Abstract Introduction Clinical sleep studies typically rely on one night for OSA detection and diagnosis. However, uncertainty exists regarding the degree of AHI stability across different nights. Population studies collecting continuous nightly data on a large scale enable detection of night-to-night variability in OSA severity; this study is the largest to date for evaluation of the number of nights to achieve high sensitivity/specificity for OSA detection. Methods Sleep-disordered breathing was analyzed by a commercially available home monitoring device (Sleeptracker-AI Monitor, Fullpower Technologies Inc., California, USA) that passively monitors sleep using piezo-electric sensors. Validated sleep/respiratory parameters were derived from device data. De-identified data were analyzed, following review and exemption of the study (#57681) from Stanford University IRB. Data from 10/01/2021 to 09/30/2022 were reviewed in 96,228 individuals with 19,148,323 recorded nights. Individuals with at least 300 nights with a computed AHI (TST≥4hrs) were included in the analytic dataset. Results A total of 23,105 individuals (10,711 men, 49.6±12.93 years; 9,430 women, 49.4±12.85 years; 2,964 no gender provided, 50.4±14.37 years) with 7,947,840 recorded nights met the inclusion criteria. Maximum (but not mean) AHIs for each individual for 3 randomly-selected (vs. consecutive) nights revealed sensitivities (≥95%)/specificities [and CIs] of 94.9%[94.8,95.1]/82.5%[82.2,82.7], 95.2%[95.0,95.5]/94.3%[94.1,94.5], and 96.2%[95.8,96.6]/97.9%[97.8,98.0] for detecting any, moderate-to-severe, and severe OSA, respectively, as categorized by mean AHI over 1 year. Sensitivities for a single night were in the low 70s, and, for maximum AHI over two randomly-selected nights, near 90%, with a tradeoff for decreased specificities for 3+ nights. For moderate-to-severe and severe OSA subjects over 1 year, means of 5.1±7.7% and 5.4±7.3% nights, respectively, showed OSA severity decreases by 2+ categories (i.e., moderate to normal, severe to mild/normal). Conclusion Use of a noninvasive in-home monitoring device enabling collection and analysis of a large sample of sleep/respiratory data on a continuous nightly basis showed that a single night yields low sensitivity for detection of OSA and on average OSA severity decreased by 2 categories on 5% or more nights. Maximum (but not mean) AHI across 3 non-consecutive nights yields good sensitivities/specificities for moderate-to-severe and severe OSA detection. Ongoing work explores statistics for nights beyond 3 to further improve sensitivity/specificity. Support (if any)