The Growing Role of AI in Claims Review
After many years working in workers compensation cost containment, I have no doubt that artificial intelligent has a useful role in the claim review process. AI can help organize the clinical records, summarize the medical information, and if necessary, identify specific patterns. However, within the confines of our Workers Compensation ecosystem the value of the report is not determined by the speed in which it is produced, or how polished it appears, the value is that this report explains the mechanism of injury, the clinical objective findings, the results of diagnostic imaging studies, and the rationale for specific treatment interventions in a way that all parties can easily understand. The bottom line is whether the report can be defended on medical, factual, and logical grounds.
The Risks of AI-Generated Medical Reports
Recently I have come across several reports from various sources, and to my informally trained eye I could tell these were AI generated reports. As an experiment, I ran these reports through three separate LLM platforms asking the same prompt “were these reports generated via AI?” Each of the output content from these prompts noted, to a reported 75%-85% certainty (with the understanding that the veracity of this statistic is questioned), that AI was utilized to generate the reports.
However, the more significant statistic identified was that these reports lacked significant support in terms of medical defensibility. From a litigation perspective, the vulnerabilities were not subtle. The reports left openings that any capable opposing attorney would recognize quickly.
Evidence-Based Medicine Is Still the Foundation
I have long been a proponent of the implementation of evidence-based medicine when making any clinical determination. The clinical literature supports those determinations. Thus, at the time of adjudication, there is no fear that the defensibility of the clinical report would be challenged. Unfortunately, if one relies to an 80% certainty that this report was generated by AI, the defensibility issue appears most probably at the worst possible moment. There was no clear clinical application that the mechanism of injury supports the diagnosis as a function of the compensable event. The findings on MRI may use the phrase of “consistent with” but that is a far distance from being a function of the reported mechanism of injury.
AI Supports the Process, Not the Clinical Opinion
There is no doubt that AI has a place in the process. However, given that these reports may affect the appropriate treatment rendered, it remains my opinion that human intervention and assessment based on individual education, training and experience should be the hallmark of moving forward. After all, isn’t this why the records were sent out for independent review? The claim file handler wants competent clinical information based on the records answering the questions posed.
Questions Every Claims Professional Should Ask
To be clear, a review of that clinical report should answer a number of questions.
- Were the records summarized but not analyzed?
- Were the findings noted on diagnostic imaging studies associated with the reported mechanism of injury with a specific clinical correlation?
- Is the phraseology of the report too sterile or automatic?
- Do the statements contain generic narrative that could fit any claim file?
- And most importantly, do the responses to the particular claim file handler questions provided reflect a specific answer and not medical phraseology?
The Final Report Must Be Medically Defensible
The practical question that each claim file handler should have with each report they receive should note “was this report generated by, reviewed by, and supported with someone with the appropriate clinical judgment to reach the noted conclusion?” Artificial intelligence can clearly assist in the creation of the more mundane aspects of the report and find the most recent clinical citations to support the determinations made.
Even a report written personally by a qualified clinician can still fall short if it doesn’t meet the standards above. The issue is not the tool use, it is if the final product provides answers that are medically grounded, relevant to the specific claim, and are medically defensible at each level of the adjudication process.
Please do not take this as a complete rejection of advancing technology. The point is acknowledging what the technology can and cannot provide. A medically defensible report would help you determine the best path to take in handling this claim file.
You are a claim file professional, and not a medical professional making a diagnosis. No one is asking you to make a diagnosis, only to endorse the clinical responses and information presented, if that medical information meets the standards as listed. Understanding that a rapid report is a good thing, it cannot be substituted for medical judgment, clinical reasoning, and that objective basis for any determination. When the report meets each of these standards, it is a particularly useful tool. If the report you have received generates more questions than answers, there is a clear need for a separate medically defensible opinion that meets the standards listed. The issue is not whether software helped prepare the report. The issue is whether competent clinical judgment controlled the final opinion.
Tools are useful. Clinical reasoning is indispensable. Defensibility is the standard.
