Objective. To determine the utility of pulse oximetry for diagnosis of obstructive sleep apnea (OSA) in children.
Methods. We performed a cross-sectional study of 349 patients referred to a pediatric sleep laboratory for possible OSA. A mixed/obstructive apnea/hypopnea index (MOAHI) greater than or equal to 1 on nocturnal polysomnography (PSG) defined OSA. A sleep laboratory physician read nocturnal oximetry trend and event graphs, blinded to clinical and polysomnographic results. Likelihood ratios were used to determine the change in probability of having OSA before and after oximetry results were known.
Results. Of 349 patients, 210 (60%) had OSA as defined polysomnographically. Oximetry trend graphs were classified as positive for OSA in 93 and negative or inconclusive in 256 patients. Of the 93 oximetry results read as positive, PSG confirmed OSA in 90 patients. A positive oximetry trend graph had a likelihood ratio of 19.4, increasing the probability of having OSA from 60% to 97%. The median MOAHI of children with a positive oximetry result was 16.4 (7.5, 30.2). The 3 false-positive oximetry results were all in the subgroup of 92 children who had diagnoses other than adenotonsillar hypertrophy that might have affected breathing during sleep. A negative or inconclusive oximetry result had a likelihood ratio of .58, decreasing the probability of having OSA from 60% to 47%. Interobserver reliability for oximetry readings was very good to excellent (κ = .80).
Conclusions. In the setting of a child suspected of having OSA, a positive nocturnal oximetry trend graph has at least a 97% positive predictive value. Oximetry could: 1) be the definitive diagnostic test for straightforward OSA attributable to adenotonsillar hypertrophy in children older than 12 months of age, or 2) quickly and inexpensively identify children with a history suggesting sleep-disordered breathing who would require PSG to elucidate the type and severity. A negative oximetry result cannot be used to rule out OSA.
Obstructive sleep apnea (OSA) in children is a disorder of breathing during sleep characterized by upper airway obstruction that disturbs sleep and disrupts normal respiratory gas exchange.1–3 Sequelae of OSA, including growth failure, cor pulmonale, and developmental delay, are thought to be attributable to sleep disturbance and repetitive hypoxemia.1,,4,2 Ali et al5 estimated the prevalence of OSA in 4- to 5-year-old children at .7%, whereas 7% to 9% of children have simple snoring, a milder form of sleep-related upper-airway obstruction.5–7Usually, pediatric OSA is attributable to adenotonsillar hypertrophy.
The most accurate and comprehensive method of diagnosing OSA is nocturnal polysomnography (PSG).3 Unfortunately, such testing is expensive and time-consuming. Thus, an abbreviated testing modality would be very helpful. We originally proposed that parental responses to questions about snoring, difficulty breathing during sleep, and observed obstructive apnea might distinguish normal children from those with OSA.8 However, several subsequent studies indicated that questionnaire responses could not reliably distinguish children with OSA from those with simple snoring.9–12
Pediatric OSA sometimes, but not always, results in dips in hemoglobin saturation (Sao2). In adults being evaluated for OSA, apparently conflicting results on the usefulness of pulse oximetry have been reported.13,,14 For adults with a low pretest probability of OSA, a negative oximetry result may be sufficient to exclude the diagnosis; for adults with a high pretest probability of OSA, a positive oximetry result may be sufficient to confirm the diagnosis of OSA.15 The accuracy of nocturnal pulse oximetry compared with PSG for diagnosing pediatric OSA has not been evaluated.
Since 1993, when patients were referred to our pediatric sleep laboratory, we have: 1) had the parents complete a questionnaire, 2) performed nocturnal PSG, and 3) printed the pulse oximetry trend and event graphics that summarize nocturnal oxygenation status.8,,16,17 The purpose of the current report is to assess, from this large series of referrals to our pediatric sleep laboratory, the potential utility of nocturnal pulse oximetry for diagnosing pediatric OSA. A secondary aim was to determine the accuracy of questionnaire data for diagnosing OSA.
