INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Shortly after the Multiple Sleep Latency Test (MSLT) was developed as a measure of physiological sleep tendency, investigators observed that narcoleptics, in contrast to normal subjects, often had rapid eye movement (REM) sleep during 2 or more daytime nap attempts (1, 2). Two or more sleep onset REM periods (2omSOREMPs) on an MSLT were thought to be highly sensitive and specific for narcolepsy. However, in more recent years, obstructive sleep apnea (OSA) has become the predominant cause of excessive daytime sleepiness among patients tested at sleep disorders laboratories (3), and the diagnostic predictive value of 2omSOREMPs for narcolepsy has decreased because they also occur in OSA, though much less frequently than in narcolepsy (4).

In narcoleptics the occurrence of 2omSOREMPs is consistent with other signs of dysfunction in the regulation of REM sleep, but no similar neurophysiological construct is available to explain why some apneics also have 2omSOREMPs. Recent reports that normal subjects sometimes have 2omSOREMPs suggest that their occurrence in apneics may not reflect any abnormality (5, 6). However, two other reports have suggested that among patients with OSA, both apnea severity (7) and excessive daytime sleepiness (4, 7) are associated with an increased likelihood of 2omSOREMPs. One of these reports appeared in preliminary form only, the other applied no statistical analysis to the data in question, and because neither used statistical or other techniques to control for possible confounding variables, neither could indicate whether or to what extent sleepiness, apnea severity, or nocturnal sleep architecture were independently predictive of 2omSOREMPs.

One potential explanation for increased REM sleep pressure among some apneics could be nocturnal, selective REM sleep deprivation due to the tendency of OSA to be most severe during this sleep stage (8). However, preliminary work has suggested that sleepiness, in addition to apnea severity, also correlates with the likelihood of a SOREMP; that the amount of nocturnal REM sleep is not predictive of 2omSOREMPs; and that apnea severity during non-REM sleep, not REM sleep, is associated with 2omSOREMPs (7).

Finally, the possible influence on SOREMPs of other important variables, such as age and sex, has not been adequately explored. Among normal subjects, youth and male sex were reported to be predictive of 2omSOREMPs (6), but whether these variables might affect the occurrence of 2omSOREMPs among patients evaluated for OSA has not been adequately studied.

To better understand underlying physiology and to facilitate clinical interpretation of 2omSOREMPs in patients with suspected or confirmed OSA, we used polysomnographic, MSLT, and demographic data from a large series of patients and examined associations between the occurrence of 2omSOREMPs and variables that could influence REM sleep tendency.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

We have previously used our sleep laboratory patient database to investigate relations between sleepiness and polysomnographic characteristics of OSA (8, 9), and to describe the value of MSLT findings in the diagnosis of narcolepsy (4). Our series of patients with suspected or confirmed OSA is the largest reported, to our knowledge, that contains both polysomnographic and MSLT data, and this report extends our analysis to include detailed data relevant to SOREMPs. The sample size for this report is n = 1,145 patients, which differs from two previous reports on the same group (8, 9) only in that one patient's data are now excluded because multiple SOREMPs in this case were attributed to recent discontinuation of methylphenidate taken for attention-deficit/hyperactivity disorder.

Subjects were originally selected, from a database of over 9,000 studies performed in our laboratory, if they (1) had diagnostic polysomnography between January 1, 1988 and February 1, 1997; (2) had MSLTs on the next day, generally because excessive daytime sleepiness was suspected and objective assessment was desired by the referring physician; (3) were free of psychoactive medications (tricyclic antidepressants, benzodiazepines, other hypnotics, and antipsychotic agents); and (4) were studied for suspected OSAS, had a final diagnosis of some form of sleep-disordered breathing, or were studied for suspected excessive daytime sleepiness and had no final sleep-related diagnosis that explained the symptom. Patients were excluded if they (1) were studied for suspected narcolepsy or were found to have narcolepsy; (2) had other indications for study or other major diagnoses that could explain excessive daytime sleepiness, such as periodic limb movement disorder, idiopathic hypersomnia, or significant medical conditions such as congestive heart failure; or (3) had either no REM sleep or no non-REM sleep on their polysomnogram. Our criteria for narcolepsy were described in detail previously (4) and differed from International Classification of Sleep Disorders (ICSD) (10) criteria mainly in that we used an MSLT mean sleep latency (MSL) < 8 min rather than an MSL < 5 min as supportive of the diagnosis.

