A small insomnia trial asked when in the day a sleeping pill helps — and when it hurts

Insomnia is diagnosed partly on what happens during the day — the fatigue, the fogginess, the irritability that follow a bad night. Yet drug trials for insomnia have mostly measured the night: how fast people fall asleep, how long they stay asleep, and how they rate the whole two-week experience on a questionnaire at the end.

A randomized trial published in JAMA Network Open on January 5 tried measuring the day instead, and the two measurement approaches disagreed [s1].

What was done

Forty older adults with chronic insomnia were randomly assigned 1:1 to suvorexant or placebo [s1]. Mean age was 67.9 years and 90% of participants were women [s1]. The dosing schedule was 10 mg for two nights followed by 20 mg nightly for 14 nights, for 16 days of treatment in total [s1]. Data were collected between October 9, 2023 and August 8, 2024 [s1].

Alongside the usual end-of-study questionnaires, participants answered short smartphone surveys four times a day — a method called ecological momentary assessment, which asks people how they feel now rather than how they remember feeling over the past fortnight. Completion was 93.3% across all surveys [s1].

The two measurement systems gave different answers

On the Insomnia Severity Index, the standard instrument, the suvorexant group improved by 9.6 points (SD 5.4) and the placebo group by 5.5 points (SD 6.8); the between-group difference was statistically significant at p=.04 [s1].

Every other conventional questionnaire found nothing. Between-group differences were not significant on the Epworth Sleepiness Scale (p=.11), the PHQ-9 for depressive symptoms (p=.97), the GAD-7 for anxiety (p=.22), or the Fatigue Severity Scale (p=.28) [s1]. Read on its own, that panel would support a conclusion many insomnia trials have reached: the drug improved sleep, and daytime function was unchanged.

The smartphone data did not support that conclusion. Fatigue and sleepiness differed significantly by time of day between groups (χ²(4)=21.43, p=.003; adjusted p=.001), as did alert cognition (χ²(4)=11.12, p=.03; adjusted p=.05) [s1]. Negative mood did not (χ²(4)=3.82, p=.43) [s1].

The direction matters more than the p-values. Compared with placebo, suvorexant increased fatigue in the morning but reduced it in the afternoon and evening [s2]. Alert cognition was lower early in the day in the suvorexant group and normalized as the day went on [s2]. Participants taking suvorexant reported numerically worse mood at all four times of day, though that difference was not statistically significant [s2].

A single end-of-day or end-of-study rating averages those two opposing effects into approximately nothing. That is a plausible explanation for why the conventional questionnaires were flat.

What this cannot establish

Forty participants is a small trial, and the authors are explicit about the limits. They write that the generalizability of a nonrandom sample of generally well-educated individuals, primarily women, is unknown, and that future work should seek to replicate the findings among both women and men [s1]. Ninety percent female enrollment in a 40-person study leaves four men [s1].

The trial was also short — 16 nights [s1] — so it says nothing about what the morning fatigue signal looks like after months of use, which is how insomnia medication is often taken in practice.

And the finding is about measurement as much as about the drug. The claim the data best support is narrow: in this sample, repeated within-day sampling detected a time-varying pattern that recall questionnaires did not. Whether that pattern is clinically meaningful — whether morning grogginess offset by afternoon relief is a trade patients would accept — is not something a 40-person trial answers.

The study was funded through an investigator-initiated studies programme run by the drug's manufacturer [s2]. That is disclosed rather than hidden, and investigator-initiated funding is common, but it belongs in the reader's assessment.

Why the method is the story

Suvorexant belongs to a class that works by blocking orexin signalling, and questions about next-morning residual effects are not new for sleep medicines generally. What is new here is a trial design that can locate an effect in time rather than averaging it away.

If the pattern replicates in larger and more balanced samples, it implies that "no daytime effect" conclusions from older insomnia trials may partly be an artifact of how daytime effects were measured. That would be a substantive change in how this literature should be read.

If it does not replicate, this remains what it currently is — a feasibility result showing that older adults will answer smartphone prompts at a 93.3% rate [s1], attached to an exploratory signal.

What to watch

Whether regulators or trial sponsors adopt within-day sampling as a secondary endpoint in insomnia studies; whether the morning-fatigue and afternoon-relief pattern appears in trials with more men and longer exposure; and whether the same method, applied to other insomnia treatments including behavioural ones, produces comparable time-of-day structure.

This article describes a single small randomized trial and is informational only. It is not medical advice and does not recommend any medication, dose, or change to treatment.

Sources

• [s1] Smartphone-Based Real-Time Assessment of Daytime Insomnia Symptoms With Suvorexant: A Randomized Clinical Trial, JAMA Network Open, 2026-01-05.
• [s2] Improving sleep isn't enough: researchers highlight daytime function as key to assessing insomnia treatments, University of Maryland School of Medicine, 2026-01-08.