Skip to content

Red flags and safety concerns

This page is research knowledge, not medical advice. It records what the literature quantifies about danger signals, iatrogenic harm and misdiagnosis in Alzheimer's disease.

TL;DR — Six things account for most avoidable harm. (1) ARIA: 19% pooled incidence with lecanemab (95% CI 16–23), 3% symptomatic, with an APOE4 gene-dose effect (RR 1.45 heterozygotes, 3.54 homozygotes) and a stage gradient — 27% symptomatic ARIA in mild dementia versus 1.8% in MCI/very mild dementia in a real-world clinic (Qi 2026, PMID 41478817; Paczynski 2025, PMID 40354064). (2) Antipsychotics: OR 1.54 (95% CI 1.06–2.23) for death in 10–12-week trials, widening on continued exposure — 24-month survival 46% versus 71% after withdrawal in DART-AD (Schneider 2005, PMID 16234500; Ballard 2009, PMID 19138567). (3) Anticholinergic burden: adjusted OR for dementia rises to 1.49 (95% CI 1.44–1.54) at >1,095 total standardised daily doses, with a population-attributable fraction of 10.3% (Coupland 2019, PMID 31233095). (4) Delirium: in-hospital mortality 32% with delirium superimposed on dementia versus 8% with neither (adjusted HR 2.14, 95% CI 1.33–3.45) (Avelino-Silva 2017, PMID 28350792). (5) Missed treatable disease: in a prospective rapidly progressive dementia cohort, autoimmune encephalitis was the largest diagnostic category (39%) and the commonest treatment-responsive cause (61% of responsive cases) (van Steenhoven 2026, PMID 42139653). (6) Suicide risk around diagnosis: standardised mortality ratio 1.53 (95% CI 1.42–1.65) in the first year after diagnosis, rising to 3.40 (2.94–3.86) at ages 65–74, and highest in the first 90 days (Schmutte 2022, PMID 34036738).

1. ARIA — the treatment-era emergency

Question Evidence
How common? Pooled ARIA 19% (95% CI 16–23) across 2 RCTs and 5 real-world studies (1,576 lecanemab patients); symptomatic ARIA 3% (2–4); infusion reactions 26% (19–34) (Qi 2026, PMID 41478817). Trial ARIA-E: 12.6% lecanemab, 24.0% donanemab, 24.9% gantenerumab
Who is at risk? APOE4 gene dose: RR 1.45 (heterozygotes), 3.54 (homozygotes) versus non-carriers (PMID 41478817). Any baseline microhaemorrhage increases risk; dose is the other major determinant (Hampel 2023, PMID 37280110)
Does stage matter? Yes, and sharply: symptomatic ARIA in 27% of patients with mild dementia versus 1.8% of those with MCI or very mild dementia in 234 consecutively treated patients (Paczynski 2025, PMID 40354064)
What is it, pathologically? Autopsy of ARIA regions showed microinfarcts with haemosiderin, complement activation and CD68-positive vessel walls arising from Aβ-laden leptomeningeal and penetrating vessels — iatrogenic destabilisation of cerebral amyloid angiopathy (Boon 2025, PMID 41109234)
How is it monitored? Donanemab AUR: baseline MRI within 12 months; surveillance before the 2nd, 3rd, 4th and 7th infusions, before the 12th dose in higher-risk patients, and whenever ARIA is suspected clinically. Exclude if >4 cerebral microbleeds, cortical superficial siderosis, or major vascular contribution to cognitive impairment (Rabinovici 2025, PMID 40155270)
Anticoagulation The lecanemab AUR recommends that patients requiring anticoagulants not receive lecanemab until more data are available (Cummings 2023, PMID 37357276)

Most ARIA is asymptomatic and resolves within 3–4 months or on stopping treatment; symptomatic cases cluster at higher doses (PMID 37280110). In the largest single-clinic real-world series there were no macrohaemorrhages and no deaths among 234 treated patients, with 4.3% withdrawing for ARIA (PMID 40354064) — but the pivotal donanemab trial recorded three treatment-related deaths. Any new headache, confusion, focal deficit, visual disturbance, gait change or seizure in a treated patient is an ARIA question until imaging says otherwise. Full context on anti-amyloid immunotherapy.