Procedures

Nocturnal polysomnography included four electroencephalographic leads (C3-A2, C4-A1, O1-A2, O2-A1 of the 10-20 international electrode placement system), two electro-oculographic leads (right and left outer canthi), chin and bilateral anterior tibialis surface electromyograms, two electrocardiographic leads, nasal and oral airflow (thermistors), thoracic and abdominal excursion (piezoelectric strain gauges), and finger oximetry. A small proportion of the subjects also had esophageal pressure monitoring with a water-filled catheter (11) that has little or no effect on sleep architecture (12). Patients generally chose their own bedtimes and then had approximately 8 h to sleep. Sleep stages were scored in 30-s epochs according to standard criteria (13) by technologists who, after an extensive training program, had correctly scored at least 90% of epochs in a set of reliability records. Technical and scoring procedures have remained equivalent in our laboratory during the past 9 yr under the direction of the same medical director and chief technologist. An apnea was defined as 10 or more seconds of complete airflow cessation during sleep, regardless of any change in oxygen saturation. An hypopnea was defined as a reduction in airflow, chest excursion, or abdominal excursion that led to a 4% or greater oxyhemoglobin desaturation, an arousal, or an awakening. Minimal oxygen saturation for each study was defined as the lowest artifact-free level recorded during sleep, but was available for only 1,097 of the 1,145 polysomnograms.

The MSLTs followed standard methods for collection of electroencephalographic, electro-oculographic, and chin electromyographic data (14). Four nap attempts generally occurred at 2-h intervals, the first nap starting 1.5 to 2 h after termination of the polysomnogram. Patients who, at the completion of 4 naps, had had exactly one SOREMP underwent a 5th nap attempt to determine whether they would qualify as having 2omSOREMPs (15). Each patient's MSL was calculated as the time, averaged across all nap attempts, from "lights out" to the first epoch of stage 1 sleep. Patients who fell asleep within 20 min of the beginning of a nap attempt were left undisturbed for up to an additional 15 min, during which the appearance of REM sleep caused the nap to be scored as a SOREMP. On the night before the MSLT, most patients also completed a sleep clinic questionnaire that included a question used with identical (16) or similar (17) wording in previous studies as a subjective measure of problem sleepiness: "How often do you have a major problem with sleepiness in the daytime?" Answers were provided on a five-point Likert scale from 1 (never) to 5 (almost always).

Data Analysis

Results were summarized as means ± standard deviations or as frequencies and were compared, for patients with and without 2omSOREMPs, using t tests and chi-square tests. The main outcome variable, presence or absence of 2omSOREMPs on the MSLT, was modeled by logistic regression with four categories of explanatory variables: (1) Demographic variables included age, gender, and body mass index (BMI, weight in kilograms divided by height in meters squared); (2) Measures of sleepiness included one that was objective (MSL) and another that was subjective (problem sleepiness, as previously described); (3) Nocturnal polysomnographic measures of sleep architecture included minutes spent in each sleep stage, percentage of total sleep spent in each sleep stage, sleep efficiency (SE, total sleep time divided by total recording time), latency to sleep (LS, time from "lights out" to first epoch of any sleep stage), and latency to REM sleep (LR, time from first epoch of sleep to first epoch of REM sleep); and (4) Measures of apnea severity, derived from full-night polysomnography, included the number of apneas and hypopneas per hour of sleep (apnea/hypopnea index [AHI]), that number per hour of non-REM sleep (AHIN), that number per hour of REM sleep (AHIR), and the minimal oxygen saturation recorded (minO₂).

Data were analyzed with SAS version 6.12 (SAS Institute Inc., Cary, NC). Logistic regression p values were calculated from the likelihood ratio chi-square (for overall model results) and the Wald chi-square (for explanatory variables in multiple logistic regressions). Odds ratios (OR) and 95% confidence intervals (95% CI) were calculated with the profile likelihood method. In statistical tests, the significance level was set at 0.05.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Overview

The 1,145 subjects had a mean age of 45 ± 13 yr (range 6 to 85) and 342 (30%) were female. Nocturnal polysomnographic, MSLT, and demographic data are summarized in Table 1. Fifty-four of the subjects (4.7%) had 2omSOREMPs; 980 (85.6%) had no SOREMPs, 111 (9.7%) had one, 44 (3.8%) had exactly two, 10 (0.9%) had 3, and none had more than 3. Subjects with 2omSOREMPs, in comparison to those without 2omSOREMPs, were naturally more likely to have a SOREMP on any given nap, but the relative distribution between naps of the SOREMPs, within each group of subjects, was nearly identical.

