Oximetry

Sensitivity, specificity, and/or correlation of oximetry results to standard PSG results were reported in 12 studies. In 3 of these studies (#10453- Rodriguez Gonzalez-Moro, de Lucas Ramos, Sanchez Juanes, et al., 1996, #10636- Series, Marc, Cormier, et al., 1993, and #9255- Gyulay, Olson, Hensley, et al., 1993), oximetry was measured separately (in time and different setting) in place from the PSG, including overnight at home in 2 studies (#10636- Series, Marc, Cormier, et al., 1993 and #9255- Gyulay, Olson, Hensley, et al., 1993) and on 2 different nights in 1 study (#9255- Gyulay, Olson, Hensley, et al., 1993). In the other 9 studies oximetry was measured during the nocturnal PSG. MetaWorks investigators did not include in this analyzable set any studies where results from the oximetry channel on a multi-channel portable device were compared with PSG results. All analyzable oximetry studies are listed in Evidence Table 4. The publication dates spanned 1986 to 1997. There were 1,784 patients in total, 1,756 of whom were suspected of having SA. The number of diagnosed SA patients was not reported in all studies. Their average age was 50.6 (k=11 studies reporting) and the percentage of patients who were male was 83 percent (k=10 studies reporting). The average BMI was 31.7 kg/m².

Table

Evidence Table 4: Oximetry studies.

These oximetry studies used various formats for presentation of results: hourly frequency of desaturations of 3 percent or 4 percent, frequency of desaturations less than 90 percent, O₂ variability, or patients exceeding a certain number of desaturations per hour or per night. The type of probe was not consistently reported. Overall sensitivity of oximetry ranged from 36 percent to 100 percent, and specificity ranged from 23 percent to 99 percent, providing moderate sensitivity and specificity, with varying AI/AHI thresholds. The overall meta-analytic estimates (with Standard Error) of sensitivity and specificity are 87.4 percent (± 3.8) and 64.9 percent (± 6.7), and the summary ROC curve was generated (Figure 1). This curve shows that the studies all fell close to the estimated curve indicating little heterogeneity. The evidence score ranged from 16 to 32, and the mean score was 20.

Figure

Figure 1: Oximetry tests: Summary ROC curve and 95% confidence interval.

Partial Time Polysomnogram

There were 7 studies reporting results with sensitivity, specificity, and/or correlations of partial night or day PSGs relative to full night, standard PSGs. All PSGs were performed in sleep laboratories with the standard array of physiologic monitors. Four studies compared partial night to full night PSGs, and the other 3 studies compared daytime PSGs to full night PSGs. These studies included 505 patients in total, most of whom were suspected of SA. The number of patients with a diagnosis of SA was not completely reported. Their average age was 51.4 (k=7 studies reporting) and the percentage of patients who were male was 86 percent (k=6 studies reporting). The average BMI was 33.9 kg/m².

Of the 4 studies (# 8685- Carmona Bernal, Capote Gil, Sanchez Armengol, et al., 1994, #12211- Fanfulla, Patruno, Bruschi, et al., 1997, #10533- Sanders, Black, Costantino, et al., 1991, #10572- Scarf, Garshick, Brown, et al., 1990) with comparisons of partial night to full night PSGs, all used AHI as the PSG metric for diagnosis of SA. One also provided results using AI. One of these studies reported only correlations, not sensitivity or specificity. Of the remaining 3 studies, the sensitivity of the partial night PSG ranged from 42 percent to 93 percent, and the specificity ranged from 70 percent to 100 percent. However, these ranges reflect varying AHI thresholds for diagnosis of SA.

Of the 3 studies (#11854- Persson and Svanborg, et al., 1996, #10631- Series, Cormier, and La Forge, 1991, #12559- Van Keimpema, Rutgers, and Strijers, 1993) of daytime PSG compared to full night PSG, 2 used AHI and 1 used AI as the PSG metric for diagnosis of SA. The sensitivity of the daytime PSG for results of full night PSG ranged from 66 percent to 100 percent, and the specificity ranged from 50 percent to 100 percent, again depending upon the AHI or AI thresholds used for diagnosis.

