Artificial Intelligence (AI) is ripe for development and investment for health information management (HIM). The major initiatives towards Interoperability and AI is just beginning in the US. The National Institute of Health has invested $1.5B into the All of Us Research Program, which invites participants across the country to share their biology, lifestyle and environment. Aggregated data sources like this one promise to enable AI technologies to improve diagnostic accuracy, clinical and operational efficiency and the overall patient experience.1
While these innovations are promising, they are also in a preliminary phase for HIM applications from an accuracy and reliability or practical use perspective. AI has been successful for other industries like retail for several years, where it is used for supply chain planning, demand forecasting, customer intelligence and more.2 Automation works well for retail because the data is consistent and abundant, so machine learning algorithms can constantly improve performance. But that’s not the case for healthcare—an industry where regulations are constantly shifting, and data is sensitive and disparate.
Why AI is Tricky for HIM
1. Data Sets Are Not Standardized
Before AI systems can be deployed for HIM, they need to be “trained” through data that is generated from clinical activities, such as screening, diagnosis and treatment assignment, so that they can learn to recognize similar groups of subjects, associations between subject features and outcomes of interest. 3
Health data technology was built on legacy systems (paper records) that were only mandated to go digital within the last decade. Throughout the shift from paper records to EHRs, data sets were not completely standardized. Even something as simple as entering the patient name as Henry Smith vs. Henry R. Smith could confuse AI. While extracting handwritten patient files or PDFs is tedious for us, it’s downright difficult for AI with all the uniqueness that would need to be learned and constantly adapt to change.
2. Shifting Quality Measures
Healthcare is a heavily regulated industry. Even as aggregated data sources become standardized and more commonplace (which they aren’t currently), a moving regulation yardstick could cause significant issues considering AI algorithms “learn” from foundational data. As regulations change, AI could create a breeding ground for error and compliance issues. These complications could also impact quality measures that result in financial repercussions, since reimbursements are based on annual benchmarks for quality measures. If you’re utilizing a highly customizable process like AI to automate, every year you would need to adjust the AI application to meet that year’s quality measure, and the AI would need to relearn the goal to achieve. This could result in significant extra downtime for your AI support team and HIM staff.
3. Laborious Human Oversight
A major limitation to deploying AI for HIM is the accessibility to reliable data. AI applications for healthcare can only automate certain functions. For example: in filing/indexing a document, AI may be able to realize a few key items such as patient identifiers and document type (like lab results) and file to the proper folder within the EHR. However, AI will struggle to decipher items that fall into the exceptions folder. For example, an AI will be challenged to handle:
- Illegible faxes and scans
- Abstracting flat data from documents
- Structuring data in the EHR
- Data matching between organizations
A hands-on approach to ensuring the foundational data is available and spotless is crucial for gaining the most benefit and return on investment from. If the foundational data is unorganized or limited, even the most advanced data analytic tools may simply get you to the wrong outcome faster.4 With the sheer volume of mission-critical tasks on providers’ shoulders, it’s tough to devote the labor or expertise to implement, monitor and adjust AI.
Preparing for AI
Down the road, HIM may be highly automated. Many forwarding-thinking organizations are preparing for the future by opting into aggregated data pools and standardizing data sets to be interoperable. However, current interoperable platforms suffer from inconsistent or inaccessible data, and highly skilled human judgment is needed to fill these gaps through:
- Patient matching and data clean up
- Syntax error research and correction
- Registry error research and correction
- Targeted data curation
While organizations are saying “yes” to technology, they may also be missing important data within their organizations with this technology. Opting into AI now may only be applicable to the very largest organizations with the budgets, knowledge and manpower to support and manage it. For essentially all other organizations, AI needs to be approached with clarity and caution and have a clear understanding of the needed initial and ongoing investment.
At DataFile, we have the human expertise to fill these gaps. Whether it’s basic filing and indexing, data abstraction and entry into the EHR, closing the referral loop or order management, we have a team ready to handle these tasks quickly, accurately and proactively.
If you’re looking for an “easy” button to both handle your administrative tasks now and prepare your data and existing technology for the future — reach out to DataFile. We can help you offload the burden of tedious data-related tasks such as filing records, releasing information and obtaining prior authorizations. With a 99.97% accuracy rate (far better than competing AI technologies that tout up to 80% accuracy after years of training 5), we can optimize your back-office on your terms, allowing you to spend more time on care while creating a solid foundation for AI and your organization’s future.
(1) Health IT Whitepaper https://www.healthit.gov/sites/default/files/jsr-17-task-002_aiforhealthandhealthcare12122017.pdf
(2) The coming AI revolution in retail and consumer products https://www.ibm.com/downloads/cas/NDE0G4LA
(3) Administration UFaD. Guidance for industry: electronic source data in clinical investigations. 2013 https://www.fda.gov/downloads/drugs/guidances/ucm328691.pdf (accessed 1 Jun 2017).
(4) How is Artificial Intelligence Working for Health Care https://www.optum.com/resources/library/ai-working.html