We conducted a cross-sectional study of all children referred to the sleep laboratory of the Montreal Children's Hospital for evaluation of possible (OSA). Between January, 1993 and April 1998, there were a total of 552 patients evaluated. Of these, 349 were referred for nocturnal PSG to evaluate OSA, were between 6 months and 18 years of age, and had complete data available. Subjects who had lung disease, neuromuscular disorders, diagnoses associated with central apnea or central hypoventilation, and those who were not specifically referred for possible OSA were excluded (Table 1).
For each patient, data included a parental questionnaire, polysomnographic results, and simultaneously obtained pulse oximetry trend and event graphics. To define OSA for this study, we chose an objective measure, the polysomnographically determined mixed/obstructive apnea/hypopnea index (MOAHI) greater than or equal to 1 event per hour.3,,18,19 Children with MOAHI values <1 were considered to be normal or to have simple snoring.9 A sleep laboratory physician (R.T.B. or K.W.), blinded to both clinical information and polysomnographic results, read the oximetry graphs.16 Each oximetry was classified as negative, inconclusive, or positive and was compared with the polysomnographic diagnosis.
Questionnaire and OSA Score
Before polysomnographic testing, the parents of each subject completed a questionnaire.8 Questions included those about sleep, breathing during sleep, medical diagnoses, and previous surgery, including adenotonsillectomy. An additional question was added regarding degree of parental concern about breathing during sleep. From the questionnaire responses, our previously described OSA score was calculated using responses to 3 questions: 1) difficulty breathing during sleep, 2) obstructive apnea observed by parents, and 3) frequency of snoring during sleep.8
Overnight PSG was performed at home (7 channels) or in the sleep laboratory (14–21 channels). These home and laboratory recordings provide equivalent assessments of sleep/wakefulness, apnea/hypopnea indexes, and movement/arousals.17,,20,21 At home, the cardiorespiratory recording consisted of electrocardiogram, pulse rate, Sao2, pulse wave form, and calibrated respiratory inductive plethysmography (thoracic, abdominal, and sum channels).17 In the laboratory, nocturnal 14- to 21-channel polysomnographic recordings included the home study channels plus 5 channels for sleep staging: central and occipital electroencephalograms, right and left electrooculogram, and submental electromyogram. Oral/nasal airflow, transcutaneous oxygen, and end-tidal carbon dioxide measurement were always included; intercostal, abdominal and arm electromyograms were sometimes included. Signals were acquired directly onto a computerized polysomnograph (Ultrasom, Nicolet, Madison, WI) in the sleep laboratory or transferred from a portable computer after the home studies. Computerized audiovisual recordings and analysis for sleep/wake and movement/arousals were used in both settings.17,,20,21
Polysomnographic outcomes included apnea, hypopnea, and desaturation indexes, total sleep time, and sleep efficiency—the percentage of time spent asleep from sleep onset until morning wakening.3Sleep versus wakefulness was scored as previously described.20 Hemoglobin desaturation events were defined as a 4% or greater decrease in Sao2. Apnea was defined as an 80% or greater decrease in amplitude, for 1 or more breaths, on the respiratory inductive plethysmographic sum channel. Mixed and obstructive apneas were considered significant regardless of duration. Hypopnea was defined as a 50% to 80% decrease in the amplitude of the plethysmographic sum associated with a decrease in Sao2 of 4% or greater. The MOAHI and the hemoglobin desaturation index were calculated as the number of events per hour of sleep.
