M3 Information · clinician explainer

One person, four quarters

How a behavioural signal that outlives the depression it came with ends up changing a prescribing decision.

Illustrative record — not a real person. Every panel below is rendered by the platform's own functions against this record, so the document cannot drift from what the software does. · Clinician-only. None of this is shown to the person.

The record

Sitting	M3 total	Depr.	Bipolar	Q21	DVPRS	What happened
Q1 2026-01-05	48	14	5	3	6	Baseline. Depression high, bipolar domain medium, pain moderate. Treatment started.
Q2 2026-04-06	38	10	5	3	5	Twelve weeks on. The total is falling and depression with it. The irritability item has not moved.
Q3 2026-07-06	29	6	5	3	5	The M3 total is under its published line of 33. The depression has responded. The irritability has not.
Q4 2026-10-05	31	6	8	4	5	The bipolar domain is now high. Without the cognitive axis on the record, the ladder reaches an atypical antipsychotic here.

Q21 is “I have felt unusually irritable or angry”, on the M3's 0–4 frequency scale. The M3 total's published line is 33; the DVPRS line is 5.

Quarter by quarter

Quarter 1 · 2026-01-05

Baseline. Depression high, bipolar domain medium, pain moderate. Treatment started.

The cue does not fire. One sitting is not persistence, and the mood axis is still over its line.

Quarter 2 · 2026-04-06

Twelve weeks on. The total is falling and depression with it. The irritability item has not moved.

The cue does not fire. The irritability has persisted, but the M3 total is still at or above 33 — nothing has improved yet, so a signal that stays is not yet telling us anything.

Quarter 3 · 2026-07-06

The M3 total is under its published line of 33. The depression has responded. The irritability has not.

Consider the two informant instruments

The mood came under its line. The irritability did not.

This person is 58.
Q21 — “unusually irritable or angry” — has stayed raised across the last 2 sittings (3 then 3 on the 0–4 scale).
The M3 total has come under its published line of 33 (48 → 29). A behavioural signal that travels with the depression is the depression. One that stays after the mood axis improves is the reason this panel is here.
Pain is at 5 on the DVPRS. Read this twice over. Persistent pain has been associated with faster memory decline (Whitlock et al., JAMA Intern Med 2017; observational, and it could not separate the pain from its treatment). It is also one of the commonest reversible causes of late-life irritability — so it may be why the behaviour is here at all. Neither reading gates this panel.

What it suggests doing: offer the IQCODE and the IADL-C — answered by a family member or close friend — so the cognitive axis is on the record. It is worth doing before a medication decision, not after: the M3 ladder reaches an atypical antipsychotic at three of its tiers, and those carry a boxed warning for increased mortality in older adults with dementia-related psychosis.

What this is not. Not a diagnosis, not a prediction, and not a statement that anything is wrong. Late-life irritability is far more often stress, pain, sleep, bereavement, alcohol, hearing loss or a medicine than it is neurodegeneration, and those are worth excluding first. This panel offers two questionnaires and nothing else; the physician decides.
Where the rule comes from. Age 50 and the persistence requirement follow the ISTAART–Alzheimer’s Association criteria for Mild Behavioral Impairment (Ismail et al. 2016; criteria reproduced in Ismail et al., J Alzheimers Dis 2017;56(3):929–938). Persistence rather than a single reading follows Guan et al., Alzheimers Dement (Amst) 2023;15(4):e12483, where single-visit ascertainment called 54.4% of an older cohort positive at HR 2.54 against 26.7% at HR 5.77 when persistence was required. The line of 33 is the M3’s own.

Quarter 4 · 2026-10-05 — what the cognitive axis changes

The bipolar domain is now high. Without the cognitive axis on the record, the ladder reaches an atypical antipsychotic here.

Without the two informant instruments on the record

The ladder returns Tier 3B — Vraylar (cariprazine) + psychiatric referral.

That is the collision. This person is 58, has had a persisting behavioural signal for three quarters that did not respond when the depression did, and the software is reaching for an atypical antipsychotic without knowing anything about the cognitive axis — because nothing on this path measures it.

With them on the record, over their lines

The IQCODE and the IADL-C were offered at Quarter 3 and came back over their published lines. One bit crosses to this path — no score, no instrument, no reading. The ladder now returns:

Safety interlock — medication tier withheld

Cognitive axis raised on the dementia path — medication tier withheld

Clinical action: This person has a cognitive-and-function reading at or over its line on the dementia path, first recorded 8/20/2026. The M3 medication ladder reaches an atypical antipsychotic at three of its tiers, and antipsychotics carry a boxed warning for increased mortality in older adults with dementia-related psychosis. People with Lewy body disease can react severely to conventional antipsychotics. Which dementia is suspected belongs in front of the prescribing decision, not after it, so no tier, no drug and no dose is shown here. The mental-health findings remain on the report as findings to work up. The dementia review holds the reading itself; this panel carries no score.

Therapy consideration: Therapy framing is unchanged — therapy is not what the interlock is holding. Treating the mood, the sleep and the pain remains worth doing at every tier.

Evidence base: M3 CDS — Dementia Framework § the antipsychotic collision; v8.13 findings-only decision (20 Sept 2026); AGS Beers Criteria; class boxed warning, antipsychotics in dementia-related psychosis.

What this example is arguing

The cue's job is not earlier detection. It is to get the cognitive axis measured before the prescription rather than after it. Ismail and colleagues put the problem plainly in 2018: presentations like this one were historically misclassified as idiopathic psychiatric illness, “potentially exposing patients to inappropriate medications or delays in dementia diagnosis.”

Persistence is what makes it worth acting on. A single behavioural reading is nearly worthless as a trigger: Guan et al. found single-visit ascertainment called 54.4% of an older cohort positive at HR 2.54, against 26.7% at HR 5.77 once persistence was required. Quarterly sittings are what let this platform ask the question properly.

And the discriminator is the treatment response. Irritability that falls with the depression is the depression. Irritability still standing after the mood axis has come under its line is the thing worth a second look.

What this is not. Not a diagnosis, not a prediction, and not a claim that this person has or will develop dementia. Late-life irritability is far more often stress, pain, sleep, bereavement, alcohol, hearing loss or a medicine than it is neurodegeneration, and those are worth excluding first. Undertreated pain in particular is both associated with faster cognitive decline (Whitlock et al., JAMA Intern Med 2017 — observational, and it could not separate the pain from its treatment) and one of the commonest reversible causes of the behaviour itself. The platform offers two questionnaires; the physician decides.

Sources for the rule. Age 50 and the persistence requirement follow the ISTAART–Alzheimer's Association criteria for Mild Behavioral Impairment (Ismail et al. 2016; criteria reproduced in Ismail et al., J Alzheimers Dis 2017;56(3):929–938, PMID 28059789). Persistence rather than a single reading follows Guan et al., Alzheimers Dement (Amst) 2023;15(4):e12483. Mild Behavioral Impairment predicts neuropathology-confirmed Alzheimer's disease at HR 1.59 (95% CI 1.07–2.37); the impulse-dyscontrol domain at HR 2.01 (1.75–2.32) — Ruthirakuhan et al., Alzheimers Dement 2022;18(11):2199–2208, PMID 35103400. The line of 33 is the M3's own. One number in the rule has no published source: the cut at which Q21 counts as raised (3, “Often”, on the M3's 0–4 scale), chosen conservatively and held in a single constant.