Prediction of sleep-disordered breathing by unattended overnight oximetry

J. Sleep Res. (1999) 8, 51 55 Prediction of sleep-disordered breathing by unattended overnight oximetry L. G. OLSON, A. AMBROGETTI ands. G. GYULAY Discipline of Medicine, University of Newcastle and Sleep Disorders Centre, Royal Newcastle Hospital, Newcastle, NSW, Australia Accepted in revised form 1 September 1998; received 7 January 1998 SUMMARY Between January 1994 and July 1997, 793 patients suspected of having sleep-disordered breathing had unattended overnight oximetry in their homes followed by laboratory polysomnography. From the oximetry data we extracted cumulative percentage time at S a O 2 < 90% (CT 90 ) and a saturation variability index (ΔIndex, the sum of the differences between successive readings divided by the number of readings 1). CT 90 was weakly correlated with polysomnographic apnea/hypopnea index (AHI), (Spearman ρ=0.36, P< 0.0001) and with ΔIndex (ρ=0.71, P< 0.0001). ΔIndex was more closely correlated with AHI (ρ=0.59, P< 0.0001). In a multivariate model, only ΔIndex was significantly related to AHI, the relationship being AHI=18.8 ΔIndex +7.7. The 95% CI for the coefficient were 16.2, 21.4, and for the constant were 5.8, 9.7. The sensitivity of a ΔIndex cut-off of 0.4 for the detection of AHI[15 was 88%, for detection of AHI[20 was 90% and for the detection of AHI[25 was 91%. The specificity of ΔIndex [0.4 for AHI[15 was 40%. In 113 further patients, oximetry was performed simultaneously with laboratory polysomnography. Under these circumstances ΔIndex was more closely correlated with AHI (ρ=0.74, P< 0.0001), as was CT 90 (ρ=0.58, P< 0.0001). Sensitivity of ΔIndex [0.4 for detection of AHI[15 was not improved at 88%, but specificity was better at 70%. We concluded that oximetry using a saturation variability index is sensitive but nonspecific for the detection of obstructive sleep apnea, and that few false negative but a significant proportion of false positive results arise from night-to-night variability. KEYWORDS oximetry, apnea/hypopnea index, sleep disorder breathing INTRODUCTION Clinical diagnosis of obstructive sleep apnea (OSA) is inaccurate (Crocker et al. 1990; Gyulay et al. 1993) and therefore some form of monitoring procedure is generally required to make or exclude the diagnosis. Polysomnography is the gold-standard for diagnosis of OSA, but because of its cost cheaper alternatives are desirable. We have previously reported (Gyulay et al. 1993) that unattended pulse oximetry had sensitivity greater than 90% for the diagnosis of OSA. This result was achieved with a criterion of a cumulated time with arterial oxygen saturation less than 90% (CT 90 ) greater than 1% of recording time. Subsequently, Sériès et al. (1993) Correspondence: Dr LG Olson, Respiratory Medicine Unit, John Hunter Hospital, Locked Bag 1, Newcastle Mail Centre, NSW 2310, Australia. Fax:+61 2 49213998; e-mail: lolson@mail.newcastle.edu.au 1999 European Sleep Research Society reported superior results by inspecting the oximetry tracing without predetermined criteria for abnormality. Other workers have used an index of saturation variability (Lévy et al. 1996), but its evaluation has not been repeated. In order to confirm the results of earlier studies with a larger sample and to compare the CT 90 and saturation variability (ΔIndex) criteria, we have reviewed the results of 793 patients assessed both with home oximetry and laboratory polysomnography. In order to determine whether night-to-night variation in the severity of OSA accounted for any of the diagnostic inaccuracy of overnight oximetry, we studied an additional 113 patients with simultaneous oximetry and polysomnography. METHODS Newcastle is an industrial city located about 160 km north of Sydney. The population is predominantly urban and of Anglo- Celtic origin. The Newcastle Sleep Disorders Centre is the only 51