Exposure Quantified harm Source
Atypical antipsychotics, 10–12 weeks Death 3.5% vs 2.3%; OR 1.54 (95% CI 1.06–2.23); no differential risk by drug, severity or diagnosis Schneider 2005, PMID 16234500
Antipsychotics, continued beyond 12 months 12-month survival 70% (58–80) vs 77% (64–85); 24-month survival 46% vs 71%; 36-month 30% vs 59% Ballard 2009 (DART-AD), PMID 19138567
Cholinesterase inhibitors Syncope 31.5 vs 18.6 per 1,000 py (HR 1.76, 95% CI 1.57–1.98); bradycardia HR 1.69 (1.32–2.15); pacemaker insertion HR 1.49 (1.12–2.00); hip fracture HR 1.18 (1.04–1.34) Gill 2009, PMID 19433698
Strong anticholinergics, cumulative Adjusted OR for dementia 1.06 (1.03–1.09) at 1–90 TSDDs rising to 1.49 (1.44–1.54) above 1,095 TSDDs; by class at >1,095 TSDDs: antipsychotics 1.70 (1.53–1.90), bladder antimuscarinics 1.65 (1.56–1.75), antiparkinson drugs 1.52 (1.16–2.00), antiepileptics 1.39 (1.22–1.57), antidepressants 1.29 (1.24–1.34). Associations persisted in 3–13-year and 5–20-year exposure windows; PAF 10.3% Coupland 2019, PMID 31233095
Benzodiazepines and Z-drugs Pooled OR for dementia 1.33 (1.19–1.49) for benzodiazepines overall, 1.72 (1.48–1.99) long-acting, 1.43 (1.17–1.74) Z-drugs — but the association did not persist after excluding studies vulnerable to reverse causation, or after introducing a lag period (OR 1.14, 0.82–1.58) AlDawsari 2022, PMID 34679196

Two of these need care in interpretation. The anticholinergic association is observational and vulnerable to confounding by indication (the indications — depression, urinary symptoms, parkinsonism, epilepsy — are themselves prodromal or risk-associated), though the persistence across long exposure windows and the dose gradient argue against pure reverse causation. The benzodiazepine association largely disappears under protopathic-bias correction, which is the more honest summary than the headline odds ratio.

The AAN's MCI guideline recommends discontinuing cognitively impairing medications where possible (Level B) (Petersen 2018, PMID 29282327), and CCCDTD5 sets out how to deprescribe cognitive enhancers safely — gradually, with reinitiation if function deteriorates, continued where neuropsychiatric symptoms responded, and deferred until such symptoms have stabilised (Herrmann 2022, PMID 35128025).

3. Delirium

Among 1,409 hospitalisations of acutely ill patients aged ≥60 in a geriatric ward, 39% had dementia; 13% had dementia alone, 21% delirium alone and 26% delirium superimposed on dementia. In-hospital mortality was 8% (neither), 12% (dementia alone), 29% (delirium alone) and 32% (delirium superimposed on dementia); adjusted hazard ratios for in-hospital death were 2.14 (95% CI 1.33–3.45, P=0.002) for DSD and 2.72 (1.77–4.18, P<0.001) for delirium alone, while dementia alone was not significantly associated (1.69, 0.72–2.30). Twelve-month post-discharge mortality was not associated with any of the diagnostic groups (Avelino-Silva 2017, PMID 28350792).

Practically: an acute change in cognition or behaviour in a person with dementia is a delirium question, not a dementia-progression question, and it carries a doubled in-hospital mortality. This is compounded by the acute-care findings on care, caregiving and health systems — roughly doubled 30-day post-discharge mortality for people with dementia.

