Sleep Medicine

AI applied to overnight SpO2 shows high pooled sensitivity and specificity for OSA diagnosis.

Bayesian meta-analysis of 25 studies (23,171 participants) found pooled sensitivity 91.1% and specificity 88.4% for AI models using overnight SpO2 to detect OSA. Neural network classifiers performed best with sensitivity 92.7% and specificity 91.3%. Accuracy varied modestly by AHI threshold, with sensitivity decreasing and specificity increasing at higher AHI cut-offs. Authors note low–moderate bias and high GRADE quality, but recommend prospective external validation in diverse populations before clinical deployment. 1

Large wearable cohort defines age and sex patterns in objectively measured sleep duration.

In 274,128 US adults using a single consumer wearable, mean sleep duration was 7.57 hours with a 10th–90th range of 6.5–8.9 hours. Sleep was shortest in ages 40–49 and longest in 60–69 (p<0.001); women slept about +18 minutes longer than men on average. Weekend sleep exceeded weekday sleep by ~28 minutes, with the largest gap in ages 40–49. These device-based reference distributions can inform clinical expectations when counseling adults using similar consumer wearables. 2

Irregular sleep patterns linked to higher incident osteoporosis and amplify genetic risk.

In 87,231 UK Biobank participants followed for median 8.6 years, higher within-person SD of sleep parameters was associated with incident osteoporosis (top vs bottom quartile HRs ~1.17–1.21). A significant additive interaction with polygenic risk was observed; the highest sleep irregularity plus high PRS nearly doubled risk (HRs ~2.22–2.24). Improving sleep regularity attenuated the genetic risk, with the largest absolute benefit in intermediate PRS groups. Observational design limits causal inference; randomized trials are needed before recommending sleep regularity specifically for osteoporosis prevention. 3