52 L.G. Olsen et al. referral centre for sleep disorders for a population of about 500 000. Between 1991 and the end of 1993, we performed home oximetry as a screening test for patients suspected of OSA using the CT 90 criterion we had evaluated (Gyulay et al. 1993). During this period, recordings in which CT 90 was < 1% were considered to make OSA unlikely. The results of Sériès et al. (1993) suggested that this method missed some cases in which apneas caused frequent small desaturations. We could not reproduce the results obtained by Sériès and co-workers with direct inspection of the oximetry tracing, but thereafter we determined not only CT 90 but also a saturation variability index (ΔIndex, the sum of differences between successive readings divided by the number of readings 1) (Lévy et al. 1996). Figure 1. The relationship of apnoea/hypopnoea index (AHI) to From December 1993, overnight oximetry was considered cumulative time spent at S a O 2 below 90% (CT 90) in patients having unattended overnight oximetry. The vertical line is at AHI=15 and abnormal if CT 90 was 1% or if ΔIndex was 0.4, and such the horizontal line at CT 90 =1%. See text for statistical significance. recordings were followed by laboratory polysomnography. Laboratory polysomnography was also performed on all patients considered to have significant daytime sleepiness (in these cases multiple sleep latency testing was also performed) or to have a high likelihood of OSA despite normal overnight oximetry. Between December 1993 and July 1997, 1020 overnight oximetries were performed and of these 793 were followed by laboratory polysomnography. Of the remaining 227 oximetries 68 were grossly abnormal, and in the presence of a typical clinical picture, nasal CPAP treatment of OSA was initiated without further investigation, 116 were normal and in the absence of daytime sleepiness or a clinical picture strongly suggesting OSA, no further investigation was undertaken, and 43 patients were lost to follow-up without undergoing polysomnography. Of the 793 patients who had polysomnography, 348 (44%) had an apnea/hypopnea index (AHI) equal to or greater than 15/h. The mean AHI in those with Figure 2. The relationship of apnoea/hypopnoea index (AHI) to a AHI 15 was 40.0 (range 15 118; 95% CI 38.4, 43.0). saturation variability index (DIndex) in patients having unattended overnight oximetry. The vertical line is at AHI=15 and the Oximetry was performed with pulse oximeters using finger horizontal line at DIndex=0.4. See text for statistical significance. probes (Biox 4700, Ohmeda Corporation, Boulder, CO, USA). These machines record in an internal memory the lowest saturation observed in a 12-s period. These stored values were used for calculation of CT 90 and ΔIndex. RESULTS In order to identify the role of night-to-night variability in The relationship between CT 90 and AHI is shown in Fig. 1. the diagnostic inaccuracy of unattended oximetry, we studied The vertical line indicates an AHI of 15 and the horizontal a separate group of patients who did not have unattended line a CT 90 of 1%. The Spearman correlation coefficient ρ= overnight oximetry before overnight polysomnography. 0.36 (P< 0.0001). The relationship between ΔIndex and AHI Patients were eligible for this element of the study if they were is shown in Fig. 2. The vertical line again indicates an AHI of referred because of snoring or somnolence (most ineligible 15 and the horizontal line a ΔIndex of 0.4. The Spearman patients were referred because of insomnia and the rest for correlation coefficient ρ=0.59 (P< 0.0001). Both oximetry parasomnias). In most cases, unattended oximetry was not indices were related to AHI, although the relationship of carried out because the patients were referred from ΔIndex to AHI was clearly closer. CT 90 and ΔIndex were geographically remote areas. Laboratory oximetry was carried correlated with one another (Spearman ρ=0.71; P< 0.0001). out