When Physical Illness Is Mistaken for Mental Illness

For thirteen years, a man was treated for schizophrenia. Obsessive-compulsive disorder too. They gave him antipsychotics and years of therapy, and he kept getting worse. More delusional. More withdrawn. Turns out, he didn’t have schizophrenia, and doctors could have figured that out with a cheap blood test and a penlight eye exam.

He had a disease called Wilson’s, a genetic glitch that causes copper to build up in the body until it poisons the liver and the brain.¹ There’s an eye exam where you shine a light and look for a faint coppery ring around the cornea, and a blood test for the protein that impacts copper levels.² Both are cheap. Both are old. Neither is exotic. Once someone finally ran them, the thirteen-year mystery resolved in an afternoon.³

As if that weren’t bad enough, they later tested his sixteen-year-old brother (a kid who’d been written off as buckling under school stress). He had Wilson’s too.

Why weren’t they tested earlier?

When physical problems present as mental ones

This is a conundrum I’ve been drawn to for around four years now. Why do we treat mental health patients one way and “medical” or physical-health patients another way?

Depressed? See a psychologist. Migraines? See a doctor. Problem is, the body only has a limited vocabulary when it comes to telling us something is medically wrong. It can’t write us a detailed email, but it can call attention to problems through symptoms. It can make things sore or painful. It can power down parts of the body, discolor them or cause them to malfunction. It can also make us sad, tired, depressed, anxious or grumpy.

Mental health and physical health have historically been treated as separate universes. You can get caught, wrongfully, in the wrong one based on where you happened to seek help first. If you’re exhausted and have brain fog, you may wander into a doctor’s office and get a battery of blood tests. Alternatively, you could have just as easily wandered into a psychiatrist’s office and received years of talk therapy combined with antidepressants.

The Wilson’s example is a real one, but Wilson’s is rare; maybe thirty people in a million.⁴ I’m not arguing we should screen every anxious teenager for copper. I’m using it as an example of a wider pattern where, again, physical health issues often present (to caregivers and patients themselves) as mental health issues. Thyroid trouble can look like depression. Autoimmune disease can look like psychosis.⁵ There’s even a clinical name for the trap, “diagnostic overshadowing.” Once you carry a psychiatric label, every new symptom gets read as more psychiatry, and the hunt for a physical cause quietly stops.⁶

Worse yet, when you start in the wrong place, the massive engines of institutional healthcare can build momentum in the wrong directions, causing negative unintended consequences for patients far beyond the doctor’s office or therapist’s couch.

Why does this happen?

Wanting to understand this issue, my team and I did our best to map (computationally) the complex, intersecting healthcare systems around mental and physical health. What would it take for these issues to be detected? What’s working? What isn’t? Multiple distinct steps in the value chain (2,332 of them). Diagnostic tests, databases, clinical workflows, billing codes. Anything we could think of.

I expected the problem to live in the science. Was there a lack of diagnostic tests or other technology gaps causing misdiagnosed patients to slip through the cracks?

No, not really. The opposite turned out to be true. First of all, this is an old problem and pockets of researchers have looked at it over the decades. For example, in the late 1970s, researchers gave hundreds of psychiatric patients a careful medical workup and found that many of their “mental” symptoms were caused by underlying physical illnesses, and that nearly half of those illnesses had been sitting there undiagnosed.⁷ Half.

Follow-up research in a state hospital put the number even higher.⁸ That was almost fifty years ago. Today, by various counts, a physical exam happens for only a small fraction of psychiatric patients, and a large share of psychiatrists say they don’t feel competent to do one.⁹ It just isn’t something mental health practitioners routinely “do”.

Peeling back the layers, you don’t find a villain. It isn’t really about lazy or greedy therapists. Instead, you find a chain of barriers, each one individually sensible, that add up to a flawed system.

Three key systemic barriers came up in the analysis. They aren’t the “only” ones, but they kept bubbling to the top of the priority list. So, even with plenty of tests on shelves, and even if you can get mental health professionals to consider them, there are still some systemic bottlenecks that need to be unblocked before real progress can scale.

