Evaluation of the Diagnostic Capacity of a Smart Mattress Versus Conventional Polysomnography

Overview

This project aims to develop and evaluate an innovative, non-invasive diagnostic system based on a smart mattress for detecting obstructive sleep apnea (OSA), as well as assessing overall sleep quality and identifying periodic limb movements. The main goal is to improve the accuracy of sleep apnea diagnosis while providing a less invasive solution suitable for home use, ultimately enhancing patients' quality of life. A descriptive, observational, prospective study will be conducted to analyze data obtained from diagnostic polysomnographies performed at the Sleep Unit of San Pedro Hospital between November 17, 2026, and March 1, 2028. Patients will use the smart mattress, and its measurements will be compared with polysomnography results. This comparison will allow for the optimization of the mattress's artificial intelligence, training it to accurately recognize respiratory patterns and sleep-related events, including positional apneas and periodic limb movements. Key technical objectives include: Determining the sensitivity, specificity, and predictive values of the mattress in detecting apneas, hypopneas, and limb movements compared to polysomnography. Evaluating the agreement between the mattress and polysomnography for sleep variables such as total sleep time, sleep efficiency, sleep stages, micro-arousals, and patient position. Assessing whether measurement accuracy varies by sleeping position or age group (adults vs. children). Measuring subjective sleep quality using the Groningen Sleep Quality Scale (GSQS-8). Performing a descriptive analysis of patient demographics. Hypotheses: The smart mattress will detect obstructive sleep apnea, sleep quality, and periodic limb movements with accuracy comparable to polysomnography. The system will provide a reliable, non-invasive, home-friendly diagnostic method. Measurements of the apnea-hypopnea index (AHI) and limb movements will show high sensitivity, specificity, and predictive values, both overall and according to OSA severity. There will be good agreement between mattress measurements and polysomnography for most sleep variables. Accuracy may vary depending on the patient's sleeping position. Measurements will correlate well across adults and pediatric patients. Subjective sleep quality scores (GSQS-8) will be consistent with objective mattress data. This project seeks to develop a more accurate, accessible, and non-invasive diagnostic system for OSA, combining advanced technology with ease of home use. By training the mattress's AI to recognize sleep patterns and events, it aims to optimize the detection of positional apneas, providing patients with better monitoring, early intervention, and improved quality of life.