with the same instruments as unattended oximetry and the Cases not detected by both indices were relatively common, data were processed in an identical fashion. In this group, 25 and although the great majority of false negative tests patients (22%) had AHI 15, and the mean AHI in those with represented mild OSA, some cases with high AHI were missed AHI 15 was 38.0 (range 15 110, 95% CI 28.2, 49.4). by both indices. Overall, the sensitivity of CT 90 equal to or Statistical analysis was performed with the Stata package greater than 1% for the detection of AHI more than 15 was (Release 5.0, Stata Corporation, College Station, TX, USA). 75% and the specificity was 46%. The sensitivity of ΔIndex

Oximetry and SDB 53 Table 1 Sensitivity and specificity for a range of apnea/hypopnea indices (AHI) for a range of cut-offs of saturation variability index (ΔIndex) and cumulative time spent at S a O 2 below 90% (CT 90 ) AHI 5 15 30 Specificity Specificity Specificity Sensitivity Sensitivity Sensitivity ΔINDEX 94.5 27.0 96.8 17.8 97.5 14.4 0.3 ΔINDEX 89.3 31.3 94 31.2 97.0 26.1 0.35 ΔINDEX 82.7 54.2 88.5 39.6 92.6 34.1 0.4 CT 90 0.75 73.6 49.1 78.2 40.9 85.8 38.9 CT 90 1.0 69.9 54.7 75.3 46.1 84.3 44.0 CT 90 1.25 66.3 58.4 72.1 50.1 81.9 48.0 Table 2 Likelihood ratios for the presence of a range of apnea/hypopnea indices (AHI) associated with positive (+ve LR) and negative ( ve LR) results for a range of cut-offs of saturation variability index (DIndex) and cumulative time spent at S a O 2 below 90% (CT 90 ) AHI 5 15 30 +ve LR ve LR +ve LR ve LR +ve LR ve LR ΔIndex 1.29 0.20 1.18 0.18 1.13 0.17 0.3 ΔIndex 1.30 0.34 1.36 0.19 1.31 0.11 0.35 ΔIndex 1.80 0.32 1.48 0.29 1.40 0.22 0.4 CT 90 0.75 1.44 0.54 1.32 0.44 1.41 0.36 CT 90 1.0 1.55 0.52 1.39 0.54 1.50 0.36 CT 90 1.25 1.58 0.58 1.44 0.56 1.58 0.39 equal to or greater than 0.4 for the detection of AHI more Fig. 3. Diagnostic accuracy was also improved, especially as than 15 was 88% and the specificity 40%. Sensitivity was regards false positive results. Sensitivity for CT 90 [1% for relatively unaffected by changing the AHI cut-off, although detection of AHI[ 15 was 92% and specificity was 73%. For specificity fell as the cut-off was moved to lower AHI. ΔIndex [0.4 sensitivity for detection of AHI[ 15 was 88% Sensitivity, specificity and likelihood ratios for a range of CT 90, and specificity was 70%. ΔIndex and AHI cut-offs are shown in Tables 1 and 2. The regression equation relating AHI, ΔIndex and CT 90 was Stepwise linear regression was used to obtain an equation in this case AHI=26.9 ΔIndex 0.24 CT 90 1.08. The 95% CI relating AHI to CT 90 and ΔIndex, which was AHI=17.4 for the coefficient of ΔIndex were 17.9, 35.9, for the coefficient ΔIndex 0.03 CT 90 +8.2. The 95% CI for the coefficient of of CT 90 0.86, 0.38 and for the constant term 4.68, 2.51. ΔIndex were 14.5, 20.2, for the coefficient of CT 90 0.13, 0.08 The frequency of oxygen desaturation was also calculated for and for the constant term 6.0, 10.3. The proportion of the patients having simultaneous oximetry and polysomnography. variance of AHI accounted for by this model was 30.3%. ΔIndex Oxygen desaturation index (ODI) was calculated for significantly contributed to the model (t=12.0, P< 0.0001) but desaturations of 1%, 2% and 3% (ODI1, ODI2 and ODI3, CT 90 did not (t= 0.46, P=0.647). respectively). The Spearman correlation coefficients for AHI In the 113 patients who had oximetry simultaneously with and ODI1, ODI2 and ODI3 were 0.55, 0.54 and 0.57, laboratory polysomnography, the correlations between both respectively (P< 0.0001 in all cases). The sensitivities of CT 90 and ΔIndex and AHI were closer. For ΔIndex Spearman ODI[15 for AHI[15 for ODI1, ODI2 and ODI3 were 72%, ρ=0.74 (P< 0.0001) and for CT 90 ρ=0.58 (P< 0.0001). The 36% and 12%, respectively, and the specificities 74%, 90% and relationship of ΔIndex to AHI for these subjects is shown in 98%, respectively.

