American Health & Medical Research Association
Brain & memory

Memory screening tests are smoke detectors, not diagnoses

Brief cognitive screens are tuned to catch problems, so they raise false alarms, and their cutoffs misjudge some people; here is how to read a score.

By the Health Research USA editorial team

Published Last updated October 2026

Written from the sources listed below; links checked October 2026. It touches on medicines, doses or health conditions and has not yet been reviewed by a licensed clinician, so use it to prepare for a conversation with yours. How we research and review

Sources28, listed below Reading time14 minutes
On this page
  1. What a screen measures, and what it can't do
  2. Why the accuracy on the label beats the accuracy in the clinic
  3. Who the cutoff misjudges
  4. Why a second test can mislead in either direction
  5. What the federal screening position actually says
  6. Online and do-it-yourself tests
  7. When a full neuropsychological evaluation helps
  8. The one thing to remember about a screening score

Your father comes home from his annual checkup with a number. The nurse asked him to remember five words, copy a drawing of a cube and list as many words as he could in a minute. He scored 23 out of 30, and he was told that is below the normal cutoff.

A number like that can feel like a verdict. It was never designed to be one. A brief cognitive screen works like a smoke detector: it is tuned to go off whenever there might be a fire, so it also goes off for burnt toast. And one of the best-known versions had its alarm threshold set on a fairly narrow group of people.

Here is the figure that reframes it. When researchers gave that 30-point test to 2,653 adults in a population-based heart study, average age 50, 66% scored below its standard cutoff for impairment.

What a screen measures, and what it can’t do

A screen is a short set of tasks, usually 3 to 15 minutes, that samples several thinking skills at once: recall of a few words, attention, orientation to date and place, language, and drawing tasks that tap planning and visual skills. The names you are most likely to hear:

  • The mini-mental state examination (MMSE), the older 30-point test, used in large US population surveys since at least the early 1980s.
  • The Montreal cognitive assessment (MoCA), a 30-point test introduced in 2005 to catch milder problems. Its developers designed it as a 10-minute tool; MedlinePlus says it takes about 15 minutes.
  • The Mini-Cog, about 3 minutes: recall three words and draw a clock.
  • Informant questionnaires, such as the AD8, which ask a family member or friend what changes they have seen.

The Alzheimer’s Association, which lists brief patient and informant tools for primary care, says no single one is recognized as the best.

What none of them can do is diagnose. The US Preventive Services Task Force states that screening tests are not intended to diagnose mild cognitive impairment or dementia, and that a positive result should prompt further evaluation. MedlinePlus, the National Library of Medicine’s consumer site, adds the other half: a normal score does not rule out impairment, and these tests alone cannot diagnose dementia or any other cause of thinking problems. Computerized versions are held to the same line: under federal regulation, a computerized cognitive assessment aid is a prescription device whose labeling must warn that it does not identify the presence or absence of a diagnosis.

Why the accuracy on the label beats the accuracy in the clinic

Two settings define any detector. Sensitivity is how reliably it catches real cases. Specificity is how reliably it stays quiet when nothing is wrong. Many brief screens lean toward sensitivity. The arithmetic behind false alarms is worked through in our primer on reading medical test results; here are the numbers for these particular tests.

When the MoCA was introduced in the Journal of the American Geriatrics Society, its developers in Quebec tried it on 94 people with mild cognitive impairment, 93 with mild Alzheimer’s disease and 90 healthy older volunteers. At a cutoff of 26, it caught 90% of the mild impairment and all of the mild Alzheimer’s cases, and it correctly cleared 87% of the healthy group.

That design compares people already known to be impaired with people known to be well. Reviewers at Cochrane, an international network that produces systematic reviews of medical evidence, exclude such studies because of their high risk of bias. Looking only at studies that tested everyone drawn from the same population, their 2021 review found that at the recommended cutoff of 26, sensitivity was 0.94 or higher but specificity was 0.60 or lower. In plain terms, at least 40 of every 100 people without dementia were flagged. The reviewers judged the evidence too thin to support recommendations, for or against, on its usefulness in different settings, and suggested a lower cutoff would likely work better, though that needs confirming. Their searches ended in 2012, despite the review’s date.

