Lucidity Caught on Camera

For months she had not said her daughter's name. Then, looking at family photographs, she named several relatives. The episode lasted about five minutes. A clinician who reviewed the recording said it was like having a conversation with anyone.

That description comes from Table 2 of a study by Andrea Gilmore-Bykovskyi, Haley Kottler, Amy Cochran and colleagues, published in The Gerontologist in 2025. It is the first prospective attempt to detect, characterise and validate lucid episodes in advanced dementia using longitudinal audiovisual observation rather than later recollection alone.

What the camera can hold

Twenty people living with advanced dementia were enrolled between December 2021 and April 2024 in a twelve-bed dementia hospice unit. Eligibility required a medical diagnosis of dementia, a Dementia Severity Rating Scale score of 37 or higher (severe), and a legally authorised representative able to consent. Across the cohort the team collected 1,433 hours of audiovisual material in 539 observations. One participant died before filming began; among the other nineteen, the number of sessions ranged from 2 to 127.

Potential lucid episodes were not left to a single impression. Verbal output and nonverbal communicative actions — coherence, eye gaze, gestures, touch — were coded against each person's baseline. Sixteen candidate episodes, involving six of the twenty participants, went to structured case review. Forty-two reviews involved ten family or friend caregivers and eight clinicians. Nine of the sixteen candidates (56%) were validated, and they belonged to only three people (15% of those enrolled).

Not only a farewell

Lucidity near death has a long anecdotal literature, and the hospice setting was chosen partly because earlier case reports suggested more episodes close to dying. The validated events in this sample do not all fit that frame. For one participant, three episodes occurred about twenty, nineteen and eighteen months before death. For another, four episodes fell between four and two months before death. A third had a day-long recovery about fifty-two days before death, and then, after a two-week decline that clinicians read as approaching the end of life, two further days of fuller speech, walking and recognition — about nine to ten days before death — including the sentence «I'm leaving soon».

Across every validated episode the shared, statistically significant change against that person's baseline was an increased rate of meaningful vocalisations. Nonverbal shifts varied: eye gaze, gesture and touch did not move in a single pattern. Heterogeneity is part of the result, not a footnote.

Observation is not explanation

The study documents recoveries of communication that informants recognised as unexpected. It does not establish why they happen. The authors write that measurement advances are needed to elucidate underlying mechanisms; they do not claim that the episodes prove a consciousness independent of brain activity. One caregiver, reviewing a longer episode, wondered aloud whether the person's «soul» had returned. The paper records that sentence as testimony. It is not offered as a finding.

The sample is small, single-site and demographically narrow — limits the authors state. Fitting a Poisson model to 556 observation days, of which ten included a lucid episode, they estimate a mean of about 55.6 days between episodes, with a wide confidence interval (35.3 to 115.9). Under an assumed mean of sixty days, roughly 3,742 observation days would be needed for a confidence interval no wider than thirty days — more than five times the volume they have. The claim that lucid episodes «may not be as rare as initially hypothesized» is their inference from this thin rate estimate, not a population prevalence.

What remains after checking

An earlier Gallup survey piece on this site asked how often witnesses place unexpected lucidity in mid-course dementia rather than only at the end. This study asks a different question: what can still be sustained when the evidence is a recording, a baseline, and a review that can fail. Seven of sixteen candidate episodes did not survive that review. Three people out of twenty accounted for every validated event.

The door that opens is modest. Communication once presumed gone can return in a form that others recognise, can be timed against a person's own ordinary days, and can happen months before death as well as near it. Closing that door would require ignoring the tapes. Walking through it as proof of a mind that no longer needs a damaged brain would go beyond anything the authors measured.