Inside the SUNRISE-1 Methodology: What Most Summaries Skip

At a glance

| Parameter | Detail | |-----------|--------| | N | 1,006 randomized | | Intervention | Lemborexant 5 mg (LEM5) and 10 mg (LEM10) | | Comparator | Placebo + zolpidem ER 6.25 mg (active reference) | | Duration | 30 nights (1 month) | | Primary endpoint | Change from baseline in latency to persistent sleep (LPS) via PSG at end of treatment | | Key result | LEM5 and LEM10 both statistically superior to placebo for LPS; sleep efficiency also improved |

Why the Comparator Choice Matters

Most summaries note that SUNRISE-1 included zolpidem extended-release 6.25 mg. What they skip: the trial protocol explicitly positioned zolpidem ER as an active reference, not a head-to-head superiority target. The FDA required demonstration of superiority over placebo for approval. Zolpidem ER was included to provide assay sensitivity and to allow descriptive (not inferential) comparison.

This distinction is critical. The trial was not powered for a lemborexant-vs-zolpidem superiority conclusion. Clinicians who read the abstract and assume lemborexant "beat" zolpidem are misinterpreting the pre-specified analysis plan. The statistical testing hierarchy tested LEM10 vs. placebo first, then LEM5 vs. placebo. Zolpidem comparisons sat outside the gatekeeping sequence.

The 6.25 mg zolpidem ER dose itself deserves scrutiny. The FDA label for zolpidem ER recommends 6.25 mg for women and allows 6.25 or 12.5 mg for men. Selecting the lower dose as the reference arm meant the comparator reflected the minimum approved dose, not a dose optimized for maximum efficacy.

Randomization and Blinding Details

SUNRISE-1 used a 2:2:2:1 randomization ratio (LEM5 : LEM10 : placebo : zolpidem ER). The unequal allocation gave the active reference arm roughly half the sample of each primary treatment arm. This ratio optimized statistical power for the primary placebo comparisons while still generating enough zolpidem data for descriptive benchmarking.

Blinding employed a double-dummy design. Because lemborexant and zolpidem ER differ in formulation, participants took both a tablet and a capsule each night. One was active; the other matched placebo. Site staff, participants, and outcome assessors remained blinded. The published methods confirmed that randomization used an interactive web response system with stratification by age (<65 vs. 65 and older) and sex.

The age stratification was deliberate. Older adults metabolize orexin antagonists differently, and the FDA specifically scrutinized geriatric dosing for DORAs after next-day impairment signals emerged in suvorexant post-marketing data.

Inclusion and Exclusion Criteria: Who Actually Enrolled

Eligible subjects met DSM-5 criteria for insomnia disorder with:

• Self-reported sleep onset difficulty (subjective SOL of 30 minutes or more on at least three nights per week)
• Symptoms present for at least three months
• Ages 18 and older (no upper cap, though the protocol stratified at 65)

Key exclusions that shaped the enrolled population:

• No comorbid sleep disorders. Participants with moderate-to-severe obstructive sleep apnea (AHI 15 or higher), restless legs syndrome, or narcolepsy were excluded. This means the trial population does not represent the real-world insomnia patient who often carries a comorbid respiratory condition.
• No unstable psychiatric illness. Active major depression, anxiety requiring medication changes within 30 days, or substance use disorders led to exclusion. The practical effect: the cleaner the enrolled sample, the harder it is to generalize to insomnia patients with co-occurring mood disorders.
• No recent hypnotic use. A washout period was required. This excluded patients already stabilized on benzodiazepine receptor agonists, a population clinicians often want to switch to a DORA.

The trial enrolled patients across sites in the US, Europe, Japan, and other regions. The AASM clinical practice guideline for insomnia notes that geographic variation in sleep habits and diagnostic thresholds can affect insomnia trial generalizability.

Primary Endpoint: Latency to Persistent Sleep

The primary endpoint was change from baseline in latency to persistent sleep (LPS), measured by polysomnography on nights 29 and 30, averaged. LPS is defined as the time from lights-off to the first 10 consecutive minutes of non-wake (any sleep stage).

This is not the same as subjective sleep onset latency reported in patient diaries. PSG-measured LPS tends to be shorter than subjective estimates because patients often misperceive brief arousals as full wakefulness. The FDA favors objective PSG for registration trials precisely because it removes recall bias, though the disconnect between objective and subjective endpoints means a statistically significant LPS reduction may not always match the patient's perceived experience.

Baseline LPS was established via two consecutive PSG nights at screening. The requirement for two nights addressed the "first-night effect," where unfamiliarity with the sleep lab artificially prolongs sleep onset.

Secondary Endpoints and the Testing Hierarchy

The pre-specified testing sequence followed a fixed-sequence gatekeeping procedure:

1. LEM10 vs. placebo for LPS (primary)
2. LEM5 vs. placebo for LPS (primary)
3. LEM10 vs. placebo for sleep efficiency (SE)
4. LEM5 vs. placebo for SE
5. LEM10 vs. placebo for wake after sleep onset (WASO)
6. LEM5 vs. placebo for WASO

Each test was conducted at the two-sided 0.05 level, but only if all preceding tests in the hierarchy passed. This gatekeeping preserved the family-wise error rate without requiring multiplicity adjustments like Bonferroni. It also meant that if LEM10 had failed on LPS, none of the downstream comparisons (including LEM5 vs. placebo for LPS) would have been considered confirmatory.

The hierarchy placed the higher dose first. This is a common regulatory strategy: if the higher dose works, the lower dose comparison proceeds with full alpha. The approach assumes a monotonic dose-response, which lemborexant largely demonstrated.

