In Clinical Documentation Integrity, a primary focus is which technology is the most beneficial – Natural Language Processing (NLP) or a deeper learning Artificial Intelligence (AI). There are different avenues healthcare systems can take when determining how to structure the technology needs for CDI programs. Consider if you want the assistance from information already written and key words noted by NLP or rather, suspected conditions and queries prompted based upon diagnostic pathways or algorithms noted by AI, or a little of both. Most healthcare systems have NLP that focuses on what the physicians have written, medications, labs, and imaging. But many times, it cannot read or recognize all connotations of negative words and may flag a review or prompt coding for an inaccurate diagnosis.
According to the ACDIS podcast on January 5, 2022, the ethical concern regarding healthcare systems using AI in clinical documentation is that there are frequent suspected conditions that could be considered leading and prompt an accidental capture of an invalid diagnosis. A primary issue with AI technology is its inability to recognize key phrases that may help identify what is considered supporting documentation and could confirm the specificity of certain diagnoses. When suspected diagnoses are suggested, it should be based on all hospital and visit encounter notes, labs, flowsheets, pharmacy, therapy, and other healthcare notes pertinent to the patient’s healthcare record from other medical centers. Even if we add all types of keywords, phrases, or algorithms within the technology, it will still not read between the lines like a human with clinical knowledge can. Furthermore, in our current state of focusing on health equity and equality for all, it is important to note that there is a recognized bias in AI technology due to costs and the ability for certain geographical locations and demographic subsets to benefit more than others.
NLP is used more with CDI software, adapted to each CDI program’s request, but software recommended for CDI programs using AI is not as adaptable, making it difficult for programs to adjust the software to their specific needs. So, if the AI software cannot be modified to implement each healthcare system’s requirements, this could decrease production and increase loss in revenue. Working as a CDI for the last ten years in inpatient and outpatient has shown me different ways to capture suspected or more specific diagnoses. The challenges seen with current software is employing that which shows opportunities for CDI professionals to query but does not read the negative validating words and therefore cannot recalculate and request a more specific query such as present on admission (POA) status.
The most important aspect of AI technology is automated, real-time data to help provide vital automated processes allowing healthcare professionals to focus primarily on patients and quality care. When productivity goes up, this allows healthcare systems to show an increase in revenue and faster capture of diagnoses to improve communication with all clinical staff and other medical centers.
AI technology will change how CDI professionals capture the accuracy and compliance of documentation, but there will still be opportunities where the software will miss a diagnosis that a human can recognize despite specific algorithms assigned. Machine learning technology is still several years away before it can be implemented in all aspects of healthcare, especially in CDI programs.
Published By: Jessica Vaughn MSN, RN, CCDS, CCDSO, CRC and Sandra Love BSN, RN, CCDS, CCDS-O, CPC
References:
https://jamanetwork.com/journals/jama/article-abstract/2787207
Related News & Insights
Compliant capture of SDOH and chronic conditions a healthy imperative
We know about the bad examples. Insurance companies adding diagnoses solely to inflate risk scores and…
Outpatient CDI adoption, denials prevention and risk adjustment trending high in CDI
By Brian Murphy CDI Week might be winding down … but I’m keyed up for the…