Results for all 7 studies are displayed in Evidence Table 5, and the summary ROC curve derived from these studies is presented in Figure 2. Most studies were quite homogeneous with one exception which had low sensitivity and extremely high specificity. Sensitivity at AI/AHI threshold of 5 was 69.7 percent (± 5.3) and improved at a threshold of 10 to 79.5 percent (± 5.2). Specificity at AI/AHI threshold of 5 was 87.4 percent (± 5.4) and at the higher threshold of 10, changed little, at 86.7 percent (± 4.6). At still higher AI/AHI thresholds, there were too few studies with analyzable results. The average evidence score of all 7 studies was 18.6, with a narrow range, from 17 to 20.

Table

Evidence Table 5: Partial time polysomnogram studies.

Figure

Figure 2: Partial PSG tests: Summary ROC curve and 95% confidence interval.

Partial Channel Polysomnogram

In 3 studies (#9778- Lloberes, Montserrat, Ascaso, et al., 1996, #8687- Carrasco, Montserrate, Lloberes, et al., 1996, #9098- Garcia Diaz, Capote Gil, Cano Gomez, et al., 1997) results of a partial set of PSG channels monitored for a full night were related to the full channel, full night PSG results. In all 3 studies, oximetry, airflow, and thoracoabdominal movement were recorded. In 2 studies, patients were monitored on 2 different nights, and in the third study, same night results were compared using respiratory channels versus full PSG. These studies totaled 213 patients with suspected or confirmed SA. Their average age was 51.7 (k=3 studies reporting) and the percentage of patients who were male was 81 percent (k=3 studies reporting). The average BMI was 32.0 kg/m². Sensitivity ranged from 82 percent to 94 percent and specificity from 82 percent to 100 percent. The sensitivity and specificity of partial channel and partial time PSGs appeared most promising as possible prescreening tests or replacements for full PSG. There were too few studies to meta-analyze. The average evidence score was 17.7 (17, 17, and 19). The results are displayed in Evidence Table 6.

Table

Evidence Table 6: Partial channel polysomnogram studies.

Portable Devices

In total there were 25 portable device studies with sensitivity, specificity, and/or correlations to standard PSG. In terms of sensitivity and specificity, these studies were variable due to study and device heterogeneity. These studies enrolled 1,631 patients, and 1,368 were suspected SA patients at entry. Their average age was 48.5 (k=24 studies reporting) and the percentage of patients who were male was 84 percent (k=21 studies reporting). The average BMI was 30.0 kg/m².

Of the 854 suspected SA patients whose subsequent diagnosis was reported, 500 (58.5 percent) were diagnosed with SA, using an AI/AHI threshold of > 5/hr. In all except 2 studies (#12154- White, Gibb, Wall, et al., 1995 and #10565- Schafer, Ewig, Hasper, et al., 1997), the portable device results were only available as measured in the setting of a sleep laboratory, and not at home, where they are generally intended for use. Devices were issued from different manufacturers, thresholds used for diagnosis of SA varied from 5 to 40 (AI or AHI) per hour, and results were reported in different formats. The average evidence score was 22.1 (range, 16 to 34). Details of these studies are listed in Evidence Table 7.

Table

Evidence Table 7: Portable device studies.

Studies listed in Evidence Table 8 were categorized according to the channels monitored by portable devices.

Table

Evidence Table 8: Portable device measurement parameters.

The first 5 studies measured oximetry, snoring sounds, heart rate, and body position (#8967- Esnaola, Duran, Infante-Rivard, et al., 1996, #11620- Koziej, Cieslicki, Gorzelak, et al., 1994, #10464- Roos, Althaus, Rhiel, et al., 1993, #10786- Stoohs and Guilleminault, 1992, and #10565- Schafer, Ewig, Hasper, et al., 1997). These studies included 444 patients total. Figure 3 gives the summary ROC curve for these studies.

