The man across from me in clinic last month was 67, a retired finance executive, and he had been running an online dual n-back task for forty minutes a day for two years. He pulled up the spreadsheet on his phone to show me. He wanted to know whether it was working. His sister had been diagnosed with Alzheimer's at 71, and he was determined the same thing would not happen to him.
I told him the truth. Yes, brain training does something. The evidence is now strong enough to say that plainly. But what he was doing alone, hammering away at one app in isolation, was leaving most of the protective effect on the table.
A study published in February 2026 followed 2,802 adults who had been randomized in the 1990s into one of four arms: memory training, reasoning training, cognitive speed training, or a no-contact control. Each training arm got roughly ten one-hour sessions over six weeks, with a handful of booster sessions at 11 and 35 months. The participants were then tracked for twenty years.
Two decades later, the people who did speed-of-processing training had a 25 percent lower rate of being diagnosed with dementia. Memory and reasoning training showed no protective effect. The benefit was specific, durable, and arose from a startlingly modest dose of brain work performed when these adults were already in their 70s.
Why Speed Training and Not the Others
The lead investigator on this analysis, Jerri Edwards at the University of South Florida, offers a mechanistic interpretation. Memory training and reasoning training rely on explicit strategy: the participant consciously learns a technique, then applies it. Speed training works differently. It pushes the brain to extract information from a brief visual scene faster than feels comfortable, and over many trials the processing gets automated. The learning is implicit, the way swimming or tying a knot becomes implicit. You cannot tell someone how you do it. Your nervous system just does it.
That distinction matters because implicit learning recruits the basal ganglia and cerebellum, regions that show different vulnerability patterns in early Alzheimer's pathology than the strategy-dependent prefrontal networks. The skill outlasts the lessons. It is neural infrastructure laid down once and used for decades.
The original 10-year ACTIVE follow-up, published in 2017 in Alzheimer's & Dementia, found a 29 percent reduction in dementia diagnoses among the speed training arm. The 20-year update brings that number down to 25 percent, which is what you would expect as a cohort ages. The effect is real and it persists.
The Number That Doesn't Tell the Whole Story
Twenty-five percent risk reduction is a finding any neurologist would notice. It is also misleading, because brain training was never going to be the only lever.
The Finnish Geriatric Intervention Study to Prevent Cognitive Impairment and Disability, known as FINGER, is the trial that anchors how I think about prevention with my patients. Researchers randomized 1,260 adults at risk for cognitive decline into either a two-year multidomain intervention or a control arm receiving general health advice. The intervention combined four pillars: structured aerobic and resistance exercise, a Nordic anti-inflammatory diet, computerized cognitive training, and active monitoring of vascular and metabolic risk factors.
Two years in, the intervention arm performed measurably better on global cognition, executive function, and processing speed. The protection held regardless of baseline genetic risk, including in APOE-ε4 carriers, the group with the highest inherited Alzheimer's risk. Follow-up data through the World-Wide FINGERS consortium now extends past a decade, and the protective effect has not faded.
What FINGER adds to ACTIVE is the demonstration that the pillars combine in a way single interventions cannot match. Exercise raises BDNF, the brain's primary growth factor for new synapses. Vascular optimization improves cerebral blood flow, which is the substrate the brain needs to lay down those synapses. The Nordic dietary pattern lowers neuroinflammation, which otherwise corrodes synaptic plasticity. Cognitive training then directs the resulting plasticity into specific circuits. Pull any pillar out and the other three lose efficiency. Stack them and the protective effect compounds.
What This Looks Like Inside a Real Protocol
When I sit with a patient at high risk for cognitive decline, with family history, mild metabolic dysfunction, sleep apnea on overnight oximetry, and declining processing speed on neurocognitive testing, the conversation is never about one intervention. It is about what stack the next twelve months should look like.
The cognitive piece is the easy part. Twenty to thirty minutes of speed-of-processing training a few days a week, on a validated platform, is dose-equivalent to what ACTIVE participants received. That is the floor.
The rest of the stack is what most patients miss without a clinician quarterbacking it. Cardiovascular conditioning at 70 percent of maximum heart rate for forty-five minutes, four times a week, raises serum BDNF in a measurable way, which I have written about in detail in the latest aerobic exercise RCT. A Mediterranean or Nordic dietary pattern, paired with tight glycemic control, lowers the inflammatory load the training is fighting against. Sleep architecture has to be intact, because slow-wave sleep is when the day's training consolidates into long-term circuit changes. Without it, the rep does not count. Targeted brain metabolism support, including the kind of NAD+ and omega-3 protocols our affiliate Action Potential Supplements formulates, gives the mitochondria the substrate to perform this work.
This integration is what the Intensive Brain Health Program at NGP is built around. The diagnostic workup measures the actual gaps, vascular, metabolic, inflammatory, structural via volumetric MRI, and cognitive via processing speed and executive function batteries, and the protocol fills them in sequence. Brain training is one of four pillars. Pulling it out would weaken the whole protocol. Running it alone, the way my finance executive was, captures maybe a quarter of what the science actually offers. That same logic shaped the way I have written about precision neuromedicine as six converging data streams rather than any one biomarker on its own.
The Asset Frame
Cognition is the asset that compounds every other asset. Career capital, relationship capital, creative output, decision quality under uncertainty: all of it runs on the same neural infrastructure that the 20-year ACTIVE data tells us is trainable. The question I would put to anyone past 50 who is not actively maintaining that asset is not whether the data justifies action. The data has been in for some time. The question is whether the protocol they are running is one pillar wide or four.
Ten hours of speed training in 1995 cut the dementia rate for those participants by a quarter over the next twenty years. The next twenty years look meaningfully different when training is one pillar inside a protocol designed to make all four of them work together.