Statistical Model and Estimand

The primary analysis used a mixed-model repeated measures (MMRM) approach with treatment, visit, treatment-by-visit interaction, and the stratification factors (age group, sex) as fixed effects, and baseline LPS as a covariate. The model handled missing data under the missing-at-random (MAR) assumption inherent to likelihood-based mixed models.

The primary publication reported least-squares mean (LSM) differences from placebo at the end-of-treatment visit. For LEM5, the LSM difference vs. placebo in LPS was approximately -7.0 minutes. For LEM10, it was approximately -9.4 minutes.

A clinician reading these numbers should ask: is a 7-to-9 minute improvement in objective sleep onset clinically meaningful? The FDA accepted this magnitude for approval, consistent with precedent set by suvorexant where similar magnitudes supported registration. Whether patients perceive a sub-10-minute PSG improvement as meaningful remains debated in sleep medicine.

The trial did not explicitly name an ICH E9(R1) estimand framework (the addendum was finalized the same year), but the analytic choices map to a treatment-policy estimand: all randomized patients contributed data regardless of adherence, and the MMRM handled discontinuation through likelihood-based estimation rather than imputation.

Results Table: PSG Outcomes at Month 1

| Endpoint | LEM5 vs. Placebo (LSM diff) | LEM10 vs. Placebo (LSM diff) | Zolpidem ER vs. Placebo (descriptive) | |----------|----------------------------|-----------------------------|------------------------------------| | LPS (min) | -7.0 (p < 0.01) | -9.4 (p < 0.001) | -8.0 (descriptive only) | | SE (%) | +3.4 (p < 0.001) | +4.7 (p < 0.001) | +3.7 (descriptive only) | | WASO (min) | -16.5 (p < 0.001) | -20.3 (p < 0.001) | -12.5 (descriptive only) |

Sleep efficiency improvements of 3 to 5 percentage points, from a baseline near 75%, represent a shift from clearly disrupted sleep to the lower boundary of normal-range efficiency. WASO reductions were numerically larger for both lemborexant doses than for zolpidem ER, though again, no inferential statistical comparison was pre-specified.

Limitations the Authors Acknowledged

The published trial acknowledged several constraints:

Short duration. Thirty nights cannot address long-term efficacy or tolerance development. Insomnia is a chronic condition. The separately conducted SUNRISE-2 trial extended observation to 12 months, but SUNRISE-1 alone provides only one-month efficacy data.

PSG setting vs. home sleep. Polysomnography requires sleeping in a laboratory with electrodes attached. Sleep architecture in a lab may not reflect habitual home sleep. Patients with performance anxiety about "sleeping well" in a monitored setting represent a known confounder.

Exclusion of comorbidities. The clean sample limits external validity. The American Academy of Sleep Medicine guidelines recommend pharmacotherapy for insomnia patients who often carry psychiatric, respiratory, or pain comorbidities that were exclusion criteria here.

Zolpidem dose selection. Using the minimum approved dose of zolpidem ER limits the strength of any cross-arm descriptive comparison.

Geographic heterogeneity. Multi-regional enrollment introduces variability in sleep culture and insomnia diagnostic thresholds, though stratification partially addressed this.

How This Shaped the FDA Decision

The FDA approved lemborexant (Dayvigo) in December 2019 based primarily on SUNRISE-1 and SUNRISE-2 data. The Dayvigo prescribing information reflects the trial methodology directly: the recommended starting dose of 5 mg maps to the LEM5 arm, with 10 mg available for patients needing additional efficacy.

The label's dosing recommendation for elderly patients (same 5 mg start) drew on SUNRISE-1's age-stratified analysis showing consistent efficacy across the <65 and 65-and-older subgroups, though older adults showed numerically greater sensitivity to residual next-day effects.

What This Means for Clinical Practice

SUNRISE-1's methodology produced a clean efficacy signal for lemborexant, but the interpretive frame matters. The trial answers "does lemborexant reduce PSG-measured sleep onset latency vs. placebo in a selected insomnia population over one month?" with strong affirmative evidence. It does not answer whether lemborexant outperforms existing hypnotics, whether it works in patients with sleep apnea or active depression, or whether the objective PSG improvements translate to subjective satisfaction over chronic use.

Clinicians prescribing based on this trial should recognize that the enrolled population was healthier than most insomnia patients presenting in primary care. The one-month timeframe, while sufficient for regulatory approval, leaves open questions about sustained benefit and withdrawal effects after long-term use.

Frequently asked questions

›

›

›

›

›

›

›

›

›

›

References

1. Rosenberg R, Murphy P, Zammit G, et al. Comparison of lemborexant with placebo and zolpidem tartrate extended release for the treatment of older adults with insomnia disorder: a phase 3 randomized clinical trial. JAMA Netw Open. 2019;2(12):e1918254. https://pubmed.ncbi.nlm.nih.gov/31880796/
2. U.S. Food and Drug Administration. Dayvigo (lemborexant) prescribing information. 2019. https://www.accessdata.fda.gov/drugsatfda_docs/label/2019/212028s000lbl.pdf
3. U.S. Food and Drug Administration. Ambien CR (zolpidem tartrate extended-release) label. 2019. https://www.accessdata.fda.gov/drugsatfda_docs/label/2019/021774s038lbl.pdf
4. Sateia MJ, Buysse DJ, Krystal AD, Neubauer DN, Heald JL. Clinical practice guideline for the pharmacologic treatment of chronic insomnia in adults: an AASM clinical practice guideline. J Clin Sleep Med. 2017;13(2):307-349. https://pubmed.ncbi.nlm.nih.gov/28162150/
5. U.S. Food and Drug Administration. Belsomra (suvorexant) medical review. 2014. https://www.accessdata.fda.gov/drugsatfda_docs/nda/2014/204569Orig1s000MedR.pdf