A deep dive into machine learning in healthcare

Healthcare in the digital age is an orchestra of people, machines, and data all working together in service of achieving a singular goal: health. Mostly unseen, but no less profound, are the underlying technologies that are transforming how we understand health and deliver care. Machine learning is one of them. 

Machine learning has recently been thrust into the spotlight due to the development of large language models (LLMs) and generative pre-trained transformer (GPT) technology. LLMs are essentially machine learning algorithms that are designed to comprehend, generate, and manipulate human language. For more, check out our companion piece on large language models in healthcare. 

In this article, we explore what machine learning is, unpack how it can be applied to the administration of healthcare, share examples of how health systems are putting it to use, and detail the critical considerations for its successful implementation. 

What is machine learning?

Let’s start with a simple definition. Machine learning is a branch of artificial intelligence that equips computers to learn from data to make decisions or predictions. 

Imagine teaching a child to identify different shapes. You might show them multiple instances of squares, circles, and triangles until they recognize the distinct characteristics of each shape. Machine learning follows a similar process, albeit on a much grander scale. It’s about computers learning from vast amounts of data, detecting patterns, and making predictions. The more quality data these systems digest, the more accurate and nuanced their predictions become. 

The power of machine learning in healthcare administration

By its very nature, healthcare is a data-intensive industry. According to the World Economic Forum, the average hospital produces some 50 petabytes of data per year. For purposes of comparison, that means the average hospital produces, on an annual basis, more than double the amount of data that exists in the entire Library of Congress (20 petabytes). And it doesn’t stop there. An explosion of smartphones, wearables, and connected medical devices has seen the amount of data generated in healthcare increase at a rate of 47 percent every year. The challenge? Some 80% of all data in healthcare is unstructured. That means it does not have a pre-defined data model or is not organized in a pre-defined manner, making it more difficult for machines to understand. It is estimated that a whopping 97% of all data produced by hospitals every year goes unused.

Consider the multitude of data sources that exist in healthcare, including patient medical records, claims documents, clinical notes, clinical trial results, consent forms, imaging reports, lab results, prescriptions, and waivers just to name a few. Hidden within this sea of information are patterns and insights that are not easily discernible by human analysis alone. Machine learning, with its ability to sift through and learn from these complex data sets, can uncover and surface previously hidden insights. However, the potential for machine learning in healthcare extends well beyond identifying patterns for disease diagnosis or treatment strategies. One of the most potent applications is in increasing operational efficiency within healthcare organizations of all shapes and sizes. 

From optimizing scheduling and managing patient flow to predicting which patients are likely to miss appointments, machine learning can help healthcare organizations run more smoothly and cost-effectively. Operational improvements such as these can lead to better resource allocation, improved patient experiences, and ultimately, higher quality care. 

In a world where machine learning is accessible to all, true differentiation happens with the ability to tailor solutions to specific patient needs, an approach that fosters cross-disciplinary collaboration, an emphasis on ethical use, and through a single, unified platform that integrates with existing workflows. 

How machine learning is used in healthcare

Today, machine learning is being deployed across health systems to drive efficiency and save money. The following are sample use cases where machine learning is helping health systems analyze and learn from the vast data sets that exist within them.

• Claims processing. One of the more tedious aspects of the administration of healthcare is processing billing and insurance claims. It is a complex process that is fraught with opportunities for human error. Machine learning is streamlining these processes by analyzing historical data to identify patterns and anomalies in claims submissions, which helps detect potential errors or fraud. Similarly, predictive models are being used to assist in estimating the cost of care based on factors such as the patient’s condition and recommended treatment plan. 
• Patient communication. Healthcare administrators are putting machine learning to work in understanding patient sentiment, preferences, and behavior by analyzing communication data across various channels, including email, text messages, and online portals. With insights from these analyses, health systems are able to tailor their communication strategies for different patient groups, helping to improve engagement, satisfaction, and overall health outcomes.
• Patient records management. Machine learning is assisting in managing electronic health records (EHRs) by automating data entry, flagging inconsistencies, and even providing care recommendations, including preventative care. This alone can save significant administrative time and enable healthcare providers to focus more on delivering care to their patients.
• Scheduling. Machine learning algorithms can analyze a multitude of factors, including patients’ past appointment history, no-shows, demographic information, weather conditions, and more to predict the likelihood of patients missing their appointments. Using these insights, health systems are able to proactively push out appointment reminders and generally better optimize their scheduling.

Key considerations for implementing machine learning in healthcare

Machine learning has immense potential to streamline healthcare operations and help deliver a better patient experience, but its implementation requires careful consideration. Healthcare leaders should keep in mind these key aspects:

Data privacy and security

In healthcare, machine learning applications often depend on vast amounts of personal health information (PHI), making data privacy and security essential. Not only is there a potential brand impact associated with a data breach, but there are also regulations such as the Health Insurance Portability and Accountability Act (HIPAA) that set strict guidelines on how patient data can be used and shared. 

Tip: One step to putting your health system on a path to compliance is to ensure that any data used to train machine learning algorithms is thoroughly de-identified and encrypted.

Quality and availability of data

In order to provide accurate predictions and deliver valuable insights, machine learning models require large volumes of data. However, it’s not just data, but high-quality, reliable, and accessible data that will ensure the model can operate as expected. Quality data is accurate, complete, consistent, and relevant to the use case. A diverse, representative data set, safely and securely extracted from multiple systems is key to maximizing the potential of a machine learning model.

Tip: Data should be consistently recorded and standardized across various sources, which includes things like using consistent terminology and units of measure.

Data freshness

Having accurate, quality data is part of the equation. However, in healthcare static data can quickly become problematic. The reality is that practices change, regulations evolve, codes are revised, and rules are rewritten. Machine learning models are only as current as the data they are trained on, meaning that a continuous or at least frequent updating of the data and the underlying models is required in order to maintain performance. For example, if a machine learning model used for coding medical procedures is not updated to reflect new procedure codes, it may lead to errors in billing or reporting.

Tip: Agile development methodologies can help address data freshness by allowing for rapid iteration and adaptation of machine learning models as changes occur.

Integration with existing systems

Building a comprehensive view of each patient starts with the ability to unify disparate data sources. Machine learning solutions need to integrate with existing systems like the EHR, practice management software, or billing systems. It’s important to plan and collaborate with IT teams for integration into disparate systems and to ensure minimal disruption to existing workflows. 

Tip: Consider solutions that have pre-built EHR integrations, proven interoperability with your source systems, and that are fully capable of working with your partner ecosystem.

What’s next for machine learning in healthcare?

There is little doubt, the rapid and increased use of machine learning in healthcare signals a paradigm shift. For too long, healthcare has been stuck in a one-size-fits-all approach, however, machine learning and other advanced AI technologies are already helping deliver a more personalized, efficient, and proactive healthcare experience. 

As leaders increasingly understand new technologies like machine learning, they can envision a future where technology, algorithms, and humans work together to deliver higher-quality care for everyone.