Figure

Figure 3: Portable device with oximetry, heart rate, snoring, and body position: Summary ROC curve and 95% confidence interval.

In 2 studies (#10785- Stoohs and Guilleminault, 1990 and #10380- Rauscher, Popp, and Zwick, 1991) totaling 103 patients, only 2 channels were monitored: snoring sounds and heart rate.

Four studies (#8605- Bradley, Mortimore, and Douglas, 1995, #9063-Fleury, Rakotonahary, Hausser-Hauw, et al., 1996, #9218- Gugger, 1997, #9575- Kiely, Delahunty, Matthews, et al., 1996) recording airflow and oximetry enrolled a total of 178 patients. Figure 4 gives the summary ROC curve for these studies.

Figure

Figure 4: Portable device with airflow and oximetry: Summary ROC curve and 95% confidence interval.

The next 6 studies tested portable devices monitoring oximetry, airflow, respiration, and heart rate (#11170- Zucconi, Ferini-Strambi, Castronovo, et al., 1996, #12225- Parra, Garcia-Esclasans, Montserrat, et al., 1997, #10390- Redline, Tosteson, Boucher, et al., 1991, #9836- Man and Kang, 1995, #12154- White, Gibb, Wall, et al., 1995, and #8953- Emsellem, Corson, Rappaport, et al., 1990). Some of them also included measurements of body position, body movement, and snoring sounds. There were 436 patients enrolled in these studies. The summary ROC curve for the 4 studies which included airflow, respiration, oximetry, heart rate, and body position are shown in Figure 5.

Figure

Figure 5: Portable device with airflow, respiration, oximetry, heart rate, body position, and snoring: Summary ROC curve and 95% confidence interval.

The remaining 8 studies could not be grouped by channels.

Non-sleep Tests

There were 17 studies that provided sensitivity, specificity, and/or correlations of results of some clinical measure in relation to standard PSG results. Pulmonary function tests and spirometry were included in this set of studies.

Flow volume loops. Four studies reported results of flow volume loops (#9373- Hoffstein, Wright, and Zamel, 1989, #9646- Krieger, Weitzenblum, Vandevenne, et al., 1985, #11908- Rauscher, Popp, and Zwick, 1990, and #12381- Shore and Millman, 1984) and are listed in Evidence Table 9. These studies included 595 patients total, of which 286 were diagnosed with SA. One patient with pure central apnea enrolled on study #12381- Shore and Millman, 1984, was excluded from all analyses. Their average age was 50.0 (k=4 studies reporting) and the percentage of patients who were male was 79 percent (k=4 studies reporting). The average BMI was 29.0 kg/m². PSG results were expressed as AI in 2 studies (diagnostic cutoffs, 5 and 10) and AHI in 1 study (diagnostic cutoff 10). One study did not state the PSG metric used for SA diagnosis. Sensitivity of FEF₅₀/FIF₅₀, a measure of extrathoracic airway obstruction, ranged from 12 percent to 67 percent, and the specificity ranged from 29 percent to 86 percent. In 2 studies FEF₅₀/FIF₅₀ was reported as a mean, and in 2 studies the percentage of patients with results exceeding 1.0 was reported. The presence of the sawtooth sign, indicative of pharyngeal fluttering during respirations, had a sensitivity ranging from 29 percent to 61 percent and a specificity ranging from 54 percent to 85 percent. Using both measures combined, the sensitivity ranged from 7 percent to 86 percent, and specificity from 13 percent to 89 percent. The average AI of the SA patients and of the non-SA patients was comparable, despite differing diagnostic cutoffs for SA. The meta-analysis of sensitivity and specificity yielded pooled estimates and ROC curve, as shown in Figure 6. The sensitivity of FEF₅₀/FIF₅₀ was 19.6 percent (± 9.6) and the specificity was 79.2 percent (± 9.7). Thus, flow volume loops provided the least accurate sensitivity and specificity.

Table

Evidence Table 9: Flow volume loop studies.

