Andrew Gallant described a frightening medical episode that started with flu-like symptoms, panic, and severe anxiety, then escalated into delusions, balance problems, vision issues, and psychiatric hospitalization before a transfer to Brigham and Women’s led to an abnormal MRI, a spinal tap, and a diagnosis of anti-NMDA receptor encephalitis. This is a rare autoimmune disease identified only in 2007. It can look like a primary psychiatric break even though the underlying problem is brain inflammation, and it is often treatable if someone thinks to look for it. Gallant said his primary care doctor’s first pass was understandable given the early symptoms and negative basic workup. The real failure was the handoff into emergency and psychiatric care, where a recent anxiety diagnosis and a “medically cleared” label made it harder to reconsider a neurological cause until a lucky personal connection forced a transfer.
Rare but treatable conditions still get lost inside overworked systems and anchoring bias, which means better escalation paths, stronger patient-advocacy workflows, and cautious use of AI as a second opinion are now operational issues, not just medical ones.
Strongly sympathetic and unsettled. Many people were grateful the author survived, but the dominant mood was frustration with misdiagnosis, especially the way psychiatric labels, rushed care, and specialty silos can trap patients with treatable neurological or autoimmune disease.
01 Anchoring bias, not just lack of knowledge, is what made this case dangerous.
Once the author had a fresh generalized anxiety diagnosis and a basic outpatient workup that looked negative, the emergency department had an easy story available and stopped reopening the differential. That reframes the failure from “nobody knew this rare disease” to “the system locked onto the first plausible explanation and made it hard to escape.”
The biggest risk was the chart telling the next doctor what to believe. Rare disease failures often start as workflow failures, not knowledge failures.
02 What finally triggered escalation was the trajectory, not a single definitive test.
A neurologist pointed to rapid decompensation in a previously healthy young adult plus a subtle focal sign, the author’s squinting left eye, and the wife’s detailed timeline that proved the change was accelerating. That is a sharp reminder that careful chronology from family can be diagnostic data, not just background color.
A precise symptom timeline can be as valuable as a lab result. Fast worsening and focal changes are what should break the “probably psych” frame.
03 Patient advocacy came through as a concrete survival tool, not generic emotional support.
People highlighted advanced healthcare directives, separate psychiatric directives, powers of attorney, and the basic need for someone in the room who can ask questions and push back when the patient cannot. The author confirmed that even with paperwork, navigating psychiatric care was unusual enough that staff said they had rarely seen it handled that way.
If a case could impair judgment or speech, treat advocacy paperwork like disaster recovery planning. You do not want to improvise it during the emergency.
04 LLMs look more useful as search assistants for neglected syndromes than as reliable diagnosticians for acute rare disease.
Several people said ChatGPT helped surface MCAS, POTS, dysautonomia, or Ehlers-Danlos after years of dead ends, usually by turning a messy symptom list into a plausible referral path. The author’s own use during illness did not help and mostly deepened uncertainty around medication side effects. That difference matters. AI seems strongest when the problem is widening the search space, not when the user is cognitively impaired and the case needs decisive escalation.
Use AI to generate possibilities and vocabulary, not to adjudicate a crisis. It is better at pattern expansion than clinical judgment under pressure.
05 The claim that biomedicine needs a software-style boom got a sharp reality check.
Commenters with domain knowledge pointed out that medicine is already advancing fast in areas like cancer, HIV, and vaccines, but it is less visible because progress is gated by biology, regulation, and the impossibility of debugging human bodies like code. That pushes against the common startup instinct that the bottleneck is mostly missing openness or tooling.
Biomedicine is not stalled in the way software people often imagine. The constraints are the substrate itself, not just lack of hacker energy.
01 Some of the apparent incompetence is the unavoidable cost of triage in a world full of vague symptoms and imperfect tests.
Doctors see far more benign complaints than zebras, and even ideal diagnostics are probabilistic because disease states are spectra, normal ranges are statistical, and imaging thresholds are judgment calls. That does not excuse this case, but it does argue against the idea that a better scanner or AI will eliminate misdiagnosis.
Medicine cannot get to zero ambiguity. Any system still has to balance missed zebras against drowning in false alarms.
02 Self-diagnosis culture is now creating its own distortion field.
Specialists dealing with internet-popular conditions like MCAS are getting swamped by patients who arrive convinced of a diagnosis, sometimes without even basic screening, while some clinics are happy to monetize that demand by validating whatever the patient already believes. That complicates the pro-LLM and pro-patient-research narrative. More empowered patients do not automatically mean better signal.
Patient initiative helps when the system misses things, but it also creates noise and incentives for diagnosis vending. More search power cuts both ways.
03 The popular framing of dismissal as “medical misogyny” drew pushback from people who think the term adds more heat than light.
They argued the problem is broader diagnostic bias and base-rate reasoning under uncertainty, not always gendered animus, and noted that some parts of clinical research are no longer male-skewed in the way people often assume. That does not deny sex and gender disparities. It challenges the label being used to explain every failure.
Bias in medicine is real, but the vocabulary matters. If the term implies motive where the mechanism is structural, it can muddy the fix.