Why some people with MCI develop Alzheimer's and others never do

What MCI is, and is not

Mild cognitive impairment is a description, not a disease. It means that testing shows a measurable decline in memory or another thinking skill, larger than expected for age, while the person still manages daily life on their own. It can be caused by early Alzheimer's. It can also be caused by depression, poor sleep, medicines, small strokes, thyroid problems, or several of these at once. Which cause is behind it is the main reason the outcomes vary so much.

The numbers that follow come from two kinds of source. The first is this site's own analysis of the Alzheimer's Disease Neuroimaging Initiative, ADNI, a volunteer research cohort of 2,430 people, 1,113 of whom had MCI when they enrolled. The second is the published literature in the research library, which includes studies of community populations that look different from a research cohort. Both matter, and they do not always agree.

How often it progresses

Of the 1,113 ADNI participants with MCI at baseline, 988 had at least one follow-up visit. Over a median follow-up of 35.8 months, 379 progressed to dementia and 609 had not when they were last seen. Using survival methods that account for people who left the study early, the estimated share who had converted was:

• 8.2%by 12 months (95% CI 6.6–10.1)
• 20.9%by 24 months (95% CI 18.4–23.7)
• 30.2%by 36 months (95% CI 27.2–33.4)
• 37.7%by 48 months (95% CI 34.3–41.3)

Read the other way round, roughly 70 in 100 people with MCI in this cohort had not developed dementia three years later, and about 62 in 100 had not after four. The estimated median time to conversion was 79 months. The full curve, with the confidence band, is on our ADNI conversion page.

Those are research-cohort numbers, and ADNI recruits people who are worried enough to volunteer for brain scans and lumbar punctures. The library's broadest answer comes from a meta-analysis in Acta Psychiatrica Scandinavica of 41 inception cohorts. It found an adjusted annual conversion rate from MCI to dementia of 9.6% in specialist clinical settings and 4.9% in community studies, and a cumulative proportion progressing to dementia of 39.2% in specialist settings against 21.9% in population studies. Its conclusion is worth quoting: the annual rate is roughly 5 to 10%, and most people with MCI will not progress to dementia even after ten years of follow-up. Where you are counted from changes the answer by a factor of two.

What predicts progression

In the ADNI analysis, each baseline measurement was tested on its own for how well it separated the people who converted within 36 months from those who did not. The measure is the area under the curve, or AUC: 0.5 is a coin toss and 1.0 is perfect separation.

The order is the interesting part. The three spinal-fluid markers lead: amyloid-beta 42, phosphorylated tau and total tau, the chemical signatures of the Alzheimer's process itself. Hippocampal volume on MRI, which measures damage already done, comes next. Carrying the APOE4 gene variant is weaker than any of those. Age at baseline is weakest of all, barely above chance within this cohort.

The published literature lines up. A meta-analysis in the Journal of Neurology, Neurosurgery and Psychiatry pooled 60 cohort studies with 14,821 participants from 16 countries. The strongest predictors of progression from MCI to Alzheimer's were an abnormal spinal-fluid tau-to-amyloid ratio (relative risk 3.77), hippocampal atrophy (2.59), abnormal p-tau (2.43), and atrophy of the medial temporal lobe and entorhinal cortex. Carrying one or two APOE4 alleles, depression, diabetes, hypertension, older age, female sex and lower scores on the standard bedside tests also raised the risk. A higher body-mass index and a better delayed-recall score were associated with lower risk.

Simple cognitive tests do more work than people expect. A meta-analysis in Neuropsychology Review of 28 studies and 2,365 participants found that many cognitive domains predicted progression with sensitivity and specificity of 0.7 or above, with verbal memory and language tests the strongest, and that combining memory tests with a small number of other domains did best. In the ADNI analysis too, the single biggest improvement in a conversion model came from adding baseline cognitive scores, which we chart on a separate page.

The combination is what counts. An early ADNI study in Neurology of 85 people with amnestic MCI found that those with abnormal results on both a glucose-metabolism PET scan and an episodic memory test were 11.7 times more likely to convert than those normal on both. In that small sample the annual conversion rate was 17.2%, far higher than the community figure above, which is the selection effect at work again.

Why the best markers still get people wrong

An AUC of 0.739, the best single marker in ADNI, sounds impressive. In practice it means that if you pick one person who converted and one who did not, the marker ranks them correctly about 74 times in 100 and wrongly the rest. Combining the markers helps: a joint model reached a concordance of 0.676 on the data it was fitted to, but only 0.637 under cross-validation, which is the honest number. These are useful for deciding who to scan next or who to enrol in a trial. They are not good enough to tell one person what will happen to them.

There are three reasons for the ceiling, and they are interpretation on our part rather than findings of any one paper. The first is that MCI has many causes, and a marker of the Alzheimer's process says nothing about the people whose impairment comes from something else. The second is timing: a three-year window cuts through a process that runs over decades, so people on the same track are counted differently depending on where they started. The third is that the measurements themselves are noisy, and in this cohort 192 baseline amyloid values hit the assay's ceiling and had to be treated as equal to it.

Blood tests are changing the practical side of this. A 2023 study in Nature Aging of 348 memory clinic patients with MCI showed that a blood model based on p-tau217, age and APOE4 status could sort patients into low, intermediate and high risk of amyloid positivity, so that only the intermediate group needed a spinal-fluid test, cutting confirmatory tests by 61 to 86% at an overall accuracy of 88 to 92%. And a 2025 head-to-head comparison across nine cohorts and 1,474 cognitively unimpaired people found that plasma p-tau217 predicted future decline about as well as a tau PET scan. The marker that leads the ADNI ranking, spinal-fluid chemistry, is on its way to being a blood draw.

Why some people never convert

The clearest answer is the dull one: many people with MCI do not have Alzheimer's. Their impairment comes from depression, sleep disorders, medicines, vascular disease or something else, and when that is treated or stabilises, so does the memory. The gap between the clinic rate and the community rate in the 41-cohort meta-analysis is consistent with this: a specialist clinic sees the people whose impairment is most likely to be progressive, and a population survey sees everyone.

A second answer is that even among people whose MCI is due to Alzheimer's changes, the pace varies. In ADNI, the markers with the weakest predictive power were APOE4 and age; the strongest were measures of how far the biological process had already gone. Two people with the same diagnosis can be at different points on the same curve.

A third answer, which the data can only hint at, is reserve. The meta-analysis of 60 cohorts found that higher BMI and better delayed recall were associated with slower progression, and that depression, diabetes and hypertension were associated with faster. Those are associations in observational cohorts, not proven causes, and the library's own prevention page is the place for what trials have actually shown. But they are consistent with the idea that general health and the brain's spare capacity change how long a given amount of pathology can be tolerated.

What this means if it is you or someone you love

Ask what kind of MCI it is. Amnestic, meaning memory is the main problem, carries a different outlook from impairment in attention or language. Ask whether reversible causes have been looked for.

Ask whether a biomarker test is appropriate. Spinal fluid, amyloid PET or, increasingly, a blood test can say whether the Alzheimer's process is present. That changes eligibility for the newer drugs and for trials, and it changes the odds in the table above.

Treat the odds as odds. Three in ten converting by three years in a research cohort means seven in ten had not. A person with MCI and a negative amyloid test has a different future from one with a positive test, and neither is a certainty.

Use the time. MCI is the stage at which the paperwork, the conversations and the planning described on our stages page are easiest. Whatever happens next, none of that is wasted.