Machine Learning Can Solve Healthcare’s Data Problem – If It’s Implemented with Care

107
abstract defocus digital technology background,represent big data and digital communication technology concept

By Sachin Patel, CEO, Apixio 

The healthcare industry has the wealth of data needed for innovations like AI and machine learning to solve historic pain points. Between notes on patient visits, claims and compounding documentation built over years of care, all of the raw ingredients exist to inform better, more efficient decision-making for diagnosing and effectively treating patients. 

While this data has the potential to improve critical aspects of healthcare, the process of inputting it and extracting valuable insights has a long way to go. A new study from JAMA Network Open found that clinicians spend an average of over two hours in EHR systems every day, making data input a top complaint of primary care doctors, and reinforcing a persistent theme from the provider community over the past decade.

New technologies have the power to help clinicians use their time more efficiently, but they face problems of adoption, consistent engagement, and clear, demonstrated value to both healthcare providers and patients. 

Here are three ways that machine learning can help solve healthcare’s most potent pain points while avoiding the common pitfalls that threaten the adoption of new technologies:

1. Machine Learning Can Help Doctors Optimize their Time 

With physicians spending so much of their time on inefficient, manual data input, healthcare providers have explored alternatives like shifting labor to an outsourced partner or having another provider or staff member perform these activities in-house. These alternatives can be costly and substantially increase the likelihood of errors. Importantly, patients also receive less attention as a result of this approach. Details like past prescriptions need to be well-organized so they are not missed in a new treatment plan. Nuances like patient preferences and fears need to be well-documented, and doctors run the risk of not understanding notes written by someone else during the crucial moments of a patient visit. 

Rather than removing the doctor from the role of data implementation, machine learning features like processing automated dictation can help them stay in control and waste less time.  As clinicians interact with the machine learning tools, patterns are learned allowing for identification of what information is most critical to display – in other words, the “art of medicine” is preserved for each physician. This information can then be dynamically rendered for an easier workflow experience. 

With clear time savings on input and less time spent deciphering EHRs, machine learning-powered technology can quickly show their value and earn engagement, which will increase long-term adoption rates.

2. Machine Learning Can Organize Decades of Unstructured Data

While 20% of healthcare data is structured, its rigid format often fails to capture the nuances of patient visits. The remaining 80% of healthcare data is unstructured, consisting of PDF charts and forms, physician notes, and other scanned documents. Unstructured healthcare data has been historically difficult to combine with structured data, and doesn’t allow for quick, easy digestion of important information during a patient visit. 

Freehand notes, years of different compliancy standards and the unique styles of all parties inputting data has led to EHRs that are all over the map. Manually deciphering diagnoses and past treatments in different formats is costly both in time and outcomes. As such, the risk of missing critical details is high, leaving doctors to rely on the latest patient narratives to fill in important information needed to correctly identify and treat.

Machine learning has the capability to transform historically unstructured data into accurate, digestible and actionable information for clinicians. It can also provide more efficient systems of input so that future data is structured and consistent – meaning, more accurate and easy to work with – while still containing the vital nuances of patient experiences. While regulatory restrictions still limit access, machine learning provides easier ways to share information across platforms. As the legal landscape catches up to technology solutions without compromising patient privacy, machine learning can transform these two types of data into one format to drive more efficiency and accuracy for clinicians and patients. With a high level of trust established, transfer of key information can be done in an implicitly non-PHI (protected health information) fashion. 

3. Machine Learning Can Fill Critical Gaps in Care

Machine learning doesn’t just have the potential to allow physicians to spend more time with patients and less time at the computer. This technology can learn how physicians document and identify missing or incomplete diagnoses or even identify potential conditions that the physician was unaware of through the ingestion and machine reading of labs, biometric data, and the unstructured medical text of a patient’s medical history. Finding previously unknown or undocumented conditions and loading them into the physician workflow enables more effective and timely care. 

The combined potential of a skilled doctor and machine learning-powered technology that is also focused on accurately diagnosing the same patient can have a profound impact on the quality of care. A JMIR study investigated how machine learning can help healthcare providers measure the quality of care for heart failure patients. The study found that the machine learning application used, compared to the hospital’s manual review process, identified mentions of left ventricular ejection fraction (LVEF) with a 97.8 – 98.6% success rate. This study showed that the risks of human error due to the intensity of care delivery are reduced significantly with utilization of machine learning applications. Machine learning means more lives saved through better healthcare outcomes. 

How to Improve Machine Learning Adoption in Healthcare 

While the applications are clear and abundant, machine learning technologies often enter the market without extensive knowledge of the healthcare industry. Ignoring years of context and history can delay the adoption of healthcare innovation. 

Without carefully incorporating this history, and comprehensive models that factor in unique nuances like the habits and characteristics of the populations in any given area, machine learning tools threaten to move healthcare innovation backwards by burning providers with less accurate diagnoses and forcing them to adopt new habits, only to drive worse outcomes.  

It is important to approach healthcare innovations with care, rolling them out after extensive testing and providing clear, proven solutions that demonstrate an understanding of pain points on all sides. That means more efficiency for doctors, better organization of both structured and unstructured data for more accurate and digestible EHRs, and insights that help fill critical gaps in diagnoses for patients. 

Machine learning technologies can save lives if they’re implemented with the same level of care patients expect from a valued healthcare provider. 

This site uses Akismet to reduce spam. Learn how your comment data is processed.