Clinical sleep data such as polysomnography (PSG) offer a wealth of physiological insights, informing clinical decision-making, sleep diagnosis, and research. Sleep wellness facility databases can support large data research projects and are a natural extension of already acquired data supporting patient care, which allows valuable and limited resources to be applied at the analysis phase. PSG data contains diversity and heterogeneity that may be specifically excluded in clinical trial databases, which are often designed to reduce sources of variability that can be detrimental to power calculations and outcome testing. Heterogeneous sets may also be more amenable to exploratory methods that allow discovery of new phenotypes that can be explored in subsequent prospective studies.
Clinical data insights might identify comorbidities or predict response to treatment of sleep disorders. The opportunity for advanced analysis of the data obtained in routine clinical practice cannot be overstated. Research about normal and pathologic sleep physiology might be derived from studying the relationship between sleep-disordered breathing events and heart rate variability, or about how the amount of rapid eye movement (REM) sleep varies depending on the presence of different medications and disease states.
There are four basic categories of data that support clinical phenotyping derived from using PSG databases. Inferential analysis and insights can also be obtained by combining data across categories.
- Standard PSG metrics and data types: The standard metrics in most clinical PSG reports are readily accessible in sleep facility databases without the need for extra processing. These include basic demographics and summary statistics of PSG scoring, such as stage percentages, total sleep time, efficiency, and apnea-hypopnea index. The importance of standardization in human scoring and basic metrics has been emphasized, especially for multicenter trials and data repositories involving PSG.
- PSG scoring annotations: PSG annotations include technician-scored labels for sleep-wake stage and various events often with time stamps. These data can be exported for off-line processing and/or combining with other sources of clinical data. Aligning these files with exported time series data allows stage- or event-specific analysis of physiological signals. Event label errors include errors of omission and of commission, and they are best assessed by manual rescoring.
- PSG time series data: Each channel of a standard PSG is a time series, to which a number of signal processing techniques can be applied to extract data. Initial preprocessing can involve detection and removal of periods with prominent muscle artifact, or removal of ECG signal contaminating electroencephalography (EEG) channels.
- Self-reported clinical data: Patients undergoing in-laboratory PSG are often asked to self-report symptoms, medical problems, and medications in the facility’s questionnaire. Most facilities use a custom form as a basic symptom and history screening tool, as well as checkboxes and boxes for free-text responses. When self-reporting methods are used, the data will require manual or semi-automated review and cleaning before analysis is possible. If medications are listed as free text, spelling errors or nonstandard terminology requires reconciliation.
Large sleep datasets offer the opportunity to detect scientifically or clinically interesting differences or patterns in health and disease. Despite these benefits, data analysis in sleep medicine also carries risks and certain limitations must be recognized. Academic centers, for instance, may have different referral biases being used for complicated cases. Although most clinical laboratories have standardized recording protocols, the collection of self-reported clinical data may not be standardized. Variations across recording and scoring technologists may contribute to errors, despite the quality efforts required in accredited laboratories and centralized scoring common to large clinical trials may not be practical for clinical databases.
Understanding these issues can help mitigate the risks, whether one is conducting the analysis or reviewing research involving data analytics. Ideally, what is learned from the data can be applied to inform and make individual clinical care decisions. Since at-home sleep testing is on the rise, these efforts will be critical to justify the current and ongoing use of PSG for clinical care. Data analytics in sleep medicine is poised to provide unprecedented insights, especially as it coincides with massive shifts in reimbursement and availability of laboratory-based PSG.
Somnoware is transforming sleep health management by accelerating diagnostics and increasing therapy adherence. To learn how you can automate sleep testing processes, obtain actionable data and predictive insights, and drive greater patient engagement, schedule a demo by clicking here.