Coordinate a cancer patient's records and care with an agent
We had something similar with my father in law who just passed from pancreatic cancer.The doctors each saw one small part of the case, he was being passed between multiple doctors and NPs. Task: We had something similar with my father in law who just passed from pancreatic cancer.The doctors each saw one small part of the case, he was being passed between multiple doctors and NPs. There was no coherence. At one point he ended up on so many medications that made no sense he passed out. Only at the hospital when a doctor looked at everything together did they realize what a mess had been created and got him back on his feet.I ended up giving all of his records to an agent and making it the coordinator. It found a bunch of problems. He was being given the wrong pain medication on the wrong schedule, the moment that was fixed he felt a lot better (unfortunately he lost so much weight on the wrong schedule it's what got him in the end). They also use voice transcription software, which led to a key error about what genetic markers he had. Fixing this changed his treatment plan!In general, I wish we had been much more aggressive with this. Last time we had a major medical issue an NP almost killed my wife. At the time there was no AI, so we just had to recognize things were in decline and start being aggressive about getting in with the right doctors (the moment we found the right one she was diagnosed, had surgery and was cured in weeks).He started doing better when the AI became the central coordinator and doctors/tests/etc. became their inputs. It was an excellent advocate. And while we have science backgrounds, we aren't doctors. But it could present evidence for actions in ways that the doctors appreciated (his oncologist was impressed and called the AI a member of the team).Unfortunately his physical reserves were very drained at that point. He probably would have had another 6-12 months with his family if it wasn't for the disorganized mess of a care system that we have.; tools: custom-built or unnamed agent. Human role: The original post does not say how a person stayed involved.; limitations: The source does not specify the technical architecture, data privacy safeguards, or integration methods for the agent. Medical advice from AI requires strict human oversight and physician validation, as the system cannot replace clinical judgment. The patient's outcome was heavily influenced by pre-existing physical deterioration, limiting the agent's ability to reverse advanced disease progression..
Tools
custom-built or unnamed agent
Published on Agent Work Atlas:
Overview
We had something similar with my father in law who just passed from pancreatic cancer.The doctors each saw one small part of the case, he was being passed between multiple doctors and NPs.
Workflow
We had something similar with my father in law who just passed from pancreatic cancer.The doctors each saw one small part of the case, he was being passed between multiple doctors and NPs. There was no coherence. At one point he ended up on so many medications that made no sense he passed out. Only at the hospital when a doctor looked at everything together did they realize what a mess had been created and got him back on his feet.I ended up giving all of his records to an agent and making it the coordinator. It found a bunch of problems. He was being given the wrong pain medication on the wrong schedule, the moment that was fixed he felt a lot better (unfortunately he lost so much weight on the wrong schedule it's what got him in the end). They also use voice transcription software, which led to a key error about what genetic markers he had. Fixing this changed his treatment plan!In general, I wish we had been much more aggressive with this. Last time we had a major medical issue an NP almost killed my wife. At the time there was no AI, so we just had to recognize things were in decline and start being aggressive about getting in with the right doctors (the moment we found the right one she was diagnosed, had surgery and was cured in weeks).He started doing better when the AI became the central coordinator and doctors/tests/etc. became their inputs. It was an excellent advocate. And while we have science backgrounds, we aren't doctors. But it could present evidence for actions in ways that the doctors appreciated (his oncologist was impressed and called the AI a member of the team).Unfortunately his physical reserves were very drained at that point. He probably would have had another 6-12 months with his family if it wasn't for the disorganized mess of a care system that we have.
Reported result
The agent identified incorrect pain medication schedules and a critical error in genetic marker data caused by voice transcription software. Correcting these issues improved the patient's immediate comfort and altered the treatment plan, though the patient ultimately passed away due to prior physical decline. The oncologist acknowledged the AI as a valuable team member, but no quantitative health metrics were reported.
Human role
The original post does not say how a person stayed involved.
Limitations
The source does not specify the technical architecture, data privacy safeguards, or integration methods for the agent. Medical advice from AI requires strict human oversight and physician validation, as the system cannot replace clinical judgment. The patient's outcome was heavily influenced by pre-existing physical deterioration, limiting the agent's ability to reverse advanced disease progression.
Try it
Act as a medical record coordinator for [Patient Name] with [Diagnosis]. I will provide [List of Medical Records/Transcripts]. Your task is to: 1) Analyze all inputs for medication conflicts, dosage errors, or transcription inconsistencies. 2) Identify any discrepancies in genetic marker data or treatment plans. 3) Draft a summary of findings with evidence citations for the physician. 4) Do NOT send any messages or make changes without my explicit approval. First, perform a dry run on [Sample Record Set] and show me the output format. Wait for my confirmation before processing the full dataset. Ensure all outputs are formatted for easy review by a medical professional.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.