Figure

Figure 6: Flow volume FEF₅₀/FIF₅₀: Summary ROC curve and 95% confidence interval. FEF₅₀/FIF₅₀ =ratio of forced expiratory flow at 50% of vital capacity

For the sawtooth sign, the sensitivity was 61.9 percent (± 10.7) and the specificity 62.7 percent (± 7.2). When both measures were analyzed together, the sensitivity was 39.1 percent (± 25.3) and specificity 60.5 percent (± 23.7). The evidence scores of these studies ranged from 17 to 20 (average = 18.3).

Global impressions. There were 4 studies reporting the global impression of clinicians: 3 studies in clinic settings (#10986- Viner, Szalai, and Hoffstein, 1991, #9372- Hoffstein and Szalai, 1993, and #9255- Gyulay, Olson, Hensley, et al., 1993), and 1 study of sleeping patients in a sleep laboratory (#9276- Haponik, Smith, Meyers, et al., 1984). These studies are listed in Evidence Table 10. Together these studies included 1,139 patients total, 539 of whom were diagnosed with SA. AHI was the PSG metric used in 3 studies, with diagnostic cutoffs of 10 and 15. In the study of sleeping patients, the PSG metric was AI, and the cutoff for SA diagnosis was 5. Their average age was 47.7 (k=3 studies reporting), and the percentage of patients who were male was 67 percent (k=4 studies reporting). The average BMI was 29.4 kg/m². Sensitivity of global impressions of SA relative to PSG diagnosis of SA ranged from 52 percent to 79 percent, with a pooled estimate of 58.9 percent (±4.2); specificity ranged from 50 percent to 100 percent, the latter result from the observation of sleeping patients. The pooled estimate of specificity was 65.6 percent (±4.8), thus providing moderate sensitivity and specificity. The evidence scores of these studies ranged from 19 to 28 and averaged 23.3. The summary ROC curve for these 4 studies is in Figure 7. These show that while sensitivity was relatively constant across studies, specificity varied a great deal. For individual study results see Evidence Table 10.

Table

Evidence Table 10: Global impression studies.

Figure

Figure 7: Global impression: Summary ROC curve and 95% confidence interval.

Other clinical. Nine studies reporting sensitivity and specificity were identified for several clinical measures, but there were too few in each category to permit meta-analysis: neck circumference, airway dimensions via acoustic reflection (#11585-Katz, Stradling, Slutsky et al., 1990), nasopharyngeal airway resistance (#12075- Suratt, McTier, and Wilhoit, 1985), pulmonary function tests without flow volume curves (#11812- Onal, Leech, and Lopata, 1985), laryngoscopy (#12212- Geibel, Schonhofer, Rolzhauser, et al., 1997), snoring sound analysis (#11404- Fiz, Abad, Jane, et al., 1996), pupillary light reflex (#11892- Pressman and Fry, 1989), heart rate variability by ECG monitoring (#9567- Keyl, Lemberger, Pfeifer, et al., 1997), and body mass index alone (#11689- Lowe, Fleetham, Adachci, et al., 1995 and #10951- Vaidya, Petruzzelli, Walker, et al., 1996). No conclusions regarding the usefulness of any of these clinical measures as aids in the screening or diagnosis of SA can be made on the basis of so few studies.

Chemical. Similarly, there was only 1 study of a chemical test (urinary uric acid and creatinine) as a screen for SA: (#8610- Braghiroli, Sacco, Erbetta, et al., 1993). There were 88 patients enrolled, 49 with SA. Patients who desaturated at night differed from those who did not, but no correlation can be made on the basis of a single study.

