Those who know me understand that I have embraced AI and all that it can do, as long as it stays well within its guard rails. As a Physician Associate with many years’ experience, I note that whenever I treated a patient, I had to remain well within my limitations. The basic axiom is you have to know what you don’t know.
The Growing Presence of AI in Clinical Assessments
In our review of clinical records, at times we see prior clinical assessments of the clinical documentation that has flawless grammar, reads exceptionally well, contains far too many (-) em dashes, and the summary is markedly concise. Given our collective experience with the use of various AI tools, several questions come to mind. The first being “how much of this report was generated by artificial intelligence.” The second question to be resolved is “are the conclusions being made reflective of objective evidence-based medicine standards tempered by the understanding that all patients are different?
In seeing a scenario such as this just last week, the question in the back of my mind was “how many organizations are using AI assisted tools even if leadership never deliberately adopted an AI strategy?”
The Importance of Clinical Oversight
There are a growing number of organizations that will summarize clinical records. Also noted are automated chart abstraction tools, dictation systems, and report drafting software. All of which are incorporated into large language model integrations. Thus, the next question becomes not whether AI was used, but whether human clinical judgment was the controlling factor for any determinations made?
Improving Efficiency Without Compromising Quality
To be clear, this is not an anti-AI piece, there are specific applications where AI can assist the decision-making process. When you have a 7000-page medical record containing the same MRI report 42 times, by excluding the excessive number of files, one can expedite the completion of the needed report.
Additionally, there are several applications within AI that can identify missing documentation where reference is made to an imaging study completed, but that imaging study is not in the claim file record. Clearly the use of these functions saves time and improves productivity. Having AI cull out the excessive aspects of the clinical records is a warranted assistance with the heavy lifting required.
The Risks of AI Medical Determinations
Some of the issues that can place the claim file in a precarious position would include reliance upon AI to determine medical causation, an analysis of the mechanism of injury, the clinical significance of each of the findings of all pathology noted on diagnostic imaging studies, and the relative credibility of the competing diagnoses. While the report may look impressive, the lack of an experienced clinician providing appropriate insight negates the utility of the report.
Evaluating the Quality of a Clinical Review
With each report, several questions should be competently answered. This list includes:
- Was the clinical record merely summarized or was there a comprehensive clinical analysis completed?
- Does the mechanism of injury support the diagnosis assigned with the sequelae of the compensable event?
- Were all of the objective findings specifically identified and referenced?
- Does this report distinguish between the acute injury, the sequelae of that event, and any pre-existing, ordinary disease of life degenerative changes, identified?
- Is the author of the report able to fully explain, and clinically defend every conclusion made under oath?
Establishing a Defensible Medical Conclusion
In short, the real standard for any comprehensive review and analysis of the clinical records is the defensibility of this report during the adjudication process. Successful claims professionals simply do not purchase reports. What they are seeking is clear, competent, objective medical evidence supported determinations so that any decision made is clearly in the best interest of all parties concerned. Each question responded to in the report should clearly explain why the diagnosis assigned is supported as a function of the identified event, what treatment would be considered reasonable and necessary to address the sequelae of the compensable injury, and why any unreasonable treatment is just that. A full description of how the evidence presented leads to the conclusions reached must be provided. If AI replaces clinical reasoning, the value of that report diminishes substantially.
Accountability in AI-Assisted Reviews
In conclusion, I am still a supporter of what AI can bring to the table. The question is not “did AI prepare this report?” The question is for each of the conclusions and opinions made, will these determinations withstand scrutiny, and will every assumption survive being challenged? There are organizations out there that are taking the easy way out. While not technically incorrect, these AI generated reports do not advance the claim file or provide for the injured employee. Just as the paintbrush in my hand is a tool that helps me stain my garage doors, AI is a tool that can assist in the completion of a successful report but should not be a replacement for competent clinical judgment.