The MMSE performs better at its usual cutoffs for dementia. The task force’s 2020 recommendation reports pooled results from 14 studies with 11,972 people: sensitivity of 0.89 and specificity of 0.90 at cutoffs of 23 or 24. A 2016 Cochrane review found similar accuracy in community samples and concluded that the test “should not be used in isolation to confirm or exclude disease.” The Mini-Cog’s record in primary care is thinner: a 2021 Cochrane review found four studies, with sensitivity from 0.76 to 1.00 and specificity from 0.27 to 0.85, too little to recommend it as a primary-care screen.

Who is being tested matters as much as the test. According to the task force, when dementia is common, as among people 85 and older, more than half of positive screens can be correct. Among unselected adults 65 to 74, it can be closer to 1 in 5.

Who the cutoff misjudges

A cutoff is only as good as the people it was set on. The MoCA’s 26 was validated on that Quebec sample. Researchers who later studied African American adults described the earlier norms as derived from an elderly white Canadian population.

The same group’s 2011 study, in Neurology, gave the MoCA to 2,653 participants in a population-based cardiovascular study: 52% African American, 34% white and 11% Hispanic, aged 18 to 85. The average score was 23.4, and 66% fell below 26. The most-missed items were copying the cube, missed by 59%, delayed recall of five words, 56%, and repeating sentences, 55%. When two-thirds of a population sample with an average age of 50 falls below the line, that says more about the line than about the people, and the authors wrote that demographic factors must be considered when interpreting the test.

A 2017 follow-up of 1,419 African American adults, average age about 50, found a mean score of 22.3, with 80% below 26 and the cube drawing missed by 72%. The authors concluded that “certain aspects of this measure and previously established cutoff scores may not be well-suited for some populations.”

Schooling moves scores too. A 1993 JAMA study of 18,056 adults in five US communities found a median MMSE score of 29 for people with at least nine years of school, 26 for five to eight years, and 22 for four years or fewer. A person with little formal education can land at the median for people like them and still fall below a common cutoff. Using data from US Alzheimer’s research centers, a 2018 study of 3,895 people found that the MoCA cutoff that best separated normal thinking from mild impairment was 25 for non-Hispanic white participants, 24 for Hispanic participants and 23 for non-Hispanic Black participants, with lower cutoffs fitting people with less education. The authors suggested cutoffs may need to vary by race, ethnicity and education.

The Alzheimer’s Association’s clinical guideline takes a different angle. It tells clinicians to read the pattern of a person’s results, not simply whether the total clears a cutoff, and to strongly consider referral when extremes of education, language or culture may keep a brief test from capturing real impairment.

Language and hearing can distort results as well. In a 2000 study of 249 older adults from a multilingual community sample, about half of them non-English speakers, education and language undermined the MMSE’s accuracy, while the newer Mini-Cog held up in that sample; later primary-care studies of the Mini-Cog varied widely. Hearing loss matters because most screens are spoken aloud. A 2020 meta-analysis found that people with hearing loss scored about 1.7 points lower on the MoCA, enough to push a borderline score below the line, and its authors urged testers to consider listening conditions and record hearing status. A 2020 review of 81 studies found that no cognitive screening tool had been validated for adults who lost their hearing after learning to speak.

Why a second test can mislead in either direction

People score higher the second time they take most thinking tests. A 2012 meta-analysis of nearly 1,600 effect sizes, in The Clinical Neuropsychologist, traced the gains to memory for specific items, learned strategies and familiarity with testing, with the size depending on age, diagnosis, the time between tests and whether a different version was used. A steady score can hide real decline, and a better one can reflect practice.

Scores also bounce. The MoCA’s creator, Ziad Nasreddine, told KFF Health News in 2019 that he had seen the same patient’s score vary by as much as 5 points over a few weeks. Concern about that variability, and about possible liability for errors, led him to require most clinicians who give the test to complete a one-hour, $125 online certification from September 2019, with access limited to certified users after September 2020. A University of California, San Francisco geriatrician objected to charging for a tool clinicians had promoted as a standard.

A single score also predicts the future poorly. A 2021 Cochrane review of 11 studies, covering 1,569 people with mild cognitive impairment, found that one baseline MMSE score did a poor job of identifying who would develop dementia, and suggested that change over repeated tests might do better, an idea not yet tested. How mild impairment differs from subjective decline is spelled out in our piece on memory changes.