Systemic barriers, named

Medical record systems don’t have the data to train smarter software on. Modern medicine, at scale, increasingly runs on software that scans patient records and flags things worth a second look. The version of that software we’d need here would learn from past cases and flag where psychiatric symptoms might really be caused by a physical illness. The bottleneck is something called “labeled clinical data.” Patient records where a doctor has marked, in a way a computer can read, “this person’s psychiatric symptoms turned out to be caused by a physical illness,” with the lab work attached to prove it. If you wanted to build a tool that learns to flag “this depression feels like a thyroid problem,” you’d need thousands of past cases tagged exactly that way. They barely exist. There are a few hundred where you’d want hundreds of thousands. Here’s the vicious circle: we don’t look because nothing flags it, and nothing flags it because we never looked.

The medical knowledge exists but isn’t in a form software can act on either. Doctors already know the rules. Every guideline says some version of “rule out physical causes first.” That instruction lives in PDF documents and in the trained instincts of good doctors. Turning it into something the software can actually execute, a rule that quietly surfaces “check the thyroid” at the right moment, is a separate job that mostly hasn’t been done. The knowledge is real. It just hasn’t been translated into the language computers speak.

Now here’s the twist. There’s a much simpler version of that prompt, and it already exists. It isn’t smart. It doesn’t learn. It’s closer to a sticky note hard-coded into the order screen: when a doctor is about to prescribe an antidepressant, a small box pops up that says “consider a thyroid test, consider a B12 test.” That much, we know how to build. We’ve built it. It ships inside the dominant medical record systems in the country.¹⁰ It’s almost never switched on. Turning it on isn’t a technical act, it’s a bureaucratic one. Each hospital has to convene a committee, approve the alert, decide who sees it and when, tune it so doctors don’t drown in pop-ups and start ignoring all of them, settle who’s even allowed to order the test, and document who’s accountable when something turns up.¹¹ The capability sits one configuration switch away from the clinician. The switch requires a quorum.

Why are these problems especially bad in mental health, specifically?

None of those three barriers is really about mental health. Encoding a guideline, training a tool on good records, switching on a prompt, those are hard everywhere in medicine. So why are they so much worse here?

The answer goes back to a decision made in 2009. When the government spent billions to drag American medicine into the digital age, it paid hospitals and doctors to adopt electronic records and left mental health and addiction facilities out of the deal.¹⁸ The money that wired up the rest of the system never reached them. Fifteen years later the gap is exactly where you’d expect.

A real share of mental health facilities still can’t order a lab test through their own software, and most don’t connect to the networks that let one provider see another’s results.¹⁸ I can’t prove that single decision is why the man with Wilson’s went thirteen years without the right blood test. It’s hard not to see the link though. The ordinary frictions of medicine land on the one part of the system that was never handed the tools to absorb them. The place we most need someone to look for a physical cause is the place least equipped to do it.

Is it too expensive?

The research is genuinely mixed on blanket testing. Run every lab on every patient who walks in, and most results come back normal, which is expensive and low-yield.¹² That’s true. It argues against the wrong thing though. Nobody needs to test everyone for everything. The targeted version is what matters, and the field already named the trigger for it: look harder when the patient is atypical, or when they aren’t getting better.¹³ The man with Wilson’s got worse for thirteen years. That’s the signal. It was ignored.

The cost argument also misses something. Yes, a cheap test costs more than no test. It costs a great deal less than treating the wrong illness for decades. Thirteen years of antipsychotics, therapy sessions, hospital stays and a life that never improves is not the cheap option. It’s the most expensive option there is, and it doesn’t even work. We’ve been calling the blood test wasteful while quietly paying many times its price to not solve the problem.

The same failure, in plain sight

People with serious mental illness who take antipsychotic drugs are supposed to get routine blood tests, for blood sugar, for cholesterol. It isn’t a judgment call. The guidelines flatly require it. The tests are free, instant and available in every clinic. It still doesn’t happen. People with serious mental illness die more than twice as often as everyone else, and roughly two-thirds of those excess deaths come from ordinary medical causes, the kind these blood tests are built to catch.¹⁴ The gap in how long they live is widening, not closing.¹⁵ When the rule is crystal clear and the test costs nothing and takes a minute, the testing still falls through.

