Published by Unseen Progress, an independent publisher of caregiver research. Last reviewed 2026-05-10. Part of the dementia caregiver research overview.
Short answer. Agitation in dementia is almost never random. The dominant research framework — built on Gitlin, Kales, and Lyketsos (2012) and the broader behavioural intervention literature (DICE, TAP) — treats the overwhelming majority of agitation episodes as communications of unmet need: pain, hunger, thirst, fatigue, overstimulation, understimulation, or a disrupted routine. The episode feels random because the precipitant is invisible from inside the moment, not because there isn't one.
Gitlin, Kales, and Lyketsos's 2012 JAMA synthesis of nonpharmacologic management of behavioural symptoms in dementia is the foundational reference. The paper reframes neuropsychiatric symptoms — agitation, resistance to care, wandering, calling out — as the visible end of an unmet-need pathway, rather than as core disease symptoms to be suppressed pharmacologically. The DICE approach (Describe, Investigate, Create, Evaluate) operationalises this: every episode is treated as data, with a structured decoding pass before any intervention.
The implication for caregivers is significant. The default mental model — "this is just the disease" — leads to either passive endurance or reflexive medication. The unmet-need frame leads to investigation, which the research consistently shows produces both better outcomes and lower caregiver burden. The Alzheimer's Association (2024) summarises a large body of evidence that nonpharmacologic approaches grounded in this frame should be tried before antipsychotics for most agitation, given the well-documented harms of antipsychotic use in dementia.
When caregivers say "the agitation came out of nowhere," they almost always mean one of four things:
1. *It came out of nowhere to them. The precipitant existed but was outside their attention — fading light, an unfamiliar visitor an hour earlier, a missed meal, a full bladder. 2. It built across hours of accumulated load. The caregiver saw the eruption; what they did not see was the accumulating overstimulation through the morning. 3. A bodily signal was present but uncommunicated. Pain, constipation, urinary discomfort, or thirst — all common and all hard for a person with moderate-to-severe dementia to name. 4. A change in environment or routine that did not register as a change.* A new aide, a different time for lunch, a rearranged room.
Real caregivers describe the inside of this with painful clarity. "I'm having trouble coping with the sense that my actions are somehow responsible for this," one writes after a difficult evening. "I doubt we'll see improvement; she's only getting more combative," writes another after a UTI-driven escalation. The first reading is rarely correct; the second is often the unmet-need signal that decoding is designed to catch.
The DICE approach (Kales, Gitlin, Lyketsos and colleagues) gives caregivers a structured pass to run after each episode. Adapted for non-clinical use, it has four steps.
Write down the episode in concrete terms. Time of day. Duration. What the person said or did. What you said or did. What stopped it. The act of describing forces precision and produces the data the next steps need.
Run a checklist of common precipitants:
The research suggests roughly 80% of episodes resolve into one of these once the caregiver runs the pass two or three times across a week.
Build a small set of responses you trust for each common precipitant. Pain → analgesic and a quiet environment. Overstimulation → drop demands, lower voice, move to a softer-lit space. Sundowning → close curtains, turn on lights, protect a quiet hour. The goal is not a perfect script; it is to remove decision load when the next episode arrives.
Track which interventions actually worked across multiple episodes. Most caregivers find two or three reliable levers within a month — not by inventing them but by noticing which of their existing instincts produced calm faster.
For an episode in progress, the literature consistently endorses a low-arousal, validating approach:
"I can see you're uncomfortable. I'm right here. Are you hurting anywhere? Are you hungry? Are you too warm?"
Lower the voice half a step; do not match volume. Drop demands (no bath, no medication, no phone call) until the activation passes. If nothing decodes within five minutes, stay with calm presence — sit, breathe audibly, hold a hand — until the wave breaks. Log the precipitant afterwards, not during.
1. After the next three to five episodes, run a written DICE pass — ten minutes, not perfection. 2. Build a one-page list of your person's common precipitants and the responses that worked. Most caregivers find this list stabilises within a month. 3. If episodes spike, treat infection (especially UTI), pain, constipation, sleep loss, and medication review as the top hypotheses before assuming progression. 4. Loop in the GP or memory clinic for any sustained increase in episodes that has not resolved with a decoding pass — but bring the log, not the impression.
---
Unseen Progress publishes long-form caregiver research. See the full dementia caregiver research overview for the complete framework.