ICU SLEEP

Abstract

Background. Sleep deprivation is common and severe among ICU patients and is hypothesized to be a key modifiable risk factor for delirium. Dexmedetomidine (Dex), an α_2A agonist, has been shown in prior trials to reduce ICU delirium, but the dosing strategy that best balances efficacy and safety is not established.

Trial. ICU-SLEEP (ClinicalTrials.gov NCT03355053) was a single-center, phase 2, double-blind, parallel-arm, placebo-controlled, three-arm randomized superiority trial conducted at Massachusetts General Hospital between January 2019 and March 2022. Non-ventilated adults aged 50 and older expected to remain in the ICU overnight were randomized 1:1:1 to one of three intravenous overnight (20:00-07:00, up to 7 consecutive ICU nights) infusions: low-dose dexmedetomidine (0.3 mcg/kg/h), very-low-dose dexmedetomidine (0.1 mcg/kg/h), or normal-saline placebo. The primary endpoint was in-hospital delirium-free days (IH-DFDs) over the first 14 hospital days; the secondary endpoint was ICU delirium-free days (ICU-DFDs) over the first 7 ICU days.

This release. 482 patients were enrolled, 347 received the intervention, and 312 entered the modified intention-to-treat analysis. This data and code release accompanies the primary trial manuscript (currently in preparation for peer-reviewed submission). It contains the de-identified per-patient dataset together with the analysis pipeline used in the paper, so that the published findings can be independently reproduced from the raw inputs. Please consult the published paper for the full results and clinical interpretation.

---

Background

Sleep in the ICU tends to be light, non-restorative, severely fragmented, and distributed across the day rather than consolidated overnight. Sleep deprivation is hypothesized to be a key modifiable risk factor for ICU delirium — a manifestation of acute encephalopathy that affects up to half of non-ventilated ICU patients and is a leading cause of preventable morbidity, mortality, and long-term cognitive impairment in hospitalized older adults.

Pharmacological options currently used to treat ICU sleep problems (benzodiazepines, antipsychotics, propofol) act on GABA_A circuits, do not produce natural sleep, and are themselves associated with increased ICU delirium. Dexmedetomidine, an α_2A adrenergic receptor agonist, engages sleep circuits without disrupting cortical dynamics in the same way and has been associated in prior work with reduced postoperative and ICU delirium incidence and duration. It is unknown whether this benefit is mediated through improved sleep, and the optimal nocturnal dosing strategy has not been established.

The ICU-SLEEP trial was designed to inform both questions by collecting, alongside delirium endpoints, dense longitudinal physiological, environmental, and neuropsychological data from each enrolled patient. The dataset released here is intended to support reproduction of the primary trial analysis and to enable secondary investigations of the relationships between sleep physiology, delirium, and cognitive outcomes in critically ill adults.

---

Methods

Trial design. Single-center, prospective, phase 2, double-blind, placebo-controlled, three-arm parallel-group, mechanistic, randomized superiority trial. Three nightly study arms (1:1:1 allocation): very-low-dose Dex (0.1 mcg/kg/h), low-dose Dex (0.3 mcg/kg/h), or normal-saline placebo, all delivered as intravenous overnight infusions from 20:00-07:00 (11 hours) for up to 7 consecutive ICU nights starting from the first study intervention.

Setting and population. Three Massachusetts General Hospital ICUs (an 18-bed medical ICU, a 20-bed surgical ICU, an 18-bed mixed medical/surgical ICU). Eligible patients were non-ventilated adults aged 50 and older, expected to remain in the ICU overnight. Detailed inclusion/exclusion criteria are given in the paper (Table S1) and the SAP (released alongside the data). Major exclusions: severe baseline cognitive impairment, focal neurologic disease/injury, serious cardiac disease, severe liver/renal dysfunction, blind/deaf or unable to use approved languages, pregnancy/breast-feeding, dexmedetomidine allergy or already on dexmedetomidine, on anticholinergics scopolamine/penehyclidine, or on alpha-2 agonist clonidine.

Randomization and blinding. Block randomization (varying, concealed block sizes; 1:1:1) was prepared by the trial biostatistician using R. Independent research pharmacists dispensed study drug or placebo per the randomization list. Drug and placebo were visually indistinguishable; allocations remained concealed from the trial team until trial completion. Unblinding was permitted only when essential for patient management or for DSMB safety review.

Trial periods. (1) Pre-treatment baseline (consent to first study intervention); (2) treatment (up to 7 consecutive ICU nights); (3) post-treatment in-hospital follow-up (through hospital day 14 from first intervention, or hospital discharge); (4) long-term follow-up (3, 6, 12 months post-enrollment).

Per-patient measurements.

