Real-world deployment of remote sleep monitoring technologies reveals distinct patterns associated with cognitive decline

This is the repository associated with the paper "Real-world deployment of remote sleep monitoring technologies reveals distinct patterns associated with cognitive decline".

Files

Here, we provide a description of the files made available in this repository.

The Models

To promote the sharing of resources, we provide the pre-trained age estimation model ('age_estimator.json') and final risk prediction model ('risk_predictor.pkl') described in the paper.

Scripts

This folder contains all associated code (including scripts for data pre-processing and stratification).

Data

The data presented in this study came from three separate sources. The Withings dataset was provided under a data-sharing agreement for research with Imperial College London and is not publicly available. A subset of the Minder dataset has been made publicly available and can be found on Zenodo at: https://zenodo.org/records/7622128. A full description of this data subset is published in Nature Scientific Data and can be found here: https://doi.org/10.1038/s41597-023-02519-y. The extended Minder dataset is available from the corresponding authors upon reasonable request. The Resilient dataset has been made publicly available and can be found on Zenodo at: https://zenodo.org/records/16755408.

Experiments

Code for experiments and figures presented in this study will be made available by the corresponding author upon reasonable request.

Set-up

For this, you will need to have conda installed (find more information here: https://docs.conda.io/projects/conda/en/latest/user-guide/install/index.html)

Create the environment from the environment.yml file:

conda env create -f environment.yml

Activate the environment:

conda activate sleep-age

Verify that the environment was installed correctly:

conda env list

Running the Model

To generate outputs on your own data, you can run the notebooks 'age_estimation.ipynb' and 'risk_prediction.ipynb'. These notebooks can be found in their respectively named folders.

The 'age_estimation.ipnyb' notebook allows you to load our pre-trained 'age_estimator.json' model, with which you can then estimate age on your own dataset. You can then calculate Sleep Age Index (SAI) for each of your inputs using the pre-calculated age-group weighted mean estimation errors in 'weighted_means_age.csv'.

The 'risk_prediction.ipynb' notebook allows you to to generate risk scores on your unlabelled SAI data using our pre-trained 'risk_predictor.pkl' model, to which you can then assign stratified group labels. Finally, you can then calculate adjusted probability scores for each of your inputs using the pre-calculated age-group weighted mean risk scores in 'weighted_means_dementia.csv', to which you can then assign updated stratification group labels.

Citation

If you use this code in any way, please refer to it by citing my paper "Real-world deployment of remote sleep monitoring technologies reveals distinct patterns associated with cognitive decline":

• Bibtex:

@techreport{Fletcher-Lloyd,
	author={Nan Fletcher-Lloyd and Nathalia Céspedes Gómez and Alexander Capstick and Antigone Fogel and Marirena Bafaloukou and Mahan Heydari and Alexandra Cairns and Chloe Walsh and Jessica True and Behnam Shariati and Ramin Nilforooshan and Payam Barnaghi},
	year={2025},
	month={Sep 3},
	title={Real-world deployment of remote sleep monitoring technologies reveals distinct patterns associated with cognitive decline},
	institution={Cold Spring Harbor Laboratory Press},
	url={https://www.medrxiv.org/content/10.1101/2025.08.29.25334735},
	doi={10.1101/2025.08.29.25334735}
}

Contact

This code is maintained by Nan Fletcher-Lloyd.