Pulse Oximetry: Methods and Classification
Sao2 was recorded during PSG using a Nellcor, N-200 pulse oximeter set to use 2- to 3-second averaging (mode 2). In the morning, trend and event graphics were printed (Fig 1).16 Trend graphs present 12-hour summaries of Sao2 and pulse rate. Event graphs present more detailed information on Sao2, pulse rate, and pulse amplitude. We have previously demonstrated the usefulness of event graphs to discriminate true hemoglobin desaturation from artifacts secondary to movement or a low pulse signal.16 In pediatric OSA, obstructive apneas and hypopneas occur more in rapid eye movement (REM) than non-rapid eye movement sleep20,,22; desaturation events, therefore, cluster during REM sleep periods. For purposes of this study, a sleep laboratory physician (K.A.W. or R.T.B.) classified each oximetry as positive, negative, or inconclusive using the following definitions and criteria: 1) A desaturation was defined as a decrease in Sao2 of 4% or more; 2) A cluster of desaturations was defined as 5 or more desaturations occurring in a 10- to 30-minute period; 3) On the oximetry trend graphs, periods of relative tachycardia, usually 10 to 25 beats per minute, at the beginning and end of nocturnal pulse oximetry, and periods of relative tachycardia and increased heart rate variability exceeding 30 minutes were regarded as wakeful time and not considered; 4) Event graphs for Sao2 were used to distinguish true desaturations from movement artifacts and low signal amplitude artifacts using a method we have termed pulse amplitude modulation range16; 5) A positive oximetry trend graph had 3 or more desaturation clusters and at least 3 desaturations to <90%19; 6) A negative oximetry trend graph had no desaturation clusters and no desaturations to <90%; and 7) An inconclusive oximetry trend graph was 1 that did not meet the criteria for positive or negative.
Our objective was to determine the accuracy with which pulse oximetry graphs and the OSA score identify OSA. We used the likelihood ratio approach that indicates by how much a diagnostic test result will raise or lower the pretest probability of a target disorder.23,,24 Pretest probability of having OSA was taken as the percentage of patients referred for OSA who were polysomnographically proven to have OSA. Likelihood ratios were used to calculate posttest probabilities of having OSA based on the OSA score (positive, negative, or inconclusive) and on pulse oximetry (positive or negative/inconclusive). The data were first analyzed for all patients and then subgroup analyses were performed for patients with and without diagnoses other than adenotonsillar hypertrophy that might have affected breathing during sleep.
Statistical analyses were conducted using SPSS and SigmaStat programs (SPSS Inc, Chicago, IL). For normally distributed data, results were expressed as the mean ± standard deviation. For data that was not normally distributed, results were expressed as median (interquartile ranges).
Interobserver Reliability of Oximetry
One hundred oximetries were selected randomly and independently classified as negative, positive, or inconclusive by 2 pediatric sleep laboratory technicians (A.M. and R.L.) and a sleep laboratory physician (R.T.B.). Thus, there were 3 sets of 100 possible agreements. The κ statistic was used to estimate the level of agreement beyond that expected by chance.25
The study group of 349 subjects consisted of 218 (62%) boys and 131 girls. Median age was 4.5 (2.9, 7.1) years. There were 10 patients 6 to 11 months of age, 36 patients 12 to 23 months of age, 185 patients 2 to 5 years of age, 75 patients 6 to 9 years of age, and 43 patients 10 to 17 years of age. Ninety-two patients had medical diagnoses other than adenotonsillar hypertrophy that could have had an influence on breathing during sleep (Table 2).
Total sleep time averaged 8.1 ± 1.4 hours and sleep efficiency averaged 89 ± 10%. Of the studies, 260 were performed in the home and 89 were performed in the laboratory. The median MOAHI was 2.0 (.3, 9.6). Considering a MOAHI greater than or equal to 1 as abnormal, 210 (60.2%) children had OSA and 139 (39.8%) were considered to have simple snoring or normal breathing during sleep. The median desaturation index was 2.4 (.7, 7.5). The desaturation index was highly correlated with the MOAHI: MOAHI = .8 (desaturation index) + 1.3 (r2 = .78; P < .001; Fig 2).