4. Diagnoses that should not be missed

Rapidly progressive dementia — dementia developing within a year of symptom onset — has a different differential from typical AD (Geschwind 2007, PMID 17659190). In a prospective multicentre cohort of 147 such patients (median age 67, range 36–86), the distribution was (van Steenhoven 2026, PMID 42139653):

Diagnosis Frequency
Autoimmune encephalitis 58/147 (39%) — and 61% of all treatment-responsive causes
Neurodegenerative disease 20/147 (14%)
Creutzfeldt-Jakob disease 17/147 (12%)

Discriminating features: seizures at first presentation occurred in 34% of autoimmune cases versus 10% of others (P<0.001), most often anti-LGI1 encephalitis, and were subtle focal events in 42% of those cases. Movement disorders at presentation were similarly frequent in autoimmune encephalitis (17%) and CJD (35%, P=0.11) but emerged within 3 months in 80% of autoimmune versus 25% of CJD cases (P=0.004). Among seizure-free autoimmune cases, autoimmune GFAP astrocytopathy was commonest (26%), typically with prominent psychosis or movement disorder.

Other categories to consider before settling on AD: structural lesions and normal-pressure hydrocephalus (structural MRI is part of every evaluation guideline), depression, metabolic and endocrine causes, medication effects (above), alcohol, and non-AD neurodegenerative syndromes with characteristic profiles — see clinical presentation and staging for the phenotype-to-pathology yields, particularly that 86–90% of semantic and nonfluent/agrammatic PPA cases are amyloid-negative.

Biomarker-clinical discordance is itself a red flag. A positive Core 1 biomarker in someone whose syndrome does not fit AD does not establish that AD is causing the syndrome; mixed and non-AD pathology is common (vascular and metabolic contributions).

5. Suicide risk

Population Finding
US Medicare fee-for-service, 2,667,987 newly diagnosed ADRD Suicide rate 26.42 per 100,000 person-years; overall SMR 1.53 (95% CI 1.42–1.65); ages 65–74 SMR 3.40 (2.94–3.86); risk highest in the first 90 days; rural residence and recent mental health, substance use or chronic pain conditions associated with higher risk (Schmutte 2022, PMID 34036738)
US veterans ≥50, 147,595 propensity-matched Suicide attempt in 0.7% with MCI, 0.6% with dementia, 0.4% with neither. Recent MCI diagnosis HR 1.73 (95% CI 1.34–2.22, P<0.001); recent dementia HR 1.44 (1.17–1.77, P=0.001); prior (non-recent) diagnoses not significantly associated. No psychiatric comorbidity moderated the association (Günak 2021, PMID 33760039)

The consistent signal is that risk concentrates around the time of diagnosis, is present for MCI as well as dementia, and is highest in the younger old. Both studies frame this as an argument for support at the point of disclosure — which connects directly to the disclosure literature on diagnostic criteria and the biological definition, where biomarker disclosure to cognitively unimpaired research volunteers produced mild, mostly transient distress in a self-selected population.

6. Driving

In 2,615 licensed drivers aged 65+ followed a mean 7 years and censored at dementia diagnosis, 350 (13%) had at least one police-reported crash; a 1-unit lower cognitive screening score was associated with an adjusted crash incidence rate ratio of 1.26 (95% CI 1.08–1.51), while the amount of decline between visits was not independently associated (Fraade-Blanar 2018, PMID 29667168). The finding applies to people without diagnosed dementia, which is the group in whom driving decisions are hardest and least regulated. Jurisdictional reporting requirements vary and are not summarised here.