This project corresponds to a descriptive, observational, and prospective study whose objective is to validate the functioning, accuracy, and clinical applicability of an intelligent mattress designed for the non-invasive detection of sleep-related respiratory disturbances, in comparison with level I polysomnography (the gold standard for OSA diagnosis). This phase constitutes the first stage of the global project aimed at the diagnosis and treatment of obstructive sleep apnea (OSA) through advanced monitoring and postural-adaptation technologies. During the period between November 17, 2026, and March 1, 2028, all polysomnographies performed in the Sleep Unit of Hospital San Pedro will be incorporated into a systematic registry together with the data simultaneously generated by the intelligent mattress. The aim is to determine the level of agreement between both systems, validate the diagnostic utility of the mattress, and generate the database required for the subsequent development of artificial intelligence algorithms for the automatic identification of respiratory events and sleep stages. 1. REGISTRY PROCEDURES AND DATA QUALITY The centralized registry will include all polysomnographic studies performed with Natus equipment, stored on its internal server, and the parallel recordings from the intelligent mattress. The registry structure will be designed to ensure data traceability, integrity, and quality, with specific procedures for technical validation, clinical verification, and coherence control. 1.1 Quality Assurance Plan A quality assurance system will be established based on four pillars: 1. Initial technical validation Each polysomnography will be reviewed by a sleep technician who will confirm: * Correct sensor placement. * Absence of recording failures. * Presence of at least 4 valid hours of sleep. * Adequate synchronization with the mattress. 2. Expert clinical review A certified sleep-medicine specialist will manually score each PSG according to AASM 2022 recommendations, including: * Classification of apneas and hypopneas. * Detection of micro-arousals. * Analysis of sleep architecture. * Quantification of supine and non-supine time. * Determination of global AHI, supine AHI, and non-supine AHI. 3. Internal audit Ten percent of the records will be reviewed by a second independent evaluator to estimate inter-observer agreement. Discrepancies \>10% in respiratory indices will trigger consensus sessions. 4. Mattress data integrity review The system will automatically verify: * Temporal continuity of the recording. * Presence of stable BCG sensor signal. * Absence of failures in air chambers. * Integrity of the exported file. 1.2 Automated Data Checks Mattress and PSG data will be subjected to validation through automated rules: * Range rules: * Oxygen saturation between 50-100%. * Respiratory rate between 6-40 rpm. * Heart rate between 35-180 bpm. * Total sleep time between 2-12 h. * Internal coherence rules: * Percentages of N1, N2, N3, and REM must sum ≤100%. * Number of events must be compatible with calculated AHI. * Time in bed must match total recording time. * Temporal consistency rules: * Synchronization between PSG and mattress ±5 seconds. Records generating alerts will be manually reviewed and classified as: * Valid. * Valid with warnings. * Unusable. 1.3 Source Data Verification (SDV) A source-data verification process will be implemented, including: * Comparison of respiratory events detected by the mattress with those manually annotated in PSG. * Cross-checking of position (supine/non-supine) between both systems. * Verification of demographic variables against the clinical record. * Concordance of sleep time and arousals. 2. REGISTRY DATA DICTIONARY The study will include a comprehensive data dictionary specifying: * Variable name. * Operational definition. * Unit of measurement. * Method of acquisition. * Source (PSG, mattress, or derived) * Standard coding (AASM 2022, MedDRA for events) * Expected physiological ranges. Examples: * Global AHI: total number of apneas + hypopneas divided by sleep time in hours. * Micro-arousals: defined according to AASM as EEG frequency shifts ≥3 seconds. 3. STANDARD OPERATING PROCEDURES (SOPs) The study includes documented SOPs for: Installation, recording, and disconnection of PSG and mattress. Manual scoring of respiratory and leg events. Data export, anonymization, and archiving. Synchronization and technical verification. Data inconsistency management and queries. Quality control and internal auditing. Classification of incomplete records. Security, privacy, and GDPR compliance. Each SOP specifies responsibilities, operational steps, acceptance criteria, and incident-handling mechanisms. 4. SAMPLE SIZE ASSESSMENT The planned sample size is 500 complete records. This calculation is based on: \- An estimated accessible population of \~700 PSGs in the reference period. \- A 95% confidence level. * A 1% margin of error. * The need for high statistical power for concordance comparisons. * Requirements for training AI models with a significant volume of observations. 5. MISSING DATA MANAGEMENT PLAN Missing data will be classified as: • Technical missing: signal loss, sensor failures. • Poor-quality missing: valid time \<4h. • Inconsistency missing: impossible ranges or temporal discrepancies. • Administrative missing: export errors. Criteria: \- Exclusion of records with valid time \<4h. * Simple imputation for position data when outcomes are unaffected. * Multiple imputation only for AI model training. * Documentation of missing-data reasons in the incident log. 6. STATISTICAL ANALYSIS PLAN The analysis will include: * Descriptive statistics: means, standard deviations, medians, ranges, and frequencies. * Correlation: Pearson or Spearman coefficients between PSG and mattress. • Agreement: Bland-Altman analyses for: \- Global AHI. \- Supine/non-supine AHI. \- Periodic limb movements. -Sleep efficiency. * Diagnostic analysis: \- Sensitivity. \- Specificity. \- PPV. \- NPV. \- ROC curves and AUC. * Predictive models: \- Logistic regression. * CHAID trees. * Multivariate classification. * Multiple-comparison adjustment: Benjamini-Hochberg (FDR). Analysis will be conducted using SPSS and R. The significance level will be α = 0.05. This collection of procedures ensures the registry meets the necessary requirements for scientific validity, reproducibility, and the future development of automated AI-based models, while maintaining applicable clinical and regulatory standards.