The 2omSOREMPs were associated, to a statistically significant extent, with male sex, low MSL, low nocturnal latency to sleep, low nocturnal latency to REM sleep, low percentage of stage 2 sleep, high AHI, low minO₂, and high AHIN (Table 1). The 2omSOREMPs showed no association with reduced amounts of nocturnal REM sleep. To better compare the magnitude and variation of the significant associations, results from logistic regressions of 2omSOREMPs on specified degrees of change in each variable are shown in Table 2. The degree of change modeled for each variable was selected to be clinically meaningful and to approximate one standard deviation in the distributions listed in Table 1. These results suggest that the variables with the largest and most consistent associations with 2omSOREMPs were male sex, MSL, and minO₂.

Male Sex

The OR for 2omSOREMPs and male sex was 4.380 (p = 0.0002, Table 2); while 6.1% of males had 2omSOREMPs, only 1.5% of females had this result. Possible confounds in the association between 2omSOREMPs and male sex were tested in a multiple logistic regression that included, in addition to male sex, several other explanatory variables that might be expected to covary with male sex: MSL, AHI, and minO₂. The odds ratio for male sex did not change appreciably (4.586, Wald chi-square p = 0.0042, 95% CI 1.816 to 15.443), though MSL and minO₂ did retain additional significant associations with 2omSOREMPs. To explore whether MSL, AHI, or minO₂ might act as an effect modifier in the relation between male sex and 2omSOREMPs, each of the three was tested as an interaction variable in three separate logistic regressions. None of the interaction terms approached statistical significance (all p > 0.10).

Sleepiness

The OR for 2omSOREMPs and a 5-min decrease in MSL was 1.895 (p = 0.0003, Table 2). Whereas 6.5% of subjects with MSL from 0 to 4.9 min had 2omSOREMPs, only 5.0% of those with MSL from 5 to 9.9, 2.5% of those with MSL from 10 to 14.9, and 1.0% of those with MSL from 15 to 20 had 2omSOREMPs (Figure 1). Variables tested as possible confounds of the association between lower MSL and 2omSOREMPs included male sex, percentage of stage 2 sleep, AHI, and minO₂. In a multiple logistic regression that included these variables, the OR for 2omSOREMPs and a 5-min decrease in MSL retained significance (OR = 1.667, Wald chi-square p = 0.0167, 95% CI 1.119 to 2.591). Male sex and minO₂ also retained significance (p < 0.005 for each), but AHI and percentage of stage 2 sleep did not (p > 0.10 for each). Among the sleep architecture variables (Table 1) that showed a significant or nearly significant (0.05 < p < 0.10) association with 2omSOREMPsentries to stage 1, latency to sleep, latency to REM sleep, stage 1 sleep, stage 2 sleep, percent stage 1, and percent stage 2none except entries to stage 1 and latency to REM sleep continued to show such an association (p = 0.0808 and p = 0.0202, respectively) in models that controlled for MSL.

View larger version (10K): [in this window] [in a new window]   Figure 1.   Probability of 2 or more sleep onset REM periods (P2SOREMP), with upper and lower 95% CI (UpperCI and LowerCI), plotted as a function of MSL (in minutes), as predicted by a logistic regression model.

Sleep Architecture

In view of this evidence that any potential influence of most sleep architecture variables on 2omSOREMPs might be mediated by intermediary effects on sleepiness (MSL), only latency to REM sleep was further investigated for a possible independent effect. In a simple logistic regression, the OR for 2omSOREMPs and a 90-min decrease in nocturnal REM latency was 1.614, p = 0.0086 (Table 2 and Figure 2). In a multiple logistic regression that controlled for MSL, sex, AHI, and minO₂, nocturnal REM latency retained significance (OR = 1.630, 95% CI 1.089 to 2.592, p = 0.0266); the other variables, except for AHI, also showed significant associations (p < 0.05) with 2omSOREMPs. To see whether the relation between 2omSOREMPs and nocturnal latency to REM was different for males and females, an interaction model was tested but the interaction term showed no significance (p = 0.7601).