• Baseline. Lawton iADLs, demographics, medical history (Charlson Comorbidity Index from ICD-10 codes), pre-existing dementia screen (IQCODE-SF, 16-item).
• Continuous physiological monitoring. Routine ICU vital signs (heart rate, respiratory rate, blood pressure, core body temperature), ECG, SpO₂; study-specific actigraphy and respiratory belt (Airgo CE Class IIa wearable, placed near the floating ribs) until ICU discharge.
• Environmental monitoring. Continuous bedside light and noise sensor data using the in-house sensor described in Leone et al. 2023 (Chronobiol Int 40(6):759–768).
• Twice-daily cognitive assessments (08:00–12:30 and 15:30–20:00, starting the afternoon before the first study intervention): Richmond Agitation-Sedation Scale (RASS), Confusion Assessment Method (CAM) and CAM-ICU, CAM-Severity (CAM-S).
• Daily morning self-report. Psychomotor vigilance task (PVT) via the ‘Vigilance Buddies’ iPad app; Numeric Rating Scale for Sleep (NRS-Sleep) for the prior night.
• Daily EMR extraction. SOFA score, weight, daily fluid I/O, drug administration events, and significant cardiac dysfunction.
• Neuropsychological battery (one session within 7 days of last intervention or ICU discharge): California Verbal Learning Test-2/3 Short Form (CVLT-II/III), WAIS-IV Digit Span subtest, Trail Making Test A and B (TMT-A/B).
• Long-term telephone follow-up at 3, 6, 12 months: NIA-LOAD Battery, Telephone Interview for Cognitive Status (TICS), modified Rankin Scale (mRS), EQ-5D-3L, Lawton iADLs, 6-Item De Jong Gierveld Loneliness Scale, CES-D-10 (10-item).

Endpoints. Primary: in-hospital delirium-free days (IH-DFDs) in pooled-Dex vs. placebo, defined as days without delirium during the first 14 hospital days from first study intervention. Secondary: ICU delirium-free days (ICU-DFDs), defined similarly during the first 7 ICU days. Delirium = any positive CAM or CAM-ICU on a given day. Pre-defined, rule-based imputation of missing delirium days is specified in the SAP (released).

Analysis approach. Modified intention-to-treat (mITT): all randomized patients who received the intervention and had at least one post-intervention delirium measurement. Treatment effects estimated with a multivariable proportional-odds ordinal logistic regression (POOLR) adjusted for age, baseline SOFA, ICU and hospital length of stay at enrollment, baseline delirium status, and severe cognitive impairment (IQCODE-SF mean ≥3.3). All analyses two-tailed, α=0.05; no multiplicity adjustment for secondary/exploratory outcomes (per standard practice). Per-protocol analysis on patients receiving 7 consecutive full nights (≥8h of the 11h infusion). Safety analysis on all patients receiving any intervention. All analyses in R; full SAP and code released.

Oversight. An independent Data and Safety Monitoring Board (DSMB), with critical-care, neurology, anesthesiology, and biostatistics expertise, met at least semiannually and was notified of all SAEs and AEs of special interest within 24 hours.

---

Data Description

The release contains both the analysis code and the underlying data.

Code. The consolidated analysis pipeline used to produce the primary trial manuscript is available in the GitHub repository bdsp-core/ICU_SLEEP. The repository's README documents how to clone, install dependencies, point at the data, and reproduce each figure and table in the paper. A related repository for the bedside light/noise sensor used during the trial is at bdsp-core/noiselight (see Leone et al. 2023).

Data. Data are hosted in AWS S3 at s3://bdsp-opendata-credentialed/icu-sleep/. Access is granted to credentialed users after acceptance of the BDSP Credentialed Health Data Use Agreement (DUA). The release contains the de-identified per-patient case-report-form (CRF) tables captured in REDCap during the trial:

• ICU-SLEEP_Enrolled_Deidentified.csv — 526 rows × 308 columns, one row per enrolled patient. Demographics, medical history (Charlson Comorbidity Index from ICD-10), baseline functional status (Lawton iADLs, IQCODE-SF), trial allocation, adherence summary, computed in-hospital and ICU delirium-free days (imputed and unimputed) used in the published primary and secondary analyses, neuropsychological-battery results (CVLT-II/III Short Form, WAIS-IV Digit Span, TMT-A/B), and long-term telephone follow-up at 3, 6, and 12 months (NIA-LOAD, TICS, mRS, EQ-5D-3L, Lawton iADLs, De Jong Gierveld Loneliness, CES-D-10, IQCODE-SF). This is the canonical input to the primary analysis pipeline.
• ICU-SLEEP_Delirium_Deidentified.csv — 7,364 rows × 189 columns, longitudinal delirium evaluations. Twice-daily RASS, CAM, CAM-ICU, and CAM-S assessments throughout each patient's trial timeline.
• ICU-SLEEP_Study_Drug_Deidentified.csv — 1,421 rows × 50 columns, per-night study-drug administrations (timing, dose, withholding events, adherence flags).
• ICU-SLEEP_Adverse_Events_Deidentified.csv — 2,878 rows × 18 columns, full AE / SAE log with severity, relatedness, and expectedness classifications.
• codebook.md — per-column data dictionary covering all four CSV tables.
• ICU-SLEEP_SAP_06.19.24_Signed.pdf — the pre-specified, signed statistical analysis plan as registered prior to unblinding.
• MANIFEST.txt — provenance and de-identification methods for each file.
• README.txt, README_repo.md — orientation documents.