Questionnaire as an Abbreviated Testing Modality
As reported previously, snoring, difficulty breathing, obstructive apneas seen by parents, and degree of parental concern were more common in patients found to have OSA than in those found to have simple snoring or normal breathing during sleep (Table 3). However, these individual symptoms and the OSA score, which integrates several of these individual symptoms, were only moderately helpful in changing the diagnostic probability of OSA.8–12,26,27 Likelihood ratios for OSA scores <−1 (predicted no OSA), −1 to 3.5 (inconclusive), and >3.5 (predicted to have OSA) were .617, .906, and 1.81, respectively (Table 3). Considering the pretest probability of having OSA as 60%, posttest probabilities were 44%, 58%, and 73%, respectively (Fig 3A).
Pulse Oximetry as an Abbreviated Testing Modality
Interobserver agreements among the 3 readers for pulse oximetry classification as positive versus inconclusive/negative were 91%, 90%, and 85%. κ analysis results showed that this level of agreement was excellent: .80 (95% confidence interval: .66-.88).25
The pulse oximetry trend and events graphs were classified as positive in 93, negative in 95, and inconclusive in 161. Of 93 oximetries read as positive, PSG confirmed OSA in 90 patients; therefore, the positive predictive value of oximetry was 97% (90/93). The median MOAHI in subjects with a positive oximetry classification was 16.4 (7.5, 30.2;Fig 4).
Test properties of pulse oximetry graphs for OSA are summarized inTable 4; inconclusive and negative pulse oximetry classifications were collapsed, as the results were similar. A positive pulse oximetry had a likelihood ratio of 19.4, which increased the probability of having OSA from 60% to 97% (Fig 3B). A negative or inconclusive pulse oximetry had a likelihood ratio of .58, which lowered the probability of having OSA from 60% to 47% (Fig 3B).
According to our study design, 3 of 93 pulse oximetry readings were considered false-positive, ie, correlated with a MOAHI <1 event per hour as documented by inlaboratory PSG. One 22-month-old child with asthma and adenotonsillar hypertrophy had a MOAHI of .6/hour. However, his PSG was read as compatible with mild OSA, based on paradoxical chest and abdominal movement present for 11% of total sleep time, snoring, and obstructive hypopneas that were not associated with a 4% desaturation. One 15-month-old boy with sickle cell anemia had a desaturation index of 6.6/hour. Desaturations occurred against a low baseline saturation of 93% to 94%, occurred during REM sleep, and were not related to apneas or hypopneas. Another 8½-year-old boy was suspected of having Prader-Willi syndrome, although cytogenetic testing was normal for the 15q11–13 deletion. Although his polysomnographic study revealed a high central apnea/hypopnea index (20/hour), he had loud breathing and used accessory muscles of respiration suggesting a degree of airway obstruction. There was marked clinical and polysomnographic improvement after an adenotonsillectomy.
One hundred sixty-five of the 257 patients with only adenotonsillar hypertrophy to explain their sleep-disordered breathing had OSA (pretest probability: 64%). In this group, oximetry had no false-positive results and correctly identified 71 of 165 patients with OSA. A positive oximetry result had a likelihood ratio of 43, giving a posttest probability of 99% of having OSA. A negative or inconclusive oximetry result had a likelihood ratio of .576, giving a posttest probability of 58% of having OSA.
Forty-five of the 92 patients with medical diagnoses other than adenotonsillar hypertrophy had OSA (pretest probability: 49%). In this group, oximetry had 3 false-positive results and correctly identified 19 of 45 patients with OSA. A positive oximetry result had a likelihood ratio of 6.60, giving a posttest probability of 87% of having OSA. A negative or inconclusive oximetry result had a likelihood ratio of .618, giving a posttest probability of 39% of having OSA.