View larger version (10K): [in this window] [in a new window]   Figure 2.   Probability of 2 or more sleep onset REM periods (P2SOREMP), with upper and lower 95% CI (UpperCI and LowerCI), plotted as a function of LR (in minutes), as predicted by a logistic regression model.

Apnea Severity

Although 2omSOREMPs were associated with both AHI and minO₂ in separate simple logistic regression models (Table 2), the relationship with minO₂ was stronger (Figures 3A and 3B). The association between 2omSOREMPs and a 15 percentage point decrease in minO₂ showed little change after accounting for AHI (OR = 1.689, 95% CI 1.291 to 2.203, p < 0.0001), but the relationship of 2omSOREMPs to AHI showed no significance after accounting for minO₂ (p = 0.6159). In addition, tests of possible confounding in the relation between minO₂ and 2omSOREMPs showed that this association remained (OR = 1.433, 95% CI 1.096 to 1.856, p = 0.0070) after controlling for all other gender, sleepiness, and sleep architecture variables listed in Table 2.

View larger version (10K): [in this window] [in a new window]   View larger version (10K): [in this window] [in a new window]   Figure 3.   Probability of 2 or more sleep onset REM periods (P2SOREMP), with upper and lower 95% CI (UpperCI and LowerCI), as predicted by a logistic regression model and plotted as a function of (A) the number of apneas and hypopneas per hour of sleep (AHI), and (B) the minimal nocturnal oxygen saturation (minO2).

Best Model for Prediction of 2omSOREMPs

A stepwise multiple logistic regression was used to derive a final model for the prediction of 2omSOREMPs. Explanatory variables considered were all those listed in Table 1 except for three with limited numbers of data points (BMI, problem sleepiness, and entries to stage 1) and "SOREMPs"; the total number of subjects available for this model was n = 1,095. The process identified the highly significant model (p < 0.0001) shown in Table 3 and resulted in retention of one main element from each of the major variable groups: demographics (male sex), sleepiness (MSL), sleep architecture (latency to REM sleep), and apnea severity (minO₂). Each retained variable remained significantly associated with 2omSOREMPs (p < 0.05) after adjusting for the other variables in the model. A Hosmer and Lemeshow goodness of fit test failed to demonstrate any significant lack of fit (p > 0.10).

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

This study of patients with suspected or confirmed OSA rather than narcolepsy identified four main variables that are associated with the appearance of 2omSOREMPs on MSLTs: male sex, sleepiness, short nocturnal latency to REM sleep, and minimal oxygen saturation. Many such variables vary together among OSA patients, and the current study appears to be the first to identify and quantify independent contributions likely to be most useful in clinical practice and possibly most relevant to underlying physiology. The strong associations found in this observational study can suggest but not prove causality. The absence of hypothesized associations, however, in a statistically controlled study of this size and power does provide some evidence against causal relationships between the specified variables.

Male sex had the strongest potential influence on occurrence of 2omSOREMPs: a male patient was more than 4 times more likely than a female patient to have 2omSOREMPs. Results of multiple logistic regression showed that this effect was not the result of any tendency among males to have more severe sleep apnea, more severe sleepiness as measured by the MSLT, or increased REM pressure as reflected by nocturnal REM sleep latency. The association between male sex and 2omSOREMPs among OSA patients has not been reported previously, to our knowledge, but one study of normal subjects did find males to have 2omSOREMPs almost 3 times as often as females (6). This study did not control for sleepiness or nocturnal REM latency, and the investigators did not offer any explanation for their finding. We wonder whether males and females in the current study differed in use of illicit drugs, sleep deprivation prior to their laboratory testing, or concerns about missed activities during MSLT testing. However, it is difficult to imagine that any of these factors could have accounted for a 4-fold difference in the rates of 2omSOREMPs between the two sexes. We therefore speculate that the cause may have been a neurophysiological difference, as yet unidentified, between males and females.