When researchers set out to fix this, they didn’t invent anything. They added a reminder in the medical record and did a little staff education. Monitoring jumped from a third of patients to half, and it held there for more than two years.¹⁶ A nudge. That’s all it took to move the number, which tells you exactly where it was stuck. Not in the lab. In the workflow.

The analysis also flagged a tightening of federal rules on certain lab tests as a possible obstacle. While we were working, a court struck that rule down and the FDA reversed it entirely.¹⁷ The scary regulatory gate everyone braced for simply vanished, and the testing reflex still isn’t there. The barriers that actually matter were never the dramatic ones. They’re the boring ones nobody writes headlines about. The missing records. The fuzzy data. The unwritten rule. The committee that has to meet.

What it looks like when it works

If we were waiting on a breakthrough this would be grim. We aren’t. The most encouraging evidence comes from places that simply decided the looking was someone’s job and then checked whether it happened.

England is the cleanest example. Since 2006 the National Health Service has paid family doctors to give people with serious mental illness an annual physical check, weight, cholesterol, blood pressure, blood sugar, alcohol.¹⁹ More than half a million people are covered. What makes it more than a nice idea is what happened when the policy wobbled. When those checks were dropped from the payment scheme, the number of people getting them fell within the year. When they were added back, it rose again.¹⁹ Pay attention to the looking and the looking happens. Stop paying attention and it stops.

There’s an important limit. Those same studies found the checks improved detection more than treatment. Doctors caught the high cholesterol and the high blood sugar but didn’t always do more about what they caught.¹⁹ Finding the problem turns out to be step one, not the whole staircase. We can’t even get to the hard part until we start finding it, and the finding is the cheap part we keep skipping.

You can see the same lesson closer to home. Programs that fold a mental health clinician into the primary care team, an approach with dozens of trials behind it, get people diagnosed and into treatment far faster than usual care.²⁰ None of it required a new machine. It required deciding, out loud, that someone is responsible for looking at the whole person.

What he needed

The man with Wilson’s spent thirteen years inside a system that was treating him for the wrong illness, while the test that would have named the right one sat untouched a hallway away. That’s the failure, in one sentence.

The strange, almost hopeful part is that this isn’t a science problem. We aren’t waiting on a breakthrough. We aren’t even waiting on the knowledge, we’ve had that since the 1970s. We’re waiting on a system that has treated health and mental health as separate worlds to start treating them as one. The blood test he needed was always there. The eye exam took a minute. What he was waiting for was someone, anyone, to whom the thought occurred. Could this be physical? That’s how the case ended up solved. Someone had the thought, even though the system around them wasn’t built to prompt it.

That’s the warm version of the story and the troubling one. Warm, because the thought did eventually occur to someone, and not much else was needed. Troubling, because at scale you can’t run a system on hoping the right thought occurs to the right person at the right moment. Most of the time it won’t. The fix isn’t to make it easier for clinicians to have the thought. It’s to make the test happen whether the thought occurs or not. A prompt in the software. A check on the schedule. The most unsettling thing I learned, and in a backwards way the most hopeful, is how close that fix already is. Built into the software the doctor was using. One click away.

Endnotes

1. Wilson's disease is an autosomal recessive disorder of copper metabolism caused by mutations in the ATP7B gene, leading to copper accumulation in the liver, brain, and other organs. See Merck Manual Professional Edition, “Wilson Disease.” The case described here is drawn from F. R. Bhatti and S. Liaqat, “Mental or Metabolic?: Misdiagnosis of Wilson Disease as Primary Psychiatric Disorder in Multiple Members of a Pakistani Family,” European Psychiatry 68 (2025), DOI 10.1192/j.eurpsy.2025.1222.

2. Diagnosis typically combines a slit-lamp eye examination for Kayser-Fleischer rings (copper deposits in the cornea) with a serum ceruloplasmin blood test and 24-hour urinary copper. See Merck Manual Professional Edition, “Wilson Disease”; and “Kayser-Fleischer Ring,” StatPearls (NCBI Bookshelf), 2024.