7. Unvalidated and disproven interventions

Intervention Evidence Status
Ginkgo biloba 120 mg twice daily 3,069 community volunteers aged ≥75, median 6.1 years: all-cause dementia HR 1.12 (95% CI 0.94–1.33, P=0.21); AD HR 1.16 (0.97–1.39, P=0.11); no effect on MCI-to-dementia progression (HR 1.13, 0.85–1.50) Adequately powered and negative (DeKosky 2008, PMID 19017911)
Vitamin E and memantine in mild-moderate AD TEAM-AD randomised trial of vitamin E, memantine, both or placebo on functional decline Reported; see the trial for effect estimates (Dysken 2014, PMID 24381967)
Cholinesterase inhibitors in MCI AAN: clinicians may choose not to offer; if offering, must first discuss the lack of evidence (Level A) Not supported (Petersen 2018, PMID 29282327)
Anti-inflammatory prevention (naproxen, celecoxib) ADAPT and its ~7-year follow-up: HR 1.03 (celecoxib) and 0.92 (naproxen) for AD; weak evidence of cognitive harm with naproxen Negative — see neuroinflammation and glia

The general rule this knowledge base applies: an observational association between a supplement or drug class and lower dementia risk has repeatedly failed to survive randomisation (ginkgo, NSAIDs, omega-3 in MAPT, intranasal insulin). Claims of cognitive benefit from unvalidated tests or supplements should be checked against whether an adequately powered randomised trial exists.

Direct-to-consumer biomarker testing is an emerging concern: plasma p-tau217 assays are accurate in symptomatic memory-clinic populations but their performance in asymptomatic people is untested, ~90% of published accuracy studies used data-derived thresholds, and many commercial tests do not meet the 90%/90% performance thresholds the 2025 clinical practice guideline sets for substituting for CSF or PET (fluid biomarkers; guidelines). This build's searches did not retrieve an empirical study of harms from direct-to-consumer AD blood testing; that is an evidence gap as of 2026-08-31, not a statement that harms do not occur.

Two safety transitions with measurable risk

After diagnosis: a South Korean nationwide cohort recorded 46 suicides in the first year among 36,541 newly diagnosed older adults. Dementia carried adjusted HR 2.57 (95% CI 1.49–4.44) versus matched controls; AD specifically carried HR 2.50 (1.41–4.44), and risk was higher with schizophrenia, mood or anxiety/somatoform disorders (Choi 2021, PMID 33119492). Generalisability is limited by the country’s higher background suicide rate, but the estimate supports active safety assessment rather than assuming cognitive impairment is protective.

After hospitalisation: 6.6% of 117,022 Medicare beneficiaries with dementia started a new psychotropic class within seven days of discharge, and 51% of new users continued beyond 90 days; delirium increased the odds of initiation 1.8-fold (95% CI 1.7–1.9) (Growdon 2023, PMID 36514208). Medication reconciliation should therefore ask whether an antipsychotic, sedative, antiepileptic or antidepressant was started for a time-limited inpatient syndrome.

Driving decisions require functional evidence, not diagnosis alone. In 1,051 older drivers, attentional/executive deficits were associated with both cessation and prior crashes, while future dementia was associated with self-reported crashes only (Lafont 2008, PMID 18503033). In a smaller two-year prospective study, 26 of 154 stopped driving and 18 crashed; crash history and age predicted crashes, while slower cognitive processing characterised cessation (Kosuge 2017, PMID 28605692). Neither study provides a single safe cutoff, which is precisely why categorical diagnosis is an inadequate licensing test.

