Introduction
Artificial intelligence (AI) and innovation have become synonymous in the healthcare industry. It is difficult to find a recent industry publication, trade journal, or new product launch without seeing AI referenced several times. Recently, AI has demonstrated success in the field of radiology to interpret images and produce diagnostic reports automatically. Healthcare institutions are actively using FDA-cleared image analysis applications powered by AI to care for patients. This achievement, along with related advances in AI (specifically natural language processing (NLP) and deep learning), has opened the doors to instantaneously identify diagnoses and conditions by applying AI to free form and uncodified physician visit notes and medical record documents.
However, AI is still new to people who are seeking true AI solutions for their healthcare data processing needs. Additionally, the term “AI” is being used too often to describe technologies that are far from qualifying as AI. As a result, many people are not able to distinguish actual autonomous AI (e.g. Machine Learning, Deep Learning, Natural Language Processing (NLP)) from computer programs that make a large number of simple decisions, require months of upfront setup and need constant reprogramming for every minor logic change (e.g., Keyword searching, pattern matching, hardcoded (static) “if this then that” rules and workflows which include “behind the scenes” human intervention). Not knowing the difference has taught costly lessons to many customers and investors in the healthcare industry (See Differences between Real AI and Fake AI table below).
Differences Between Real AI & Fake AI
| “Real” AI | “Fake” Ai | |
| Data |
|
|
| Analytics |
|
|
| Outputs |
|
|
What Do We Mean by Artificial Intelligence?
Artificial Intelligence refers to computing that can learn, improvise, and evolve like a human being. AI software is designed to:- Collect and combine information from a variety of different sources, including audio and visual sensory information, spatial data, and digital information
- Analyze large amounts of data instantly without any preconceived notions of how the data is organized, focusing on identifying underlying trends and relationships across the data
- Learn constantly from the data to develop ever-evolving processes for identifying increasingly efficient and optimal solutions
How Can AI Be Applied to Medical Record Reviews?
As physicians, hospitals, and other health care providers continue to shift towards electronic medical records, there is a growing wealth of information that can inform health care decision-making, quality improvement initiatives, and risk adjustment activities. While this represents a major opportunity to improve how health care is delivered and financed across the country, it also represents a major challenge. Much of the enormous amount of data in electronic medical records is unstructured—i.e., it does not have any pre-defined data model or schema. Examples include certain textual data, such as notes from text editors, productivity applications, or text messages. Electronic medical records also include data from non-textual sources such as lab images, digital photos or videos, and audio recordings. The lack of structure makes the data much harder to search and analyze. Given the amount of data and complexity of the data, the healthcare industry is turning to AI systems to help understand this data and drive better resource allocation and health outcomes for patients. There are several features of AI that are advantageous for analyzing medical records, including:- Machine Learning. The ability for computers to learn and perform perpetually different and/or better processing without any additional software programming.
- Deep Learning. An advanced sub-field of machine learning that allows for computers to learn using a method that is very similar to how the human brain acquires knowledge. When describing deep learning, the methodology of knowledge formation, as well as the related terminology, is akin to the brain (e.g., neurons, neural pathways, neural networks).
- Natural language processing. Natural language processing refers to programs used to read, decipher, understand, and make sense of human languages. Electronic medical records contain large amounts of written and spoken notes by clinicians, much of it based on esoteric medical terminology. Natural language processing that includes deep learning techniques makes it possible for computer programs to understand the vast nuances in the language used in electronic medical records.
- Adaptability and Scalability. AI programs can be developed to be configurable and integrative with various electronic medical systems. This is largely due to the ability to take in data from various sources, even if those data are not in the same format and are unstructured. This makes AI programs highly scalable.
How Can RaLytics™ Help Payers Manage the Medicare Advantage Risk Adjustment Data Validation (RADV) Process?
Risk adjustment data validation (RADV) is the process that the Centers for Medicare & Medicaid Services (CMS) uses to ensure the accuracy of the data submitted by Medicare Advantage (MA) organizations on the health status of their members. The main pieces of information used to identify the health status of their members are diagnosis codes. These codes are used to risk-adjust capitated monthly payments that MA plans receive from CMS. Members with more diagnoses or more severe diagnoses are associated with higher payments to support the greater use and intensity of services required to effectively manage their health. Hence, RADV helps ensure payment accuracy, as well as the integrity of the Medicare Advantage program.
The current RADV process requires a manual review of medical records to determine whether submitted diagnoses are supported in medical record documentation. Most Medicare Advantage plans use teams of people to review medical records. This is a highly labor-intensive process that limits the number of records that can be reviewed. It also makes it difficult for plans to comprehensively review medical records prior to submitting diagnoses to CMS. That increases the likelihood plans submit diagnosis codes to CMS that are not properly supported in medical records, which can result in the plan having to return payments to CMS.
The RaLytics™ technology platform uses AI, specifically deep learning and natural language processing, to conduct real-time medical record documentation evaluation. The RaLytics platform and its medical record ingestion engine represents a significant advancement in how medical record reviews are conducted, resulting in many advantages over the traditional RADV approach:
- More accurate payments. Using the RaLytics™ platform, it is possible for a Payer to review a significantly larger number of electronic medical records in minutes, in advance of submitting diagnosis data to CMS. That would make it possible for Payers to engage with Providers to address coding inaccuracies prior to when payments are made. These more efficient reviews would make it possible for Payers to receive more accurate payments from CMS, reducing financial uncertainty as well as waste for CMS.
- Lower expenses. Payers will save time and money on manual record reviews. RaLytics™’ solutions will be able to handle increasing enrollment and adjustments in compliance rules without requiring a huge team of internal staff to keep pace with the changes.
- Increased data security. Traditional record reviews require the exchange of medical records containing personal health information via multiple points of contact and human-to-human hand-offs risking disruption in the “chain-of-custody”. RaLytics™’ solutions does not require the sharing of any PHI data across data systems or other points of contact.
- Better identification and remediation of persistent coding issues. By increasing the scope of medical record acquisition and evaluation, RaLytics™ can identify specific health conditions or clinicians associated with persistent coding errors. These clinicians may be candidates to receive training on how to improve coding documentation.