What the federal screening position says

In February 2020 the task force issued an I statement on screening community-dwelling adults 65 and older without signs or symptoms, concluding that the evidence was too thin to weigh benefits against harms. In Task Force terms, that grade signals insufficient evidence, not advice against screening. The evidence review behind it found no empirical evidence that screening improves outcomes for patients or caregivers, or that it causes harm.

The question is open again. In June 2025 the task force published a final research plan for an update, asking whether screening improves decisions and outcomes for patients and families, including safety, and how accurate the instruments are. It excludes blood, spinal fluid and urine biomarkers and brain imaging. As of late October 2026, no draft recommendation had been posted, so the 2020 statement stands.

That statement covers people without symptoms. Once you or your family have noticed a change, testing is part of an evaluation, and the Alzheimer’s Association guideline gives its top strength rating to using a validated cognitive test in that setting. Medicare’s annual wellness visit has included detection of possible cognitive impairment since 2011; who evaluates, and what Medicare covers beyond that visit, belongs to our article on when and how memory problems get evaluated.

Online and do-it-yourself tests

Self-given tests, on paper or on a screen, perform about as well in research settings as clinician-given paper screens. A 2024 Cochrane review of 11 studies with 2,303 participants found sensitivity from 55% to 100% and specificity from 45% to 100% across six self-administered tools. The reviewers found insufficient evidence to recommend any single one and called for more research on the best test and cutoff, and on how setting, language and education might change them.

Real-world proof is thin. A 2025 meta-analysis in Aging & Mental Health pooled unsupervised computerized tests at a sensitivity of 0.84 and specificity of 0.86, but only 3 of its 28 studies checked feasibility and usability in the settings where the tests were meant to be used. A 2026 review of 48 studies of digital tools in Digital Health found most were run in controlled settings with little long-term or real-world validation.

Regulation sets a useful benchmark. Computerized cognitive assessment aids regulated by the Food and Drug Administration must summarize their performance in their labeling, including, where available, sensitivity, specificity, predictive values, repeatability and how cutoffs were set, and must warn that they do not identify a diagnosis. An online quiz promising to tell you whether you have dementia claims more than those regulated devices may. MedlinePlus advises following up with a provider about the results of any online test.

When a full neuropsychological evaluation helps

A neuropsychological evaluation is the long-form version. The American Psychological Association describes clinical neuropsychology as a specialty within clinical psychology devoted to the relationship between brain and behavior. An evaluation typically includes a review of records, an interview with the patient and, with permission, family members, and standardized tests of attention, memory, language, reasoning and problem solving, visual-spatial skills, and sensory and motor skills, sometimes with questionnaires about mood. The neuropsychologist interprets the combined results and reports back to the referring clinician.

The Alzheimer’s Association guideline recommends this kind of evaluation when office testing is not informative enough. Its examples fit this article’s problem exactly: a patient or family reports worrying changes in daily life but the brief test comes back normal, or results are abnormal but hard to interpret because of a complex health picture or factors such as limited or advanced education or language. At minimum, it says, the evaluation should cover learning and memory, attention, executive function, visual-spatial function and language.

Questions to ask about a screening result

  1. Which test was it, and what cutoff did you use? Was my schooling or first language taken into account?
  2. Which parts did I miss? The pattern often says more than the total.
  3. Could hearing, vision, language or the testing conditions have lowered the score?
  4. Was this a screen or part of an evaluation, and what is the next step?
  5. If we repeat it, will you use another version or allow for practice effects?
  6. If the score was normal but we still see changes, what other testing makes sense?
  7. Would a neuropsychological evaluation answer the question better?

When to act sooner

  • A positive or borderline screen from any source, including an online test, calls for clinical follow-up rather than a conclusion either way.
  • Changes you or your family keep noticing despite a normal score are reason to ask about another kind of testing, as MedlinePlus suggests.
  • Abrupt confusion, or speech that suddenly becomes hard to produce or follow, is a possible stroke sign. The CDC says to call 911 rather than drive to the hospital.

The one thing to remember about a screening score

A screening score is an alarm, not an answer. A low score means someone should look more closely, and a high one does not overrule what you and your family have seen. The best thing to bring to the next conversation is context the test cannot measure: your schooling, your first language, your hearing, and a concrete example of what changed and when.