A practical trigger list

Observation Consider
New headache, confusion, focal deficit, visual change, gait change or seizure in a patient on an anti-amyloid antibody ARIA — urgent MRI (PMID 37280110; PMID 40155270)
Acute or fluctuating change in cognition/behaviour, especially in hospital Delirium; doubled in-hospital mortality (PMID 28350792)
Dementia developing within 12 months of onset Rapidly progressive dementia workup; autoimmune encephalitis is the commonest treatable cause (PMID 42139653)
Seizures at presentation, subtle or overt Autoimmune encephalitis, especially anti-LGI1 (PMID 42139653)
New or worsening confusion after a medication change Anticholinergic burden; review cumulative exposure (PMID 31233095)
Falls, syncope, unexplained bradycardia on a cholinesterase inhibitor Drug-induced bradyarrhythmia (PMID 19433698)
Antipsychotic prescribed and continued past a review point Mortality hazard widens with duration (PMID 19138567)
Recent diagnosis of MCI or dementia, especially age 65–74 Suicide risk highest in the first 90 days (PMID 34036738; PMID 33760039)
Continued driving with declining cognition, no dementia diagnosis Crash risk rises with lower cognitive score (PMID 29667168)
Biomarker result that does not match the syndrome Copathology or a non-AD cause (PMID 42139653; see vascular contributions)

Open questions

  • Can ARIA risk be predicted precisely enough at the individual level to make treatment safe in mild dementia, where symptomatic ARIA reached 27% (Paczynski 2025, PMID 40354064)?
  • Is anticoagulation an absolute contraindication to anti-amyloid therapy? The AUR advises against treatment pending data that do not exist (Cummings 2023, PMID 37357276).
  • Is the anticholinergic–dementia association causal, and does deprescribing reduce incidence? Trials have tested medication-review or deprescribing interventions, but the retrieved studies measured prescribing or cognitive outcomes rather than dementia incidence: AgeWell.de did not significantly reduce continuing anticholinergic use (67.6% intervention vs 72.1% control, P=0.57), and a 12,787-person trial in people already living with dementia targeted medication dispensing (PMID 40817636; PMID 39432286). Thus the incidence question remained unanswered in the 2026-08-31 search (Coupland 2019, PMID 31233095).
  • Would routine autoimmune-encephalitis screening in rapidly progressive dementia be cost-effective, given a 39% yield in a referral cohort (van Steenhoven 2026, PMID 42139653)?
  • Does structured suicide-risk screening at diagnosis reduce the 90-day peak (Schmutte 2022, PMID 34036738)?
  • What are the harms of direct-to-consumer AD blood testing? A 2025 Spanish Society of Neurology consensus explicitly recommends against direct-to-consumer use (PMID 40685136), but a focused PubMed query on 2026-08-31 retrieved no study measuring downstream patient outcomes; the gap is empirical rather than normative.
  • Does delirium prevention in people with dementia reduce the doubled in-hospital mortality, or does delirium mark severity rather than cause death (Avelino-Silva 2017, PMID 28350792)?