Among children who were referred to a pediatric sleep laboratory for possible OSA, a positive pulse oximetry result increased the probability of a patient having OSA from 60% to 97% (Fig 3B). Furthermore, the median MOAHI of patients with a positive pulse oximetry result was 16 events per hour and 90% of these patients had a MOAHI greater than 2.5 events per hour. A negative or inconclusive oximetry trend graph did not rule out OSA; indeed, referred children with negative pulse oximetry results had a 47% probability of having OSA on PSG (Fig 3B). Clearly, some children can have obstructive apneas without accompanying desaturations. The close relation between MOAHI and desaturation index suggests that pediatric OSA with desaturation events can be considered a more severe disorder than OSA without desaturations.28
We report on the potential usefulness of pulse oximetry in a group of children at high risk for OSA, ie, those referred to a pediatric sleep laboratory for evaluation of this disorder. We took the pretest probability of OSA as 60%, the incidence of OSA in this series. Readers having children with a different pretest probability of having OSA could use Fig 3B to calculate posttest probabilities given a positive oximetry result (likelihood ratio: 19.6) or a negative/inconclusive oximetry result (likelihood ratio: .58).
Several groups have reported on the use of pulse oximetry for children having adenotonsillectomy. However, none of these groups have compared the diagnostic accuracy of pulse oximetry to nocturnal PSG, the accepted standard for diagnosis of OSA.3 Van Someren et al29 reported that 15 of 44 children undergoing adenotonsillectomy had hypoxemia the night before the operation. Stradling et al30 studied 61 snoring children who were having adenotonsillectomy primarily for recurrent tonsillitis; they found a decrease in the desaturation index from 3.6 to 1.5 events per hour after the operation. Vavrina,31 using a computerized oximetry analysis, found that 25% of children undergoing tonsillectomy and adenoidectomy showed a characteristic pattern of repeated oxygen desaturations during sleep.
Two studies have linked OSA, desaturation events, and sleep disturbance to daytime behavioral performance using psychometric testing or school performance. Ali et al32 found that sleep disturbance, daytime behavior, and vigilance improved after adenotonsillectomy in a group of children who averaged 2.9 desaturation events per hour preoperatively. Gozal33 found that low-achieving first grade students with symptoms of OSA and a mean of 5 desaturations per hour during sleep improved their school performance after having adenotonsillectomy. The median desaturation index and MOAHI of our positive pulse oximetry group were 3 to 5 times greater than the desaturation index of the groups reported by Gozal and Ali, respectively. Thus, children identified by pulse oximetry testing are likely to have clinically significant disease severity.
Pulse oximetry has been suggested in lieu of PSG for other patient populations. In children with myelomeningocele, Waters et al34 found that oximetry was useful for preliminary testing, detecting all cases of moderate to severe sleep-disordered breathing. However, their patients with a positive oximetry required PSG to clarify the type of sleep disordered breathing: central apnea, obstructive apnea, or central hypoventilation. In adults being tested for OSA, pulse oximetry has performed well for detecting individuals with severe OSA (high apnea/hypopnea indexes).35,,36,13However, Epstein and Dorlac14 found that pulse oximetry, as a screening test for adult OSA, was not cost-effective and misdiagnosed a significant number of patients with treatable sleep disorders. Adults too may have significant sleep disturbance from OSA without desaturation.37
Care is required before nocturnal pulse oximetry is implemented as a testing modality for pediatric OSA. The accuracy of oximetry as an abbreviated test for OSA depends on the patient population, technician and physician experience, methodologic details, and oximeter brands. In this series, we were careful to exclude infants and patients who might have had central apnea, hypoventilation, and pulmonary and neuromuscular diseases that might have nonobstructive causes for hypoxemia. Too few patients <1 year of age were evaluated to allow firm conclusions for young infants. For this study, sleep laboratory technicians applied the pulse oximeter probes either in the sleep laboratory or in the child's home. In practice, parents would have to come to the hospital, learn to apply the oximeter, set it up at home, and return the machine after study completion. Since completion of this study, we have attempted home pulse oximetry in 33 children, 10 of whom were referred for possible OSA. In each case, a usable oximetry trend graph was obtained.
