How Big Data and AI Will Combine to Transform Healthcare

The AI revolution is being driven by data, so those industries that produce the most of it also have the most potential for disruption. That’s especially true when, like in the healthcare sector, most of that data has to date gone unutilized, says Invisible Technologies’ GM of Healthcare, Rishi Madhok MD.  

“If we properly harness the data that healthcare generates, the sector could become unrecognizable within the next decade—not only when it comes to diagnosis, but also when it comes to administration and patient experience,” he explains. 

“However, for that to happen, healthcare organizations really need to think about how they can collect and standardize all the information they produce to create interoperability.”

An Incredible Amount of Unstructured Data

The healthcare sector now accounts for as much as 30% of all information generated globally. And the amount of data it generates is expected to expand at a Compound Annual Growth Rate (CAGR) of 36% between 2018 and 2025, according to an RBC Capital report. 

Dr. Madhok notes that the explosion of new medical-focused technologies, such as wearable medical tech, electronic records, and genomic data, is contributing to the exponential rate of data growth. However, despite this strong tech focus, he also observes that most healthcare data remains unstructured and disparate. 

“The vast majority of X-rays, scans, and medical records taken around the world are siloed in different formats and closed systems which don’t talk to each other,” he notes. “This means they can’t easily be analyzed.” 

“But within this information, there’s a wealth of patient engagement and medical information that could be leveraged to create AI-driven solutions that could transform diagnostics.”

Not All About Diagnostics 

This fact has already led to a wave of MedTech companies looking to use generative AI for imaging and other detailed tasks. 

For instance, MedTech companies are using AI to detect early-stage cancers and other abnormalities not readily visible to the human eye, substantially reducing error rates and directly saving lives. But these have often been trained on relatively small datasets—at least when compared with the enormous amount of data available worldwide.  

“There are 95 million MRI scans undertaken globally each year, including 40 million in the United States alone,” Madhok notes. “That is an incredibly large dataset, and within it, we would see virtually every disease from the common to the ultra rare.”

“If used properly, this could provide a comprehensive basis for training AI systems to recognize a wide spectrum of medical issues. The untapped potential for this when it comes to saving lives is astounding.”

But Madhok also believes that AI’s potential to transform the “process” side of healthcare could be equally as powerful - partly because healthcare can be expensive to access and provide. That’s especially true in the United States, which spends far more per capita on healthcare than any other high-income country.  

A 2021 McKinsey study found that almost half of this total cost did not relate to treatment or diagnosis but to administration, with hospital administrative costs totaling US$250 billion in 2019 and clinical services administrative costs coming in at $205 billion. This represented 21% and 27%, respectively, of total national health expenditure (NHE).

“The high cost of healthcare is largely the result of inefficiencies,” Madhok notes. “While it’s true that the health systems are often overburdened with too many patients and not enough staff, there is also a massive administrative burden”

“In the United States, this burden is often the result of a sometimes complex interplay between multiple stakeholders, including insurers, healthcare providers, pharmaceutical companies, and the government.”   

A Patient-First Future: But How to Get There

Madhok sees a future, however, where AI intervenes to put the patient at the center of the healthcare experience.  

“AI can change how we engage with technology,” Dr. Madhok notes. “We need to think past the chatbot to lead a patient into the best possible healthcare experience based on data.” 

“AI gives us the potential to anticipate patients’ needs rather than first requiring inputs. This gives us the power to make the patient experience proactive.” 

“AI could be used to guide the patient through their entire healthcare journey, flagging any warnings, providing preventative advice, and making sure they receive appropriate care.”

If this seems a pipedream, Madhok notes that it doesn’t need to be that far into the future at all, given that most of the technology to enable it already exists. Large Language Models (LLMs) could be supplemented by a process known as retrieval augmented generation (RAG) to incorporate real-time medical information. This approach involves using external data sources, like medical databases, to augment AI models, allowing them to access and use relevant information for more accurate and contextual responses. This would allow AI systems to guide patients through their healthcare journey with precision and context.

But not all AI-related tasks in the healthcare system need to be so grand, Madhok notes. There are smaller tasks along the way to achieving it that AI could be applied to for almost immediate results.

“In areas such as revenue cycle management and patient navigation, many enterprises could already apply AI at relatively little cost,” he observes. “There could be rapid gains in operational efficiency and cost reduction, while patient outcomes could improve markedly.”

“Those organizations - such as insurers, hospitals, and medical organizations - that choose to do this are likely to find they give themselves a competitive advantage over their rivals very quickly, and one that could become insurmountable as they continue down the AI pathway.”

Coming Back to Data

However, Madhok also says most healthcare enterprises need to do some serious work with their data before this can happen.

“They first need to invest in interoperability and reconcile their data,” he notes. 

While AI and automation can be used to do some of this, technology can’t do everything. Humans will still be needed to carry out non-standardized tasks such as deciphering handwritten notes or dealing with other inconsistencies. They will also need to get their data ready for training - something that involves having subject matter experts label or ‘tag’ data so that it’s understandable to machine learning algorithms and can be used to teach them how to make accurate predictions or decisions.

“These can be time-consuming and expensive tasks, especially if they’re carried out in-house. But they’re also necessary steps on the road to AI enablement.”

“That’s why it ultimately makes sense to outsource them to an experienced third party such as Invisible, which can make it happen in a very quick time and at a fraction of the cost,” Dr. Madhok concludes.