The American Health Information Management Association (AHIMA) regularly publishes practice briefs related to best practices within the health information arena. These briefs serve as references within the industry to guide organizations when creating policies and procedures for compliant health information management practices. The most recent practice brief titled, “Clinical Validation: A 2023 Update,” provides an in-depth review of clinical validation and its emerging role in healthcare reimbursement and audits. This article explores the potential value of artificial intelligence in supporting improvements in the clinical validation process.
What is clinical validation?
Clinical validation is the process of determining whether a condition can be reported based on documented clinical information that meets clinical validation criteria. The requirements for clinical validation overlap considerably with medical coding requirements; however, coding professionals have met their objective once a condition is reported based on adherence to coding guidelines. Clinical validation takes this one step further. Coding guidelines and clinical validation criteria must be met for a condition to be reported. While clinical validation requires more clinical knowledge than coding, it may be performed by a wider range of professionals, including coding professionals, clinical documentation integrity (CDI) professionals, registered nurses, health information management (HIM) professionals, physicians, and others.
Challenges in clinical validation today
For example, a coding professional may report a condition based on the condition being documented, in most cases with one or more elements of supporting documentation. However, in some instances, the health plan may have established clinical criteria for reporting a condition. If documentation falls short of these clinical criteria, the claim could be denied based on these grounds alone. A commonly cited example of this may be when the coding criteria for sepsis have been met but an organization’s or health plan’s clinical criteria for clinically diagnosing sepsis are not.
This is made even more challenging given there is no universally accepted definition for clinical validation. As discussed in the AHIMA practice brief, CMS compared coding to clinical validation in its 2011 Statement of Work (SOW) for the Recovery Audit Program: “Clinical validation is a separate process, which involves a clinical review of the case to see whether or not the patient truly possesses the conditions that were documented.”
Inpatient and outpatient guidelines also vary in their requirements for reporting chronic conditions and uncertain conditions. However, clinical validation may over time close this gap, as clinical validation criteria per condition may be used regardless of the setting of care.
The importance of clinical indicators in the clinical validation process
A broad range of clinical indicators may be applicable to a condition. Coding professionals, assuming other criteria are met, may correctly report conditions based on documentation of the condition and one or more clinical indicators. In some settings (e.g., inpatient) clinical indicators may not even be required for certain conditions. In contrast, the clinical validation process generally takes into consideration all relevant clinical indicators to determine if clinical criteria have been met.
Most records contain multiple clinical indicators for each documented condition. At the level of the disease model, there may be dozens of applicable clinical indicators, ranging from the clinical history, physical examination findings, lab results, imaging results, interventions, medications, orders (e.g., physical therapy, labs, imaging, consultations, counseling, etc.) the patient’s clinical course, temporality, context, and other criteria. Retaining this level of knowledge for thousands of conditions is beyond the ability of even the most skilled coding or clinical validation professionals. This extends all the way up to the level of a physician, in particular if they are reviewing a record created by a provider in a different specialty.
How artificial intelligence (AI) is being used in clinical validation
As noted above, CMS and commercial payers are raising the bar for reporting conditions by requiring that conditions are fully supported clinically. Clinical validation requires in-depth knowledge of potentially thousands of clinical indicators, not only to identify them but also to assign a relative weighting as to how well they support the evaluation and management of the diagnoses. . Ideally, clinical validation would be performed by a physician in the same specialty as the author of a medical record entry, but this is clearly not always feasible.
A potential solution to this inherent clinical indicator knowledge gap is to leverage the power of artificial intelligence (AI). Machine learning uses neural networks to “learn” what information represents a clinical indicator for a diagnosis. The process requires the review of tens of thousands of clinical records and input from teams of clinicians and coding professionals, with which the wrong and right identifications are used to reinform neural networks. This iterative feedback loop is the key to success.
Natural language processing and rules engines, once trained through machine learning, can easily identify valid conditions and relevant clinical indicators. This process improves over time and is continuing to make significant progress. As noted in the practice brief, some automated coding tools on the market may suggest diagnoses incorrectly or make other errors. However, these challenges are mostly rules-based, making them applicable to machine-based solutions.
Advances in the field of AI are making dramatic improvements in automating efforts to not only accurately code records, but also support the more challenging clinical validation process. Only a small percentage of records are subjected to clinical validation. When this occurs, it may be limited to the validation of a small number of conditions (e.g., sepsis). Advances in AI will allow the clinical validation process to be applicable to a marked wider range of conditions. They will also allow for improvements in efficiency and accuracy. Machine-based coding and clinical validation solutions are here to stay and may be in relative infancy. A potential key role for AI will be to empower clinical validation specialists by providing them with appropriate and in-depth clinical information, as needed.
Embracing clinical validation in risk adjustment
Risk adjustment models such as Medicare Advantage may benefit from approaches that embrace clinical validation. A primary focus of the Office of the Inspector General of the Health and Human Service Department has been assessing whether medical records contain adequate supporting documentation for reported conditions that risk adjust.
Over time risk adjustment audits may shift towards clinical validation. Clinical validation supported by AI has the potential to instantly identify all relevant clinical indicators in a record. Once validated by a coding professional this process would serve to drastically reduce the risk of negative audit outcomes.
Is your MAO prepared for the next, impending regulatory audit? Learn more about how purpose-built, clinically-tuned NLP technology can help Medicare Advantage Organizations (MAOs) increase the accuracy of risk adjustment coding, facilitate streamlined clinical validation, and optimize your risk adjustment processes.