54 L.G. Olsen et al. did not receive working diagnoses of OSA or snoring, we cannot distinguish retrospectively those unequivocally irrelevant to our purpose from those in whom OSA was in fact an important issue. The division of subjects into groups having unattended oximetry or laboratory oximetry was also not in any way random, although no bias existed after referral. The group who had laboratory oximetry contained fewer patients with OSA, although the distribution of severity was similar. The lower proportion of patients with OSA would work against finding higher specificity, however, which was the principal difference between unattended and laboratory oximetry. Not all subjects who had oximetry had polysomnography, and this is likely to have affected the results. We have no clinical information concerning the 43 patients lost to follow- up, and therefore cannot estimate the impact their results might have had on our estimates of the sensitivity and specificity of oximetry. A larger group (116) did not proceed to polysomnography because the chance of OSA was judged too low to justify further investigation. The cut-off used to identify these patients had, in the present study, a sensitivity of 88% for AHI 15, therefore the prevalence of OSA in the group not subjected to further study is not likely to have been higher than 12% and was probably less. If it was 12%, inclusion of these patients would have reduced sensitivity for the ΔIndex=0.4 cut-off to 84.5% and increased specificity to 51%. These effects are modest, but the pretest probability of OSA in these patients was very low and in many settings they would not have been referred for evaluation at all. For this reason we do not believe that including them gives an accurate reflection of the likely performance of overnight oximetry in most referral clinics. The 68 patients not studied because they had obvious OSA, had the presence of obstructive apnea confirmed during pressure setting for nasal CPAP and responded symptomatically to nasal CPAP, therefore we consider the diagnosis of OSA certain. Addition of this group would have increased sensitivity for the ΔIndex=0.4 cut-off from 88% to 90% (specificity would be unaffected). In this case we believe that excluding these patients provides a more accurate reflection of the performance of overnight oximetry in the patient group in which a screening test is likely to be used. The utility of unattended oximetry depends on the level of sleep-disordered breathing which it is desired to detect. The conventional cut-off for the diagnosis of OSA is AHI greater than 15 events per hour, but the empirical basis for this is not strong. Two general criteria for the cut-off between normal and pathological may be used: the level at which there are no adverse effects or the level at which treatment offers no net benefit. The relationship of AHI with adverse outcomes is not well-defined, but there is evidence that AHI as low as 5 is associated with higher systemic blood pressure (Young et al. 1997). If it is desired to screen for AHI in this range, unattended oximetry will be of no use, because in order to obtain useful sensitivity the oximetry cut-off has to be set so low that specificity is inadequate. There is, however, no evidence that Figure 3. The relationship of apnoea/hypopnoea index (AHI) to a saturation variability index (DIndex) in patients having simultaneous laboratory oximetry and polysomnography. The vertical line is at AHI=15 and the horizontal line at DIndex=0.4. See text for statistical significance. DISCUSSION The results of this study suggest that unattended oximetry is reasonably sensitive but nonspecific for the detection of OSA. Relatively few of the false negative results but a substantial proportion of the false positive results, obtained with unattended oximetry, are related to night-to-night variability of sleep-disordered breathing. The cohort of patients studied was subject to referral bias, whose magnitude is unknown. However, our epidemiological studies have shown that in the general community most individuals with OSA have mild or borderline disease (Olson et al. 1995). Any diagnostic test will perform less well on patients with borderline disease, and, conversely, the higher the proportion of patients with severe or classical disease the better the test will appear to perform. The mean AHI of the subjects of this study whose AHI was greater than 15 was 40 (see Results, above) compared to 32 in the community (Olson et al. 1995). There appears therefore to have been bias towards referring more severely affected patients, but overall the distribution of severity in the present sample was probably closer to that of the general population than has been the case in samples in which oximetry has performed better (Sériès et al. 1993; Lévy et al. 1996). Nevertheless, unattended oximetry would probably not perform as well in a sample of uniformly low-risk subjects or in a true general community sample as in the present sample. An element of bias, impossible to quantify retrospectively, was also introduced because not all patients had oximetry. Patients with insomnia and many of those with chronic fatigue were considered to have a likelihood of OSA too low to justify oximetry as a screening test, but some would nevertheless have had OSA (Jacobs et al. 1988). In other patients in whom the clinical picture was complex or obscure, the information obtained from unattended oximetry would have been judged to add little to formal polysomnography and these patients were not included in this study. Because many of these patients