Sources

  1. US Preventive Services Task Force. “Cognitive Impairment in Older Adults: Screening, Final Recommendation Statement (February 25, 2020).” uspreventiveservicestaskforce.org
  2. US Preventive Services Task Force. “Update in Progress: Cognitive Impairment in Older Adults: Screening (last updated June 12, 2025).” uspreventiveservicestaskforce.org
  3. US Preventive Services Task Force. “Final Research Plan: Cognitive Impairment in Older Adults: Screening (June 12, 2025).” uspreventiveservicestaskforce.org
  4. JAMA (US Preventive Services Task Force). “Owens DK, Davidson KW, Krist AH, et al. "Screening for Cognitive Impairment in Older Adults: US Preventive Services Task Force Recommendation Statement." JAMA, 2020;323(8):757-763. PubMed. DOI: 10.1001/jama.2020.0435.” pubmed.ncbi.nlm.nih.gov
  5. JAMA. “Patnode CD, Perdue LA, Rossom RC, et al. "Screening for Cognitive Impairment in Older Adults: Updated Evidence Report and Systematic Review for the US Preventive Services Task Force." JAMA, 2020;323(8):764-785. PubMed. DOI: 10.1001/jama.2019.22258.” pubmed.ncbi.nlm.nih.gov
  6. Code of Federal Regulations (Legal Information Institute, Cornell Law School). “21 CFR 882.1470, Computerized cognitive assessment aid.” law.cornell.edu
  7. National Library of Medicine (MedlinePlus). “Cognitive Testing (last updated August 27, 2026).” medlineplus.gov
  8. Alzheimer's Association. “Cognitive Screening and Assessment (for health systems and medical professionals).” alz.org
  9. Alzheimer's & Dementia (Alzheimer's Association). “Atri A, Dickerson BC, Clevenger C, et al. "Alzheimer's Association clinical practice guideline for the Diagnostic Evaluation, Testing, Counseling, and Disclosure of Suspected Alzheimer's Disease and Related Disorders (DETeCD-ADRD): Executive summary of recommendations for primary care." Alzheimers Dement, 2025;21(6):e14333 (online December 2024). PubMed. DOI: 10.1002/alz.14333.” pubmed.ncbi.nlm.nih.gov
  10. American Psychological Association. “Clinical Neuropsychology.” apa.org
  11. Centers for Disease Control and Prevention. “Signs and Symptoms of Stroke (reviewed May 19, 2026).” cdc.gov
  12. Journal of the American Geriatrics Society. “Nasreddine ZS, Phillips NA, Bédirian V, et al. "The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment." J Am Geriatr Soc, 2005;53(4):695-699. PubMed. DOI: 10.1111/j.1532-5415.2005.53221.x.” pubmed.ncbi.nlm.nih.gov
  13. Cochrane Database of Systematic Reviews. “Davis DH, Creavin ST, Yip JL, et al. "Montreal Cognitive Assessment for the detection of dementia." Cochrane Database Syst Rev, 2021;7:CD010775. PubMed. DOI: 10.1002/14651858.CD010775.pub3.” pubmed.ncbi.nlm.nih.gov
  14. Cochrane Database of Systematic Reviews. “Creavin ST, Wisniewski S, Noel-Storr AH, et al. "Mini-Mental State Examination (MMSE) for the detection of dementia in clinically unevaluated people aged 65 and over in community and primary care populations." Cochrane Database Syst Rev, 2016;(1):CD011145. PubMed. DOI: 10.1002/14651858.CD011145.pub2.” pubmed.ncbi.nlm.nih.gov
  15. Cochrane Database of Systematic Reviews. “Seitz DP, Chan CC, Newton HT, et al. "Mini-Cog for the detection of dementia within a primary care setting." Cochrane Database Syst Rev, 2021;7:CD011415. PubMed. DOI: 10.1002/14651858.CD011415.pub3.” pubmed.ncbi.nlm.nih.gov
  16. Cochrane Database of Systematic Reviews. “Arevalo-Rodriguez I, Smailagic N, Roqué-Figuls M, et al. "Mini-Mental State Examination (MMSE) for the early detection of dementia in people with mild cognitive impairment (MCI)." Cochrane Database Syst Rev, 2021;7:CD010783. PubMed. DOI: 10.1002/14651858.CD010783.pub3.” pubmed.ncbi.nlm.nih.gov
