Healthcare is packed with complex data stored in multiple places and evolving every day. That makes it a great target for the form of artificial intelligence known as machine learning.
Oxford defines machine learning as “the use and development of computer systems that can learn and adapt without following explicit instructions, using algorithms and statistical models to analyze patterns in data and draw conclusions.”
In recent years, machine learning has already proven useful in diagnosis and can aid in the efficiency of medical coding. But there are many other places where machine learning can be useful but hasn’t made any headway yet. Why is that?
Harshith Ramesh is co-CEO of Episource, a provider of risk adjustment services and software for medical groups and health plans and an expert in machine learning. We interviewed him to discuss why machine learning is so suitable for healthcare, how it has helped diagnose and coding so far, and most importantly, what’s holding it back in healthcare.
Q. You claim that healthcare is ideally suited to machine learning. Why?
A. Machine learning is a branch of artificial intelligence that uses data to imitate the way humans learn, continuously improving the performance of a given task over time. In healthcare, this technology is being used to detect patterns in patients’ health information and refine the algorithms to become more accurate as it learns from available data.
As more healthcare providers take on downside risks under value-based contract models in the coming years, it has become more important than ever to measure patient outcomes efficiently, accurately and cost-effectively. Machine learning is an important tool that providers can use to achieve this goal.
Healthcare is ideally suited to machine learning because of the exponential increase in the volume of patient data over the past two decades. Today, about 30% of the world’s data is generated by healthcare.
This is partly due to the widespread use of the electronic health record, which was first introduced in the 1990s. The digitization of patient information has not only increased the amount of data there is, but also made it easily accessible for machine learning applications.
In addition to EHRs, health data is also generated by a growing number of sources, such as medical devices, wearables, data clearinghouses, labs and supplier offices. This rich abundance of data is critical for machine learning models to become more accurate in predicting patient outcomes. This can help caregivers gain a more comprehensive picture of a patient’s health over time.
Healthcare data is also more objective in nature than data generated by other industries, making it particularly compatible with machine learning technology. This is due to standardized procedures, automated systems, medical coders, and expert physicians – all of which help remove as much subjectivity from the data as possible.
For example, industry has established standardized data sets for healthcare organizations to use, such as International Classification of Diseases (ICD-10) codes for diagnostic information or National Drug Codes (NDCs) for drug identification.
Regulations on how healthcare organizations can house and transport data in EHRs have also made it easier for machine learning-powered models to analyze the data, discover trends, and apply algorithms to improve patient outcomes.
Q. How can machine learning help healthcare organizations make diagnoses?
A. Machine learning has a variety of applications in the clinical space. One such application is predictive modeling, a widely used statistical technique that can be used to predict future behavior.
Predictive modeling allows health care providers to effectively predict whether a high-risk patient may develop sepsis or another type of complication after a procedure. This can help determine whether they want to take additional preventive measures to reduce this risk, such as calling patients for regular checkups or optimizing resources to target potential high-risk patients.
It can also support population health management by creating dynamic cohorts, which segment member populations based on a particular set of health problems or other type of pattern. These lessons can then be shared with care management teams, who then determine which interventions are most impactful for a given cohort.
Finally, machine learning models can help healthcare providers carry out clinical conjectures. This technology can be used to analyze diagnostic data to predict which patients need care most urgently and to identify gaps in their medical history.
Machine learning can also help healthcare providers determine whether a particular treatment is effective for a patient, for example by analyzing a patient’s complete health history to find the safest and most effective drug a doctor can prescribe based on the diagnosis.
Q. How can machine learning help healthcare provider organizations with medical coding?
A. Providers are often complete in their documentation processes, but it can be difficult to translate this data into just one of more than 72,000 ICD-10 diagnostic codes.
As vendor organizations strive to improve data quality, they can choose to leverage and scale AI technology to help improve the efficiency and quality of medical coding across the risk-adjustment continuum – prospective, concurrent, and retrospective.
Before and during the visit, machine learning algorithms can quickly analyze the patient’s medical information and present the healthcare provider with a real-time snapshot of the patient’s health.
Physicians can spend less time on burdensome administrative tasks and instead more time on providing targeted and timely patient care. In addition, prospective coding, powered by machine learning, can uncover chronic conditions that have been documented in the past, but not at the time of the visit.
Machine learning can intelligently and automatically parse unstructured information in the EHR to identify the most accurate code. For example, it can also be used retroactively to increase both the speed and accuracy of coding, saving time and costs for supplier organizations – allowing them to focus more resources where they are needed most.
This, in turn, helps groups of healthcare providers meet quality measures, track performance, and ensure patients are regularly assessed.
Q. What’s stopping healthcare from making more progress with machine learning?
A. The biggest contributing factor to healthcare hesitation to adopt machine learning is the barriers the industry faces to become more interoperable. Competition and the resulting lack of coordination between health systems has created many challenges.
From inconsistent technical standards to divergent health information privacy policies, from different approaches to obtaining patient consent to difficulties in coordinating key EHRs, there are many hurdles healthcare organizations must overcome in their pursuit of interoperability.
This creates a data gap between different EHR applications and networks, creating silos in the data that would inform the most urgent and impactful patient interventions.
In addition, the healthcare regulatory landscape is becoming increasingly complex, with rule revisions for government-sponsored programs taking place annually. This raises providers’ doubts about whether technologies such as machine learning can adapt to these constant regulatory changes.
There will always be skepticism about any emerging technology, especially when healthcare providers are offered one-size-fits-all black-box solutions that don’t effectively equip them to provide better care to their patients.
Technology vendors that offer solutions that leverage machine learning technology must be transparent in explaining how they can improve workflow efficiency and reduce administrative burden so that caregivers have more time to focus on delivering care.
Vendors must serve as a constant resource and partner throughout the implementation process and beyond, ensuring their solutions work continuously to better understand the provider organization’s member population and improve patient outcomes.