Prediction Equations

Eight studies (#9255- Gyulay, Olson, Hensley, et al., 1993, #12219- Kushida, Efron, and Guilleminault, 1997, #11897-Quera-Salva, Guilleminault, Partinen, et al., 1988, #10951- Vaidya, Petruzzelli, Walker, et al., 1996, #10986- Viner, Szalai, and Hoffstein, 1991, #9372- Hoffstein and Szalai, 1993, #11888- Pracharktam, Nelson, Hans, et al., 1996, #10565- Schafer, Ewig, Hasper, et al., 1997) reported the sensitivity, specificity, or correlations of multivariate models relative to PSG results. MetaWorks investigators did not capture the predictive accuracy of each separate component of each model, although it was reported in some studies. Only the predictive features of the model as a composite result were captured. These studies included 1,908 patients, 254 of whom were known at entry to have SA; an additional 841 were diagnosed during the study. Their average age was 49.4 (k=8 studies reporting) and the percentage of patients who were male was 77 percent (k=8 studies reporting). The average BMI was 31.4 kg/m². Each model included at least 3 of the following variables: gender, age, obesity, hypertension, neck circumference, overjet, BMI, cephalometry measurements, arterial blood gases, home oximetry, pulmonary function tests, apnea spells, snoring, falling asleep while driving, percentage of time spent in Stage 1 sleep. Sensitivity of the models for the PSG result ranged from 28 percent to 97.6 percent, and specificities ranged from 21.4 percent to 100 percent. The pooled estimate for sensitivity was 66.5 percent (±14.0) and specificity was 88.7 percent (±4.9), achieving high sensitivity and specificity. The evidence scores of these studies ranged from 17 to 30, and averaged 21.5. All results are displayed in Evidence Table 11, and the summary ROC curve is shown in Figure 8. In this figure, both sensitivity and specificity were high for most studies.

Table

Evidence Table 11: Prediction equation studies.

Figure

Figure 8: Prediction equations: Summary ROC curve and 95% confidence interval.

Radiologic

One study correlated MRI results with PSG. This study included 40 patients whose results (#11936- Rodenstein, Dooms, Thomas, et al., 1990). One study correlated CT scans with PSG and included 37 patients (#12021- Shinohara, Kihara, Yamashita, et al., 1997). The latter also reported cephalometry, in relation to PSG results. There were additional cephalometry studies, which reported a multitude of different measurements of patients in different positions. Most of these studies did not, however, report correlations to PSG results, and none reported sensitivity or specificity in relation to PSG results. Of the 5 radiologic studies which did report correlations to PSG results, 1 study (#11888- Pracharktam, Nelson, Hans, et al., 1996) combined cephalometric results with morphometric results in a statistical model, which is discussed in the Prediction Equations section above. Among the remaining 4 studies (#8832- Davies and Stradling, 1990, #11689- Lowe, Fleetham, Adachi et al., 1995, #11081- Will, Ester, Rameriz, et al., 1995, #12021- Shinohara, Kihara, Yamashita, et al., 1997) with 256 patients in total, there is too little overlap of measurements to pool data, or even to synthesize data in a strictly qualitative way. The average evidence score of these studies was 18.5 (range, 17 to 20). These studies are displayed in Evidence Table 12.

Table

Evidence Table 12: Radiologic studies.

Questionnaires

Three studies reported sensitivity, specificity, or correlations of focused questionnaires to PSG results. Only 1 study (#10334- Pouliot, Peters, Neufeld, et al., 1997) out of 2 studies which used the Epworth Sleepiness Scale (ESS), reported sensitivity (42 percent) and specificity (68 percent) in 354 suspected SA patients, using a PSG AI threshold of 20. The other ESS paper (#9479- Johns, 1991) reported a correlation (r=0.55) to PSG RDI. The third paper (#9279- Haraldsson, Carenfelt, Knutsson, et al., 1992) did not use standard questionnaires, but selected questions about observed apneas, falling asleep or daytime sleepiness, and snoring. One additional paper (#8557- Bliwise, Nekich, and Dement, 1991) should be noted in this category (but is not included in Evidence Table 13), since it studied the sensitivity and specificity of several questions in a large number of patients (n=1,409). However, it only reported these outcomes by patient subgroups stratified by gender; as such it was not considered analyzable with the other studies in this set. These 3 studies are presented in Evidence Table 13.

Table

Evidence Table 13: Questionnaire studies.