Machine Learning in Healthcare:
Fundamental Challenges vs. Immense Opportunities
Advances in healthcare have delivered numerous benefits to society, from increased life spans and reduced unnecessary disability to improved quality of life and health equity. Yet these improvements also put a heavy burden on the economy — the United States is projected to spend no less than $6 trillion, or about $17K per person, on healthcare in 2027.
Could AI and machine learning deliver healthcare advances while bringing down costs and improving affordability? We have reviewed the current state and scope of machine learning in healthcare, looked into the challenges organizations in healthcare face today and the benefits they expect to reap while implementing specific machine learning use cases.
Machine Learning for Healthcare: A Different Story
The US healthcare system generates no less than one trillion gigabytes of data every year, which, given AI’s ability to self-learn on large quantities of data, is good news. Both doctors and executives hope to take advantage of AI to discover and test new drugs faster, diagnose more accurately, and provide individualized care more efficiently. Yet healthcare is fundamentally different from other industries, and it may be more challenging to apply machine learning in the medical field.
The first problem is our understanding of disease
Any condition is a sum of multifactorial processes that we may or may not understand. We risk pushing vague or entirely wrong data to Ml algorithms, thus putting a patient’s health and well-being in danger. Built on a false premise, any ML solution will cause more problems than provide benefits.
The second problem is our false sense of security about the impact of interventions
We expect a well-tested system to perform as it should, yet a human body can act and react in the most inexplicable ways, which leads to medical errors and misdiagnosis. This brings into question the entire paradigm of how ML solutions are trained and evaluated (i.e. on a comparatively small, pre-defined dataset). Extensive cross-validation, monitoring, and fine-tuning of healthcare AI are a must.
The third problem is data
Despite the sheer amount of data generated in healthcare, data collection and processing remain a problem. In fact, most of the generated data is siloed and is not accessible for analytics. For instance, organizations collect administrative data like insurance claims and performance reports, which is unstructured and should be converted, but often miss insights-rich clinical data. Medical data makes for a challenging analytics environment.
Unlocking the potential of machine learning in healthcare is also challenging, because:
Data quality is often lacking, both in terms of representativeness and scale, which leads to wrong conclusions (i.e. cough is the result of respiratory tract infection only)
ML solutions operate as a black box, which means that results may not clear and transparent, and can be hard to reproduce
It is expected that the results delivered by ML should align with the clinical consensus, which blurs the line between novel insights and erroneous findings
ML solutions can be biased even if trained on a quality dataset.
All of these factors make it difficult to design, build, evaluate, and apply machine learning solutions. Organizations need to approach AI transformations in healthcare thoughtfully and with caution.
Advantages that Machine Learning Offers to Healthcare
Despite the limitations, AI and machine learning will redefine health and healthcare in one way or another. AI’s capacity to self-learn and identify patterns in data that humans cannot realistically detect is golden.
Intelligent solutions powered by machine and deep learning can help make more informed decisions (automation plus advanced analytics), on the one hand, and accurately detect anomalies (disease screening and diagnosis), on the other — and that are just two major areas to take advantage of AI. Others are administrative efficiency and FWA (i.e. fraud, waste, and abuse). Let’s look at these closely.
The consensus is...
The implementation of AI and machine learning can significantly improve healthcare delivery and administrative efficiency by:
Automating routine transactions done by clinical and non-clinical workers like managing regulatory documentation and fraudulent claims
Standardizing and streamlining performance metric reporting between 3,000+ care provider systems and insurers
Reducing administrative complexity through faster and more flexible interaction between provider systems and payers that process billing and insurance-related information
In other words, AI reduces bureaucracy that constrains both clinical labor and administrative labor. For example, John Hopkins now relies on an automated, AI-powered hospital control center that helps bed managers assign beds 30% faster — all achieved by cutting red tape from the picture.
To sum up
The impact of AI in healthcare goes beyond handling EHRs, healthcare analytics, and health prediction. Both artificial intelligence and machine learning are poised to redefine healthcare, first and foremost, by increasing its affordability.
A combination of better decision-making (both clinically and business-wise), more accurate diagnosis, improved administrative efficiency, and reduced fraud and waste that AI brings to the table drives productivity improvements, which, as projected by McKinsey, will save from $280 to $550 billion by 2028.
Hopefully, AI and ML will help the US healthcare system deliver more for less, radically changing the perception about it as of the most inefficient medical system in the world, based on health care spending and outcomes.
Thank you for reading!
Feel free to rate the article here: