Refining Sleep‐Disordered Breathing Annotations Across Multiple Public Sleep Study Datasets
Bibliographic record
Abstract
Polysomnography annotations in Sleep Heart Health Study (SHHS), Osteoporotic Fractures in Men Study (MrOS), and Multi-Ethnic Study of Atherosclerosis (MESA) scored apneas and hypopneas solely by flow reduction rather than following the American Academy of Sleep Medicine's (AASM) comprehensive criteria. To address this, we developed a standardized annotation pipeline that integrates sleep staging, oxygen desaturation, and arousal events in accordance with AASM guidelines. This retrospective study analyzed polysomnography data from SHHS1 (n = 5793), SHHS2 (n = 2651), MrOS1 (n = 2907), MrOS2 (n = 1026), MESA (n = 2054), and Korea Image-based Sleep Study (KISS) (n = 7745). We compared reported apnea-hypopnea indices (AHIs) with those derived from original annotations and recalculated values adjusted for sleep stage, desaturation, and arousal. The impact of precise annotation was demonstrated by training two deep learning models, one with original and the other with refined annotations, and comparing their performance in classifying obstructive sleep apnea (OSA) severity. AHIs from original annotations consistently overestimated reported values in SHHS, MrOS, and MESA, with mean absolute errors (MAEs) ranging from 10.3 to 23.6 events/h. After refining the annotations, MAEs were reduced significantly to 0.56-1.29 events/h. KISS, adhering to contemporary scoring guidelines, exhibited high baseline accuracy with an MAE of 0.6 events/h and required no additional refinement. With refined annotations, OSA severity classification F1 score rose from 0.5 to 0.69. Our standardized approach improves cross-cohort consistency, supports both clinical research and AI-based analysis, and enables more reliable use of existing sleep datasets in accordance with current clinical guidelines.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.010 | 0.013 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.002 | 0.004 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.004 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".