Context

Obstructive sleep apnea (OSA) is a serious public health problem. Approximately 2 percent to 4 percent of middle-aged women and men, respectively, have this condition; the majority are undiagnosed. Undiagnosed and thus untreated, OSA is associated with significant morbidity and mortality. Effective treatment modalities should not be applied without an accurate diagnosis of OSA, but medical history and physical examination are insufficient to establish the diagnosis or its severity. Using the accepted reference standard test – attended, in-laboratory polysomnography (PSG) – can be expensive and involve long waiting times for studies, so various groups have developed portable technologies to classify patients in terms of the presence or absence of OSA and, for the former, level of severity. Such devices are intended for use in sleep laboratories or in the home.

Objectives

We updated a 2002–2003 systematic review of OSA diagnostic testing to address the key questions of how portable sleep testing devices compared to PSG in diagnosing OSA and, assuming equivalent effectiveness, what sleep and physiologic factors and what patient and technician conditions were important to measure or have in place. The Centers for Medicare and Medicaid Services commissioned the Agency for Healthcare Research and Quality to provide a technology assessment that addressed the following: 1. How does the diagnostic test performance of unattended portable multi-channel home sleep testing compare to facility-based polysomnography in the diagnosis of obstructive sleep apnea? a. If unattended portable multi-channel home sleep testing is as effective as polysomnography in the diagnosis of obstructive sleep apnea, which parameters of sleep and cardiorespiratory function (i.e., sleep staging, body position, limb movements, respiratory effort, airflow, oxygen saturation, electrocardiogram) are required? b. If unattended portable multi-channel home sleep testing is as effective as polysomnography in the diagnosis of obstructive sleep apnea, what conditions (i.e., patient education, technician support) are required so that it is done correctly in the home?

Data sources

We searched for studies published since the original review (i.e., from 2002 on) in MEDLINE, The Cochrane Library, the National Guidelines Clearinghouse, and the International Network of Agencies for Health Technologies Assessment (INAHTA) database; we also handsearched bibliographies of included articles. In MEDLINE, we used the following main terms in various combinations: polysomnography, oximetry, physiologic monitoring, and sleep apnea (with limits of human, adults, and English language); we refined searches using the terms airway resistance, upper airway resistance syndrome, respiratory disturbance index, autoset, snoring, and respiratory events related arousals as well as reproducibility of results, predictive value of tests, and sensitivity and specificity.

Study selection

We included studies of humans, both sexes, ages 18 and over, with any diagnosis of OSA; studies of any type of portable device used for diagnosis that also included a reference standard test (PSG or another acceptable test for diagnosing OSA); studies in which each analysis group, after the end of the study, included at least 10 subjects; and studies published in English. Specifically excluded were studies in which results from portable devices were not compared with results from PSG. Also excluded were reviews, meta-analyses, case reports, abstracts, letters, and editorials.

Data extraction

One investigator recorded abstracted data onto data abstraction forms used for the original review and created detailed evidence tables. A second investigator checked entries against the original articles. One investigator assigned initial classifications for level of evidence and presence or absence of eight quality indicators and a second investigator reviewed these; disagreements were resolved by consensus discussion. A third investigator combined level of evidence and quality indicators into a summary quality grade; the other investigators reviewed these grades, with differences resolved by consensus.

Data synthesis

We identified 172 unique titles and abstracts from the literature searches, and excluded 157 articles as not meeting inclusion criteria; reasons included the fact that PSG studies were not performed on all patients, that the portable device was an electroencephalogram (EEG), and that the study assessed a telemedicine approach that did not compare a portable device to the PSG. We obtained 15 articles for full review and retained 12 for inclusion here. These 12 studies fell into four categories: Type 3 devices used in laboratory settings (four studies); Type 3 devices tested in homes whether or not they were also tested in facilities (two studies); Type 4 devices in laboratory settings (six studies); and Type 4 devices tested in homes (whether or not in facilities, three studies). Type 3 devices include a minimum of four channels and must monitor at least two channels of respiratory movement or respiratory movement and airflow, and heart rate or ECG and oxygen saturation to define an event; generally, no electroencephalogram (EEG) signals are monitored. Level 4 devices include only one or two channels of physiologic signals and generally use only one channel (either saturated oxygen or airflow) to define a sleep-disordered breathing event; no EEG signals are monitored. Most articles provided only comparisons of the results from portable monitoring done simultaneously with full PSG in the laboratory, i.e., “a side-by-side” study. The in-laboratory simultaneous studies, which used technologies identical or similar to those in the previous review, produced sensitivity and specificity results for diagnosing OSA similar to those reported earlier; that is, the newer studies produced no meaningful changes in the level or quality of evidence for the effectiveness for home monitoring devices in diagnosing OSA. Only four of these studies (two of Type 3 and two of Type 4 devices) were graded good or fair quality. Ultimately, we focused on the five studies with in-home testing, because the questions we were asked concerned the effectiveness of unattended monitoring in the home. Four in-home studies employed technologies similar or identical to those reviewed before; of these, two studies (one of good quality, one poor) used Type 3 devices and two (one of fair quality , one poor) used Type 4 devices. Reported sensitivity and specificity values were similar to those from older studies, so the newer studies yielded no major information that would change the previous basic conclusions about portable devices used in the home. The one in-home study using a new technology, of fair quality, produced likelihood ratios that indicated that the test had little effect in changing pretest probabilities of the presence or absence of OSA. Reported data loss in the home studies ranged from a low of 3 percent to a high of 33 percent, in a subgroup of patients who did their own hookup. Automated scoring appeared to agree less closely with the reference standard than manual scoring. Internal validity of the five in-home studies was mixed: one study of good quality, two of fair quality, and two of poor quality. In terms of external validity, the patient populations were mostly male, middle-aged, and with high pretest probabilities of OSA; comorbidities were generally not specified or taken into account in analyses. Finally, these studies typically did not evaluate the accuracy of clinical management decisions based on portable results compared to those based on the reference standard.

Conclusions

This newer body of evidence does not materially change earlier findings regarding in-home devices for diagnosing OSA. Choices of cutoffs for determining OSA by AHI or RDI differed widely across these studies, making cross-study comparisons impossible. The better studies yielded sensitivity and specificity values (or LRs) that provided modes changes in the probability of OSA over the pretest probability. In studies that directly compared automated versus manual scoring from home monitoring devices, manual scoring correlated better with data from laboratory PSG than did automated scoring.