After the dementia diagnosis, the prescription outlasts the guidance

A dementia diagnosis in UK primary care is, for a sizeable minority, followed by an antipsychotic. That much has been counted many times. What had not been counted, as the authors' own search of 4158 records across three databases found, is the pair of things guidance is most explicit about: how long the first course runs and at what dose it is started. Holly Smith, Juan Carlos Bazo-Alvarez and colleagues at University College London, Keele, Nottingham, Queen Mary University of London, Aarhus and the Karolinska Institute have now measured both across 24 years of records.

What 9819 records show

The dataset search found 108,910 people with any record indicating dementia; 99,091 were excluded for having no antipsychotic prescription between the ages of 60 and 85, a history of antipsychotics starting more than a year before the dementia record, missing deprivation data, or only one eligible prescription. That left 9819 people and 260,548 prescriptions. Risperidone was the most common first drug, in 3790 people (38.6%), then quetiapine 2175 (22.2%), other agents 1987 (20.2%), haloperidol 990 (10.1%) and olanzapine 877 (8.9%).

Two duration figures appear in the paper and they are not the same statistic. The crude median time between the first and last prescription of the first episode was 5.4 months (IQR 1.5–15.1). The headline 7.0 months (IQR 6.6–8.7) is a model estimate from parametric survival regression adjusting for age, sex and calendar year and accounting for people whose follow-up was incomplete, which is why its interval is so much narrower. Both exceed 1 to 3 months, so the conclusion does not turn on which one is quoted, but the wide observed spread is the more honest picture of practice.

Dose is the other half. At quarter 1, 63.8% (95% CI 62.8–64.7) were on a low dose of one of the four index drugs, 18.1% (17.4–18.9) above the minimum effective dose for those drugs, and 18.1% (17.3–18.9) on an antipsychotic outside the four, for which no equivalence was applied. The 18.1% above minimum effective dose is a share of the whole cohort; among the roughly 82% started on one of the four classifiable drugs it is closer to 22%. By quarter 4, of the 6559 people who had neither died nor reached the end of follow-up, 5136 (78.3%) were still on an antipsychotic: 48.9% (47.7–50.1) low dose, 14.8% (13.9–15.6) moderate or high, 14.6% (13.8–15.5) something else.

The dose people start on is largely the dose they keep

Movement between dose bands over the first year is small. For someone on a low dose at quarter 1, the adjusted probability of being on a low dose at quarter 4 was 0.45 (95% CI 0.44–0.46); for someone on a moderate or high dose, the probability of still being on a moderate or high dose was 0.35 (0.33–0.37). One number needs flagging: table 2 separates its columns into unadjusted and adjusted, and the low-dose row reads 0.44 (0.43–0.45) unadjusted against 0.45 (0.44–0.46) adjusted, while the results text prints 0.44 (0.43–0.45) and labels it the adjusted transition probability. The table is followed here. Of the 1781 who began above the minimum effective dose, 519 (29.1%) were above it in every quarter of the first year. Age separated people more than deprivation did, which showed no differences: 24.4% of those aged 60 to 70 started on a moderate or high dose against 17.2% of those aged 71 to 85, and their adjusted probability of still being there at quarter 4 was 0.47 against 0.32.

Stopping was rarely the end. Of 5547 people eligible to restart, 3125 (56.3%) did, a median of 3.1 months after discontinuation. The second course was shorter than the first, a median of 2.6 months, and by quarter 2 of that second episode 83.6% (82.3–84.9) were on nothing. Another number needs flagging: the printed summary reports 3106 restarts (56%) where table 1 and the results text both report 3125 (56.3%). The table is followed here. A sensitivity analysis that drops episodes beginning and ending on the same day lengthens both figures, to 6.5 months (IQR 2.1–16.5) for the first course and 5.3 months (1.9–14.1) for the restart.

The calendar bands are worth a practitioner's eye, because they are the closest thing here to a policy readout. Crude first-episode medians run 4.1 months in 2000–2004, 5.8 in 2005–2009, 5.7 in 2010–2014, 6.4 in 2015–2019 and 4.6 in 2020–2023, the last band having a shorter observation window and more censoring than the others. The paper reports these as baseline description and does not test a time trend, so this is not evidence that courses lengthened. It is, at minimum, no visible shortening across two decades of guidance.

What the guidance asks for, and what this study cannot settle

The 1 to 3 month ceiling the authors benchmark against comes from the 2006 NICE dementia guideline they cite. The guideline currently in force, NG97, published 20 June 2018, sets no numeric ceiling at all: recommendation 1.7.6 asks for the lowest effective dose, the shortest possible time, and reassessment at least every 6 weeks. Around it sits the rest of the sequence, which is where this material meets the rest of the issue. Recommendation 1.7.1 asks for a structured assessment of possible reasons for distress and for clinical or environmental causes, naming pain, delirium and inappropriate care. Recommendation 1.7.2 makes psychosocial and environmental interventions the initial and the ongoing management. Recommendation 1.7.3 restricts antipsychotics to people at risk of harming themselves or others, or experiencing agitation, hallucinations or delusions causing severe distress. Recommendation 1.7.8 asks that people living with dementia can continue to access psychosocial and environmental interventions for distress while they are taking antipsychotics and after they have stopped taking them. Neither the 2006 wording nor the 2018 wording reads as seven months at an unreviewed dose.

The obvious next sentence would be that non-drug care should simply replace these prescriptions, and the authors decline to write it. They note that a 2023 review found no conclusive evidence that psychosocial interventions reduced antipsychotic use in care homes, and that subsequent commentary argued this under-represented positive findings from trials such as WHELD; they point to WHELD in care homes and NIDUS-family in the community as promising rather than settled, and call for investment, staff training, better recognition of delirium, and structured medication reviews built into annual dementia review. This is a study of prescriptions in a database. It contains no clinical rationale, no symptom severity and no record of what was tried first, so it cannot say which of these 9819 courses were justified and which were drift.

What it does give a clinician is a question with a date in it. When was this started, what was it started for, does that reason still exist, and when was it last reassessed. The restart finding sharpens the point: more than half of the people who came off went back on within about three months, so a stop that is not planned alongside what replaces it is a pause, not a decision.

Funding is public: the NIHR School for Primary Care Research. In the declaration of interests, one co-author reports honoraria from Lundbeck Neuratorium and Lilly, and another reports consultancy payments from the Wellcome Trust, Swiss Re and Juli together with a US patent application filed jointly with Juli. Lilly manufactures olanzapine, one of the four index drugs in this analysis; the declared interest therefore points against the paper's conclusion rather than towards it.