that researchers are looking at advanced data analytics more closely to try and find efficiency savings. There’s a huge business opportunity here waiting to be seized: McKinsey estimates that applying big data strategies to pharmaceutical research could generate up to $100bn annually in cost savings in the US healthcare system alone.

Problems exist not just in the scale of data that is handled by medical and pharmaceutical organisations, but also with how it is organised. In medical fields the siloing of data is particularly prevalent, and this results in all the inefficiencies that one would expect. It is becoming increasingly hard for front-line doctors to pull all the information they need to treat patients, as that data is siloed across different medical data systems. A more wide-ranging and elegant umbrella solution is required.

The solution: more data?

What’s needed is a more robust system of metadata. Rather than undertaking a massive task of refactoring an entire data pool, companies are realising that a lot of work can be saved by adding a thin metadata layer over the top of the software stack, and then indexing that metadata layer (and the underlying information by proxy) with advanced database technology.

Natural language processing is an AI technology which has improved drastically over the last few years, and can now be used to identify and tag discrete entities within vast volumes of text. Using this technology, research and development experts no longer have to spend huge amounts of time poring through medical records and other text based data. The technology can isolate keywords and phrases like “pancreatic cancer” or “aspirin is a blood thinner”, tag those keywords, and break down the sentence into relationships to be saved in a specialised database called a graph database.

Graph databases have been around for a while, but are coming to the fore as a particularly useful mechanism for AI-enabled healthcare. They allow many-to-many datapoint analysis: in layman’s terms, this means that they can save the relationships between multiple data points in an intuitive web-like structure. This technology can therefore interlink various different data silos within large, data-intensive organisations in entirely new ways.

Applications and implications

Graph databases are increasingly being used by some of the world’s largest healthcare providers and pharmaceutical research institutions. On the front lines of healthcare this technology will make diagnosing conditions faster and safer. Doctors will be able to search their database for symptoms in a much more rigorous manner, because the data points will be connected more thoroughly. Through machine learning the computer system may even be able to infer a diagnosis using the information available, and suggest treatments. This may be some time in the future, but the progress towards this endpoint will always start with increasing the connectivity and accessibility of data.

When companies have a stronger grasp over their own data, interesting things can happen. For example, Ontotext’s graph database technology is used by one of the biggest biopharmaceutical companies in the world to enhance its drug discovery efforts by uncovering new, unseen connections within existing data pools. This company also uses graph databases to help with compliance: when regulators come knocking and request decades’ worth of paperwork, natural language processing and graph databases can pull together the relevant documents much faster than would have been possible before.

Looking to the future, personalised medicine is a huge application area of this technology. As we become more adept at reading and processing patients’ genetic information, we need more data analysis tools to make that information actionable. Graph databases can provide this: they’re much more robust and powerful than existing databases, and could power a future era of drug discovery and manufacture tailored to individual genetic predispositions and weaknesses.

The health industry spends vast quantities of money on drug development and research. While graph databases aren’t exactly the most glamorous application of AI technology, they have the capacity to deliver real efficiency savings, and, in this market, efficiency savings of even a few percentage points can amount to billions of dollars.

If you’re interested in finding out more about AI and healthcare, subscribe to receive our latest print magazine for free, which focuses on the topic of artificial intelligence.

Show Comments