JMIR Publications

ABSTRACT

Background:

Home-based sleep technologies enable longitudinal assessment in free-living settings, but heterogeneous sensors, proprietary algorithms, and asynchronous deployment complicate cross-device interpretation.

Objective:

We evaluated which sleep measures were most suitable for longitudinal within-device monitoring versus cross-device integration.

Methods:

We analyzed longitudinal SleepImage, Emerald, and Verisense data from the Boston University Alzheimer’s Disease Research Center digital brain health platform. Device outputs were harmonized to shared nightly measures. We summarized data yield, follow-up, and same-night and within-7-day overlap. Cross-device concordance was assessed using Pearson correlations and Bland-Altman analyses. Longitudinal monitoring suitability was evaluated using night-to-night variability, first-k-night reliability, and adjusted temporal drift models. Cross-device integration suitability was evaluated using mixed-effects translation models and variance decomposition. These components were summarized into two percentile-ranked composite scores: a long-term monitoring score and a cross-device integration score.

Results:

The sample included 227 SleepImage (median age 68 years; 61.9% female), 28 Emerald (median age 69 years; 38.9% female), and 68 Verisense (median age 70 years; 48.1% female) participants. Median recorded nights were 22.0, 87.0, and 14.5, and median follow-up time was 454.0, 92.0, and 38.5 days, respectively. Same-night overlap was highest for SleepImage-Verisense (339 pairs), followed by SleepImage-Emerald (84 pairs), and minimal for Emerald-Verisense (4 pairs). Sleep efficiency ranked highest for longitudinal within-device monitoring (score 0.78), supported by the highest 7-night reliability (0.91) and smallest mean relative drift (0.17% per 30 days). Total sleep time (TST) ranked highest for cross-device integration (score 0.78), with the highest same-night concordance (r = 0.22), statistically detectable translation in 2 of 4 estimable models, and low device-related variance (5.8%). In contrast, number of awakenings showed large device-related variance (31.0%), and wake after sleep onset showed the weakest cross-device concordance (r = 0.07).

Conclusions:

Under real-world asynchronous deployment, the three platforms provided complementary rather than interchangeable sleep information. Sleep efficiency was identified as best suited for longitudinal within-device monitoring, whereas TST was the strongest candidate for cross-device harmonization. Therefore, multimodal home sleep monitoring should prioritize intended analytic use rather than agreement with a single summary metric.