ICU-SLEEP

Code repository for the Investigation of Sleep in the Intensive Care Unit (ICU-SLEEP) clinical trial — a single-center, phase 2, double-blind, three-arm, parallel-group, randomized controlled trial of dexmedetomidine for sleep promotion in non-ventilated medical/surgical ICU patients aged 50+.

• ClinicalTrials.gov: NCT03355053
• IRB: Mass General Brigham (MGB) #2017P000090
• Statistical Analysis Plan: signed 2024-06-19; included in the data publication on bdsp.io.
• Manuscript: in preparation (peer review submission upcoming) — citation will be added on acceptance.
• Data publication: bdsp.io project page (DOI to be added once assigned)

Where the data lives

The de-identified analytic data is not in this repository. Download it from the credentialed-access AWS S3 bucket once you obtain bdsp.io credentials:

s3://bdsp-opendata-credentialed/icu-sleep/

See docs/data-access.md for credentialing and download instructions.

The identified (PHI) version of the source data is retained internally by Brandon Westover at:

Brandon - PHI/Datasets/zz_Delirium_Encephalopathy/@Delirium_Encephalopathy_Data Roundup_Identified/ICU-SLEEP (Investigation of Sleep in the ICU)/

(in case it is needed for future re-deidentification or audit; not publicly accessible).

Repository layout

ICU_SLEEP/
├── paper-code/             R Markdown analysis pipeline reproducing the paper
│   ├── z0_RR_dat_mgt_did.Rmd     data management (CSV -> rds)
│   ├── z1_RR_tables.Rmd          baseline & outcome tables
│   ├── z2_RR_primary_analysis.Rmd     primary outcome (DFD ordinal logistic)
│   ├── z2_RR_sen_analysis.Rmd         sensitivity analyses
│   ├── z2_RR_sen_forest_plot.Rmd      forest plots
│   ├── z2_RR_survival_analysis.Rmd    time-to-event analyses
│   ├── plots-design-files/      DataGraph .dgraph design files for figures
│   └── data/                    populated by user from the S3 download
│
├── deidentification/       Safe-Harbor de-identification pipeline (raw XLSX -> public CSVs)
│   └── deidentify.R
│
├── ancillary-code/         Companion code from earlier ICU-SLEEP-related repositories
│   ├── icu-sleep-vitals/         Python tooling for plotting vitals (Bedmaster/HL7/EDW)
│   └── icu-sleep-preprocessing/  notebooks and scripts kept for posterity
│
└── docs/
    ├── codebook.md         data dictionary for the public CSVs
    ├── data-access.md      how to obtain the data from S3
    ├── reproducibility.md  how to reproduce the paper's numbers
    └── related-work.md     companion publication (noiselight)

Reproducing the paper

See docs/reproducibility.md for end-to-end instructions. In brief:

1. Download ICU-SLEEP_Enrolled_Deidentified.csv from S3 to paper-code/data/.
2. Open the R project; install the package set listed in paper-code/README.md.
3. Knit, in order: z0_RR_dat_mgt_did.Rmd → z1_RR_tables.Rmd → the four z2_*.Rmd analyses (each in 2-group and 3-group toggles).
4. Numerical outputs (Table 1, Table 2, primary OR/CI/P, secondary, sensitivity, survival, forest) appear in the rendered HTML notebooks. Figures are produced by pasting write_clip() outputs into the matching .dgraph files in plots-design-files/ (DataGraph required for figure rendering).

Related work

• noiselight (sound and light recordings from the same ICU course): https://github.com/bdsp-core/noiselight — published as Leone MJ et al., Chronobiol Int. 2023. Maintained as a separate repository.

License

Code: see LICENSE. Data: distributed via bdsp.io under their credentialed-access data use agreement.

Citation

See CITATION.cff. On acceptance of the manuscript, the citation will be updated with the published reference and the assigned data DOI.