References

  1. Qi L, et al. Safety profiles of lecanemab: a systematic review and meta-analysis of randomized controlled trials and real-world evidence. J Prev Alzheimers Dis. 2026;13:100473. PMID 41478817.
  2. Paczynski M, et al. Lecanemab treatment in a specialty memory clinic. JAMA Neurol. 2025;82:655-665. PMID 40354064.
  3. Hampel H, et al. Amyloid-related imaging abnormalities (ARIA): radiological, biological and clinical characteristics. Brain. 2023;146:4414-4424. PMID 37280110.
  4. Boon BDC, et al. Neuropathological changes and amyloid-related imaging abnormalities in Alzheimer's disease treated with aducanumab versus untreated. Lancet Neurol. 2025;24:931-944. PMID 41109234.
  5. Cummings J, et al. Lecanemab: appropriate use recommendations. J Prev Alzheimers Dis. 2023;10:362-377. PMID 37357276.
  6. Rabinovici GD, et al. Donanemab: appropriate use recommendations. J Prev Alzheimers Dis. 2025;12:100150. PMID 40155270.
  7. Schneider LS, et al. Risk of death with atypical antipsychotic drug treatment for dementia: meta-analysis of randomized placebo-controlled trials. JAMA. 2005;294:1934-43. PMID 16234500.
  8. Ballard C, et al. The dementia antipsychotic withdrawal trial (DART-AD): long-term follow-up of a randomised placebo-controlled trial. Lancet Neurol. 2009;8:151-7. PMID 19138567.
  9. Gill SS, et al. Syncope and its consequences in patients with dementia receiving cholinesterase inhibitors: a population-based cohort study. Arch Intern Med. 2009;169:867-73. PMID 19433698.
  10. Coupland CAC, et al. Anticholinergic drug exposure and the risk of dementia: a nested case-control study. JAMA Intern Med. 2019;179:1084-1093. PMID 31233095.
  11. AlDawsari A, et al. Use of sedative-hypnotic medications and risk of dementia: a systematic review and meta-analysis. Br J Clin Pharmacol. 2022;88:1567-1589. PMID 34679196.
  12. Avelino-Silva TJ, et al. Association between delirium superimposed on dementia and mortality in hospitalized older adults: a prospective cohort study. PLoS Med. 2017;14:e1002264. PMID 28350792.
  13. Geschwind MD, et al. Rapidly progressive dementia. Neurol Clin. 2007;25:783-807. PMID 17659190.
  14. van Steenhoven RW, et al. Autoimmune encephalitis as treatment-responsive cause of rapidly progressive dementia: a multicenter prospective cohort study. Neurology. 2026;106:e214933. PMID 42139653.
  15. Schmutte T, et al. Suicide risk in first year after dementia diagnosis in older adults. Alzheimers Dement. 2022;18:262-271. PMID 34036738.
  16. Günak MM, et al. Risk of suicide attempt in patients with recent diagnosis of mild cognitive impairment or dementia. JAMA Psychiatry. 2021;78:659-666. PMID 33760039.
  17. Fraade-Blanar LA, et al. Cognitive decline and older driver crash risk. J Am Geriatr Soc. 2018;66:1075-1081. PMID 29667168.
  18. DeKosky ST, et al. Ginkgo biloba for prevention of dementia: a randomized controlled trial. JAMA. 2008;300:2253-62. PMID 19017911.
  19. Dysken MW, et al. Effect of vitamin E and memantine on functional decline in Alzheimer disease: the TEAM-AD VA cooperative randomized trial. JAMA. 2014;311:33-44. PMID 24381967.
  20. Petersen RC, et al. Practice guideline update summary: mild cognitive impairment [RETIRED]. Neurology. 2018;90:126-135. PMID 29282327.
  21. Herrmann N, et al. CCCDTD5 recommendations on the deprescribing of cognitive enhancers in dementia. Alzheimers Dement (N Y). 2022;8:e12099. PMID 35128025.
  22. Boekholt M, et al. Deprescribing drugs with anticholinergic effects in older patients with increased risk of dementia in the multicomponent intervention study AgeWell.de. Age Ageing. 2025;54:afaf236. PMID 40817636.
  23. Bayliss EA, et al. High-risk medications in persons living with dementia: a randomized clinical trial. JAMA Intern Med. 2024;184:1448-1457. PMID 39432286.
  24. García-Ribas G, et al. Blood-based biomarkers for Alzheimer's disease: positioning document and usage recommendations from the Behavioral Neurology and Dementia Study Group of the Spanish Society of Neurology. Neurologia (Engl Ed). 2025. PMID 40685136.
  25. Choi JW, et al. Suicide risk within 1 year of dementia diagnosis in older adults: a nationwide retrospective cohort study. J Psychiatry Neurosci. 2021;46:E119-E127. PMID 33119492.
  26. Growdon ME, et al. New psychotropic medication use among Medicare beneficiaries with dementia after hospital discharge. J Am Geriatr Soc. 2023;71:1134-1144. PMID 36514208.
  27. Lafont S, et al. Driving cessation and self-reported car crashes in older drivers: the impact of cognitive impairment and dementia. J Geriatr Psychiatry Neurol. 2008;21:171-82. PMID 18503033.
  28. Kosuge R, et al. Predictors of driving outcomes including both crash involvement and driving cessation in a prospective study of Japanese older drivers. Accid Anal Prev. 2017;106:131-140. PMID 28605692.