De-identification. The release is HIPAA Safe Harbor compliant. Direct identifiers (names, MRN, ZID) were removed at source; DOB is blanked; ages 90 and over are capped at 89; all date/datetime values are shifted by a per-patient deterministic random offset in [−1095, +1095] days (~3 years), so every within-patient interval (length of stay, days-from-X-to-Y, survival days) is bit-identical to the un-shifted source. Free-text date fields are redacted. The per-patient shift table is held only by the study PI under PHI controls and is not part of this release.

Identified version. An identified version of the dataset (containing the protected health information that has been removed from this release) is retained by M. Brandon Westover in secure institutional storage (Box, BIDMC) for compliance reference.

---

Usage Notes

To reproduce the published trial analysis:

1. Clone the analysis code: git clone https://github.com/bdsp-core/ICU_SLEEP
2. After being granted credentialed access (sign the DUA on this page, then await approval), download the data with the AWS CLI:

   aws s3 cp s3://bdsp-opendata-credentialed/icu-sleep/ ./data --recursive

3. Place ICU-SLEEP_Enrolled_Deidentified.csv at paper-code/data/ and follow docs/reproducibility.md in the repository.
4. The other three CSVs (Delirium, Study Drug, Adverse Events) are provided for secondary analyses; they are not consumed by the primary pipeline as currently shipped, but are documented in codebook.md.

Notes for follow-on work:

• The signed SAP (ICU-SLEEP_SAP_06.19.24_Signed.pdf) and the trial registration (ClinicalTrials.gov NCT03355053) are the authoritative reference for endpoint definitions, imputation rules, and analysis decisions.
• This release accompanies the primary trial manuscript (in preparation). The trial's findings are reported in that paper; this release is intended to enable independent reproduction of those findings and secondary research.
• Because date offsets preserve within-patient intervals, all interval-based and survival analyses (e.g., LOS, time-to-event) reproduce exactly from the released data.

---

Ethics

This trial was approved by the Mass General Brigham Institutional Review Board (IRB #2017P000090) and the protocol was approved by the Partners HealthCare Human Research Committee. The trial was registered prospectively at ClinicalTrials.gov (NCT03355053) prior to enrollment, with the final statistical analysis plan uploaded before group assignments were unmasked. A licensed physician investigator obtained written informed consent from each participant. When a patient was unable to consent (e.g., due to delirium or coma), written informed consent was obtained from an appropriate surrogate or legally authorized representative (LAR). An independent Data and Safety Monitoring Board provided ongoing trial oversight, including real-time review of adverse events of special interest.

---

Acknowledgements

We thank the patients who participated in this study, and the physicians, nurses, and support staff across the Ellison 4, Blake 7, and Blake 12 ICUs and general care units at Massachusetts General Hospital who contributed to their clinical care. We also thank the members of the trial Data and Safety Monitoring Board for their oversight of trial conduct and safety.

---

Conflicts of Interest

M. Brandon Westover is a co-founder of, scientific advisor to, consultant for, and has a personal equity interest in Beacon Biosignals. Oluwaseun Akeju has a provisional patent application describing the use of alpha-2 agonists for promoting N3 sleep. The remaining authors declare no conflicts of interest related to this work. This trial was supported by the National Institute of Neurological Disorders and Stroke of the National Institutes of Health (NIH R01NS102190).

---

References

1. Tesh RA, Deng H, et al. Investigation of Sleep in the Intensive Care Unit (ICU-SLEEP): A Randomized, Double-blind, Parallel-arm, Placebo-controlled Phase 2 Clinical Trial. Manuscript in preparation. ClinicalTrials.gov NCT03355053.
2. Leone MJ, Dashti HS, Coughlin B, et al. Sound and light levels in intensive care units in a large urban hospital in the United States. Chronobiol Int. 2023 Jun 3;40(6):759-768. doi:10.1080/07420528.2023.2207647. PMID:37144470. PMCID:PMC10524721.

---