Different brands of oximeters possess different performance properties. Averaging times and motion artifact rejection algorithms affect displayed Sao2 and vary across brands and even across modes in the same model. Because movement artifact can cause false decreases in saturation, any proposed technique must be able to distinguish true desaturations from movement artifact.16 The most accurate method to distinguish true from false desaturations is to have the pulse wave form available; alternatively, a strong, stable pulse amplitude signal before and during desaturation events can be used.16 New oximeters are now available with the ability to read Sao2 even during motion such as that occurring at apnea termination.38
How could pulse oximetry be used for pediatric OSA? Most children suspected of having OSA have adenotonsillar hypertrophy and sleep-related signs and symptoms of airway obstruction but are otherwise well. In this group, a positive oximetry trend graph increases the probability of having OSA to nearly 100%. For these children no additional testing should be needed before adenotonsillectomy. However, careful perioperative monitoring, including postoperative, inhospital observation is recommended.39,,40 A negative or inconclusive oximetry result would require a subsequent PSG to definitively rule out OSA.
Although a complete cost–benefit analysis is beyond the scope of this article, abbreviated testing with oximetry might be economical in certain circumstances. Equipment and supplies cost less for the simpler oximetry testing. In our experience, technician time is <2 hours for oximetry versus at least 16 hours for standard PSG. Physician interpretation time is similarly reduced.
For those children with disorders other than adenotonsillar hypertrophy that might affect breathing during sleep, pulse oximetry could be used to quickly and inexpensively determine whether PSG is needed urgently to determine the type and severity of sleep-disordered breathing.41,,42 Oximetry screening might also be particularly useful in geographical areas where pediatric PSG is not readily available. Pulse oximeters are widely available and easy to apply. We have found that reading graphical trend and event summaries is very fast (<2 minutes per study in the current series).
Several studies have found that parental responses to questions about their child's breathing during sleep are too inaccurate to rule in or rule out OSA.9 The OSA score, which integrates responses to questions about snoring, difficulty breathing, and observed apneas, has not been able to accurately distinguish between children with OSA and those with simple snoring. The data summarized in Table 3 and Fig 3A do demonstrate, however, that children with a more dramatic history for OSA are more likely to have OSA than those who are less symptomatic.
This work was supported by The Hospital for Sick Children Foundation, the Canadian Foundation for the Study of Infant Deaths, Jeremy Rill Center for Sudden Infant Death Syndrome and Respiratory Control Disorders, and the Montreal Children's Hospital, McGill University Research Institute. Dr Waters was funded by a fellowship from the Canadian Lung Association/Medical Research Council of Canada.
- Received March 19, 1999.
- Accepted July 22, 1999.
Reprint requests to (R.T.B.) Montreal Children's Hospital, 2300 Tupper St, Montreal, Quebec, Canada H3H 1P3. E-mail:
- OSA =
- obstructive sleep apnea •
- PSG =
- polysomnography •
- Sao2 =
- hemoglobin saturation •
- MOAHI =
- mixed/obstructive apnea/hypopnea index •
- REM =
- rapid eye movement
- Ali NJ,
- Pitson DJ,
- Stradling JR
- Corbo GM,
- Fuciarelli F,
- Foresi A,
- De Benedetto F
- Wang RC,
- Elkins TP,
- Keech D,
- Wauquier A,
- Hubbard D
- Poets CF,
- Stebbens VA,
- Samuels MP,
- Southall DP
- McNamara F,
- Issa FG,
- Sullivan CE
- ↵Rosner B. Fundamentals of Biostatistics. 4th ed. Belmont, CA: Duxbury Press; 1995:423
- Gozal D
- ↵Guilleminault C, Stoohs R, Clerk A, Simmons J, Labanowski M. From obstructive sleep apnea syndrome to upper airway resistance syndrome: consistency of daytime sleepiness. Sleep. 1992;15:S13–S16. Supplement
- ↵Barker SJ, Shah NK. The effects of motion on the performance of pulse oximeters in volunteers. Anesthesiology. 1997;86:101–108. Revised publication
- Rosen GM,
- Muckle RP,
- Mahowald MW,
- Goding GS,
- Ullevig C
- Copyright © 2000 American Academy of Pediatrics