3. Bhatti and Liaqat, “Mental or Metabolic?” (2025). Case 1: a 32-year-old man carrying diagnoses of treatment-resistant schizophrenia and obsessive-compulsive disorder for 13 years, whose symptoms worsened despite extensive psychiatric treatment until an underlying metabolic cause was investigated. Case 2: his 16-year-old brother, initially attributed to academic stress. Case 3: their mother, who developed major depressive disorder with psychosis.

4. Wilson's disease has an estimated prevalence of roughly 1 in 30,000. See “Wilson's Disease / Kayser-Fleischer Ring,” EyeWiki (American Academy of Ophthalmology).

5. Numerous medical disorders produce symptoms that mimic psychiatric conditions, including thyroid disease (depression) and autoimmune encephalitis (psychosis). See Merck Manual Professional Edition, “Medical Assessment of the Patient With Psychiatric Symptoms,” 2026. The best-known popular account of autoimmune encephalitis misdiagnosed as psychiatric illness is Susannah Cahalan’s Brain on Fire: My Month of Madness (Free Press, 2012).

6. “Diagnostic overshadowing” refers to the misattribution of physical symptoms to a patient’s mental illness. See M. Shefer et al., “Diagnostic Overshadowing and Other Challenges Involved in the Diagnostic Process of Patients with Mental Illness Who Present in Emergency Departments with Physical Symptoms,” PLOS ONE (2014), PMC4219761.

7. R. C. W. Hall, M. K. Popkin, R. A. DeVaul, L. A. Faillace, and S. K. Stickney, “Physical Illness Presenting as Psychiatric Disease,” Archives of General Psychiatry 35, no. 11 (1978): 1315–1320, DOI 10.1001/archpsyc.1978.01770350041003. In 658 consecutive psychiatric outpatients, 9.1% had a medical disorder judged to be producing their psychiatric symptoms; 46% of those patients had medical illnesses previously unknown to them or their physician.

8. R. C. W. Hall, E. R. Gardner, S. K. Stickney, A. F. LeCann, and M. K. Popkin, “Physical Illness Manifesting as Psychiatric Disease: II. Analysis of a State Hospital Inpatient Population,” Archives of General Psychiatry 37, no. 9 (1980): 989–995, DOI 10.1001/archpsyc.1980.01780220027002. Of 100 patients intensively evaluated, 46% had medical illnesses judged to have caused or greatly exacerbated the symptoms responsible for admission, and 80% had a physical illness requiring treatment.

9. Surveys of psychiatric practice have reported that physical examinations are performed on a minority of psychiatric patients and that a substantial share of psychiatrists report not feeling competent to conduct one. See B. P. Sharma et al., “Physical Illnesses Among Psychiatric Outpatients in a Tertiary Care Health Institution,” PMC2913645, summarizing earlier survey findings that fewer than 35% of practicing psychiatrists routinely examined patients physically and that physical examination rates were as low as ~13% of inpatients and ~8% of outpatients.

10. Standards-based, in-workflow clinical decision support at the point of order entry (HL7 CDS Hooks “order-select” and “order-sign”) is documented and available within major U.S. electronic health record systems, including Epic. See the HL7 CDS Hooks specification and Epic developer documentation.

11. Local implementation of point-of-care decision support depends on institutional governance: multidisciplinary CDS committees, alert tuning to limit “alert fatigue,” and workflow approval. See ONC/ASTP SAFER Guides (Computerized Provider Order Entry with Decision Support), 2025.

12. Multiple studies and a meta-analysis have found limited yield from routine, untargeted laboratory “medical clearance” screening of psychiatric patients with normal history, vital signs, and physical exam. See S. Conigliaro et al., “Utility of Investigations, History, and Physical Examination in ‘Medical Clearance’ of Psychiatric Patients: A Meta-Analysis,” Psychiatric Services (2021), DOI 10.1176/appi.ps.202000858; and E. L. Anderson, K. Nordstrom, M. P. Wilson, et al., “American Association for Emergency Psychiatry Task Force on Medical Clearance of Adults Part I,” Western Journal of Emergency Medicine 18, no. 2 (2017): 235–242.

13. Clinical guidance recommends targeted medical workup when a psychiatric presentation is atypical or when symptoms fail to respond to, or worsen with, psychiatric treatment. See “Physical Exam in Psychiatry and ‘Medical Clearance,’” PsychDB; and Anderson et al., AAEP Task Force (2017).