  17. Cochrane Database of Systematic Reviews. “Naeem F, McCleery J, Hietamies TM, et al. "Diagnostic test accuracy of self-administered cognitive assessment tools for dementia." Cochrane Database Syst Rev, 2024;12:CD013725. PubMed. DOI: 10.1002/14651858.CD013725.pub2.” pubmed.ncbi.nlm.nih.gov
  18. Neurology. “Rossetti HC, Lacritz LH, Cullum CM, Weiner MF. "Normative data for the Montreal Cognitive Assessment (MoCA) in a population-based sample." Neurology, 2011;77(13):1272-1275. PubMed. DOI: 10.1212/WNL.0b013e318230208a.” pubmed.ncbi.nlm.nih.gov
  19. Archives of Clinical Neuropsychology. “Rossetti HC, Lacritz LH, Hynan LS, et al. "Montreal Cognitive Assessment Performance among Community-Dwelling African Americans." Arch Clin Neuropsychol, 2017;32(2):238-244. PubMed. DOI: 10.1093/arclin/acw095.” pubmed.ncbi.nlm.nih.gov
  20. JAMA. “Crum RM, Anthony JC, Bassett SS, Folstein MF. "Population-based norms for the Mini-Mental State Examination by age and educational level." JAMA, 1993;269(18):2386-2391. PubMed. PMID: 8479064 (no DOI in record).” pubmed.ncbi.nlm.nih.gov
  21. Alzheimer's & Dementia: Diagnosis, Assessment & Disease Monitoring. “Milani SA, Marsiske M, Cottler LB, Chen X, Striley CW. "Optimal cutoffs for the Montreal Cognitive Assessment vary by race and ethnicity." Alzheimers Dement (Amst), 2018;10:773-781. PubMed. DOI: 10.1016/j.dadm.2018.09.003.” pubmed.ncbi.nlm.nih.gov
  22. International Journal of Geriatric Psychiatry. “Borson S, Scanlan J, Brush M, Vitaliano P, Dokmak A. "The mini-cog: a cognitive 'vital signs' measure for dementia screening in multi-lingual elderly." Int J Geriatr Psychiatry, 2000;15(11):1021-1027. PubMed. DOI: 10.1002/1099-1166(200011)15:11<1021::aid-gps234>3.0.co;2-6.” pubmed.ncbi.nlm.nih.gov
  23. International Journal of Geriatric Psychiatry. “Utoomprurkporn N, Woodall K, Stott J, Costafreda SG, Bamiou DE. "Hearing-impaired population performance and the effect of hearing interventions on Montreal Cognitive Assessment (MoCA): Systematic review and meta-analysis." Int J Geriatr Psychiatry, 2020;35(9):962-971. PubMed. DOI: 10.1002/gps.5354.” pubmed.ncbi.nlm.nih.gov
  24. Otolaryngology-Head and Neck Surgery. “Raymond M, Barrett D, Lee DJ, et al. "Cognitive Screening of Adults With Postlingual Hearing Loss: A Systematic Review." Otolaryngol Head Neck Surg, 2021;164(1):49-56 (online July 2020). PubMed. DOI: 10.1177/0194599820933255.” pubmed.ncbi.nlm.nih.gov
  25. The Clinical Neuropsychologist. “Calamia M, Markon K, Tranel D. "Scoring higher the second time around: meta-analyses of practice effects in neuropsychological assessment." Clin Neuropsychol, 2012;26(4):543-570. PubMed. DOI: 10.1080/13854046.2012.680913.” pubmed.ncbi.nlm.nih.gov
  26. Aging & Mental Health. “Li L, Zhang B, Yang Y, et al. "Diagnostic performance of self-administered unsupervised computerized cognitive tests for the identification of mild cognitive impairment (MCI) and dementia: a systematic review and meta-analysis." Aging Ment Health, 2025;29(8):1439-1455. PubMed. DOI: 10.1080/13607863.2025.2495794.” pubmed.ncbi.nlm.nih.gov
  27. Digital Health. “Hong S, Kim C, Lee SH. "Digital tools and biomarkers for cognitive screening: A systematic review." Digit Health, 2026;12:20552076261461194. PubMed. DOI: 10.1177/20552076261461194.” pubmed.ncbi.nlm.nih.gov
  28. KFF Health News. “Aleccia J. "Creator Of Brain Exam That Trump Aced Demands New Training For Testers" (July 29, 2019).” kffhealthnews.org