Oximetry and SDB 55 treating mild OSA will reduce blood pressure when it is elevated We have no clear explanation for the improved specificity (Rolfe et al. 1991), much less that this will lead to improved associated with laboratory oximetry compared to home outcomes. If screening for this level of sleep-disordered oximetry. This pattern suggests that patients had, overall, more breathing comes in the future to be considered worthwhile, sleep-disordered breathing at home than in the laboratory. neither unattended oximetry nor any other screening method Improved sleep quality with more REM sleep could have this presently available would have adequate performance. On the effect. The consumption of alcohol at home before unattended other hand, clear benefits from treatment have been shown oximetry might also be responsible. Patients doing this might only for AHI over 30 (Wright et al. 1997), and for detection have positive results on unattended oximetry but negative of this level of AHI unattended oximetry is sensitive, and laboratory polysomnography because on those nights they are although relatively nonspecific could significantly reduce the free of alcohol. demand for polysomnography where this is a scarce resource. The present data suggest that the utility of unattended It is possible that the addition of a measure of daytime oximetry is limited. The negative likelihood ratios in Table 2 symptoms might improve the diagnostic utility of screening. show that in patients with a relatively low pretest probability We did not systematically collect Epworth Sleepiness Scale of OSA, unattended oximetry negative by either criterion could scores during the whole period of this study. However, we have reduce the post-test probability to a level that would not observed in a different cohort of patients that no measure of warrant further investigation. However, the low specificities oxygenation during polysomnography was related to Epworth associated with these results in Table 1 show that the number Sleepiness Scale scores (Olson et al. 1998). of cases with this outcome will be relatively low. The uniformly low positive likelihood ratios show that no common positive The cases of OSA missed by unattended oximetry may result makes a useful contribution to diagnosis. represent true false negative results. Patients may have repeated In order to make unattended oximetry useful, its specificity obstructive events with no oxygen desaturation and these cases must be increased. The sensitivity of unattended oximetry, even would be missed by oximetry in any setting and however, the for the mildest grades of OSA, is adequate but poor specificity results were analysed. The failure of sensitivity to improve results in likelihood ratios for positive results that are barely when oximetry was carried out in the laboratory supports this above unity. Other forms of unattended monitoring device are explanation. not likely to improve on the sensitivity of unattended oximetry, The method of analysis may also produce false negative and it is in greater specificity that their advantages must lie. results by obscuring the presence of significant events. The CT 90 criterion is particularly likely to do this because it takes no account of desaturations that do not extend to below REFERENCES 90%. We and others have found that indices of desaturation Crocker, B. D., Olson, L. G., Saunders, N. A. et al. Estimation of the frequency using threshold criteria (usually 3% or 4% falls) probability of disturbed breathing during sleep before a sleep study. Am. Rev. Respir. Dis., 1990, 142: 14 18. perform poorly because many significant events are not Gyulay, S. G., Olson, L. G., Hensley, M. J., King, M. T., Murreeassociated with desaturation, so that the desaturation index is Allen, K. and Saunders, N. A. Comparison of clinical assessment less than the AHI (Gyulay et al. 1993; Sériès et al. 1993; Lévy and home oximetry in the diagnosis of obstructive sleep apnoea. et al. 1996). These results were confirmed in the present study, Am. Rev. Respir. Dis., 1993, 147: 50 53. Jacobs, E. A., Reynolds, C. F., Kupfer, D. J., Lovin, P. A. and ODI3 15 having sensitivity of 12% for AHI 15. The high Ehrenpreis, A. B. The role of polysomnography in the differential specificity of ODI is not a compensation for its low sensitivity diagnosis of chronic insomnia. Am. J. Psychiatry, 1988, 145: 346 349. when oximetry is proposed as a screening test. Lévy, P., Pepin, J. L., Deschaux-Blanc, C., Paramelle, B. and Brambilla, Sériès and co-workers reported a sensitivity of 100% by C. Accuracy of oximetry for detection of respiratory disturbances in sleep apnea syndrome. Chest, 1996, 109: 395 399. direct inspection of the oximetry tracing with no predetermined Olson, L. G., Cole, M. F. and Ambrogetti, A. Correlations among criteria for abnormality (Sériès et al. 1993). We have been Epworth Sleepiness Scale scores, Multiple Sleep Latency Tests and unable to reproduce these results, and since one reason for psychological symptoms. J. Sleep Res., 1998, 7: 248 253. screening tests is to allow clinicians inexpert in sleep disorders Olson, L. G., King, M. T., Hensley, M. J. and Saunders, N. A. A to avoid referring every patient in whom OSA is a possibility, community study of sleep-disordered breathing: prevalence. Am. J. Respir. Crit. Care Med., 1995, 152: 711 718. we do not think that an approach that avoids objective criteria Rolfe, I. E., Olson, L. G. and Saunders, N. A. Prolonged nasal CPAP is helpful. therapy does not ameliorate obstructive sleep apnoea. Aust. N. Z. Lévy et al. (1996) reported that ΔIndex gave a sensitivity of J. Med., 1991, 21: 235 238. 98% and a specificity of 46% with a cut-off of 0.6. In the Sériès, F., Marc, I., Cormier, Y. and La Forge, J. Utility of nocturnal home oximetry for case finding in patients with suspected Sleep present sample a cut-off of 0.6 did not give adequate sensitivity Apnea Hypopnea Syndrome. Ann. Intern. Med., 1993, 119: 449 453. (67%), although specificity was higher than this cut-off given Wright, J., Johns, R., Watt, I., Melville, A. and Sheldon, T. Health in Lévy s sample (77%). This result might arise if the present effects of obstructive sleep apnoea and the effectiveness of continuous sample had more patients with borderline or very mild OSA. positive airways pressure: a systematic review of the research evidence. B.M.J., 1997, 314: 851 860. This situation is probably more representative of the Young, T., Peppard, P., Palta, M. et al. Population-based study of circumstances in which a screening test is most likely to be sleep-disordered breathing as a risk factor for hypertension. Arch. used. Intern. Med., 1997, 157: 1746 1752.