14. All-cause mortality among people with serious mental illness is roughly 2 to 3.5 times that of the general population, and approximately two-thirds of the excess mortality is attributable to natural (medical) causes, particularly cardiovascular disease. See T. Soda et al., “Systematic Quality Improvement and Metabolic Monitoring for Individuals Taking Antipsychotic Drugs,” Psychiatric Services 72, no. 6 (2021): 647–653, DOI 10.1176/appi.ps.202000155; and “What Is Behind the 17-Year Life Expectancy Gap Between Individuals With Schizophrenia and the General Population?,” Schizophrenia (Nature) (2025), DOI 10.1038/s41537-025-00667-1.

15. The life-expectancy gap between people with severe mental illness and the general population persists and may be widening, largely because survival gains in the general population have not reached people with SMI. See “Contributions of Specific Causes of Death to Lost Life Expectancy in Severe Mental Illness,” European Psychiatry (2017), PubMed 28391102.

16. Soda et al., “Systematic Quality Improvement and Metabolic Monitoring” (2021). A combination of staff and patient education and an EHR-based reminder raised cardiometabolic monitoring (HbA1c and lipid panel) from 33% to 49% over one year, a gain sustained for 27 months after the intervention.

17. The FDA’s May 2024 final rule subjecting laboratory-developed tests to medical-device regulation was vacated by the U.S. District Court for the Eastern District of Texas on March 31, 2025 (American Clinical Laboratory Association v. FDA), and the FDA formally rescinded the rule on September 19, 2025. See U.S. Food and Drug Administration, “Laboratory Developed Tests,” fda.gov.

18. The 2009 Health Information Technology for Economic and Clinical Health (HITECH) Act provided incentive payments for electronic health record adoption to hospitals and eligible physicians but excluded substance use and mental health treatment facilities; as the Office of the National Coordinator for Health IT (ONC/ASTP) notes, “as a result, EHR adoption and interoperability gaps persist.” In 2024, 68% of substance use and mental health facilities used EHRs only and 25% still used a mix of EHR and paper; among EHR-only facilities, 71% could order lab tests through the EHR (50% of hybrid facilities), and only 19% participated in a health information exchange. See ONC/ASTP, “Electronic Health Record Adoption and Exchange Capabilities Among Substance Use and Mental Health Treatment Facilities, 2024,” healthit.gov (data brief, April 2026), drawing on SAMHSA’s 2024 National Substance Use and Mental Health Services Survey (N-SUMHSS).

19. In England, the Quality and Outcomes Framework has incentivized general practitioners to perform annual physical health checks (including body mass index, cholesterol, blood pressure, blood glucose, and alcohol use) for patients with serious mental illness since 2006, covering more than 535,000 people. A difference-in-difference analysis of national primary care data found that removing checks from the incentive scheme reduced uptake of BMI, cholesterol, and alcohol checks by 14.3, 6.8, and 11.9 percentage points respectively, while reintroducing the BMI check raised its uptake by 10.2 percentage points. See M. A. Matias et al., “Assessing the Uptake of Incentivised Physical Health Checks for People with Serious Mental Illness: A Cohort Study in Primary Care,” British Journal of General Practice 74, no. 744 (2024): e449, DOI 10.3399/BJGP.2023.0532. A separate cohort study found the 2004 incentive produced sustained increases in the recording of elevated cholesterol (odds ratio 1.37), obesity (1.21), and hypertension (1.19) in the serious-mental-illness group, but found no clear effect on the prescribing of lipid-modifying or anti-diabetic medication, indicating incentives improved identification more than treatment. See “Financial Incentives and the Recording of Cardiovascular Risk Factors in Severe Mental Illness,” PMC5466340. Pay-for-performance evidence overall is mixed.

20. The Collaborative Care Model, which embeds behavioral health management within primary care, is supported by more than 80 randomized controlled trials. One review reports that patients in collaborative care reach a diagnosis and begin treatment within six months roughly 75% of the time, compared with under 25% in usual care. See “Collaborative Mental Health Care: A Narrative Review,” PMC9803502; and University of Washington AIMS Center materials on the Collaborative Care Model.