The findings that 2omSOREMPs were predicted by MSLT-defined sleepiness but not by self-rated problem sleepiness probably reflects, in part, discrepancies between objective and subjective measures of sleepiness (16). Additional objective correlates of sleepiness, such as nocturnal sleep and REM sleep latencies, also showed associations with 2omSOREMPs. Adjustment for apnea severity had little effect on the association between MSLT-defined sleepiness and 2omSOREMPs. In previous publications, the tendency for patients with lower sleep latencies on MSLTs to have more SOREMPs has been noted in narcoleptics with cataplexy, narcoleptics without cataplexy, and normal subjects (4, 6). The consistency of this finding among different groups, along with the data from the current study, suggests that excessive sleepiness is one cause of SOREMPs. Increased sleep pressure, due to nocturnal sleep disturbance in some cases, could cause both increased sleepiness and SOREMPs. In experimental paradigms, acute sleep deprivation is usually followed by recovery of non-REM sleep before REM sleep, but chronic sleep restriction or nocturnal sleep disturbance often results in sleepiness and increased SOREMPs (19). Our study did not directly assess the possibility that chronic insufficient sleep, as opposed to disordered sleep, caused 2omSOREMPs in some subjects. However, several variables likely to reflect chronic sleep deprivationsuch as sleep efficiency, percentage of stage 3/4 sleep, percentage of REM sleep, and patterns of SOREMP occurrence across napsdid not correlate with the presence of 2omSOREMPs.

Our finding that shorter nocturnal REM latencies predict 2omSOREMPs suggests at least two possible explanations. First, some third variable not considered in our analyses could cause early REM sleep both at night and during the day. However, this study did include many of the variables that are commonly the focus in clinical and research polysomnography; sometimes the latter includes a more direct measure of arousals than our variable "entries to stage 1 sleep." A second, related possibility is that short latency to REM sleep represents a basic neurophysiological tendency that is expressed whenever sleep is obtained. Such an underlying tendency may reflect a particular genetic trait located at or near that for human leukocyte antigen DR2 (HLA-DR2), as narcoleptics without cataplexy but with 2omSOREMPs on MSLTs have an increased frequency of HLA-DR2 (20).

Among patients suspected or confirmed to have OSA, our data suggest that minimal oxygen saturation, not AHI, is most predictive of 2omSOREMPs. Research reports, review articles, and the ICSD have noted that 2omSOREMPs in a patient with OSA may not reflect comorbid narcolepsy, but the underlying explanation or assumption has often been that OSA-induced sleep disruption or deprivation, especially during REM sleep, causes increased REM pressure (10, 21, 22). This explanation is not supported by our results, which failed to show independent associations between 2omSOREMPs and any of the relevant measures, including AHI, AHI during REM sleep, amount of REM sleep, percent REM sleep, sleep efficiency, and entries to stage 1 sleep. We have also reported, in the past, that the extent of REM sleep rebound on initiation of treatment for OSA is best predicted by baseline minimal oxygen saturation (23) and that short latency to REM sleep rebound is predicted by minimal oxygen saturation during baseline REM sleep (24). Our data raise the possibility that hypoxemia promotes SOREMPs through a direct neurophysiological mechanism, perhaps by disruption of an oxygen- dependent metabolic process. We are unaware of other experimental evidence or clinical data, for example among patients with chronic obstructive pulmonary disease or obesity-hypoventilation syndrome, that might support or refute this hypothesis.

One possibility that could explain the occurrence of 2omSOREMPs in some of our patients is that some actually had narcolepsy, which often begins without cataplexy and can cause symptoms for many years prior to diagnosis (25). However, we believe that the number of our patients who had occult narcolepsy is not likely to be large. The presence of narcolepsy in some of our patients would not explain the association between apnea severity and 2omSOREMPs. In addition, narcolepsy occurs with equal frequency in males and females (10, 26), whereas among our subjects 2omSOREMPs were much more common in males.

In short, the findings of this study raise more questions than they answer about pathophysiological mechanisms responsible for 2omSOREMPs in some patients with sleep disorders. The association between 2omSOREMPs and sleepiness in non-narcoleptic subjects raises the possibility that among narcoleptics as well, 2omSOREMPs are to some extent a manifestation of excessive daytime sleepiness. Such a conclusion is supported by our previous finding that among narcoleptics with or without cataplexy, the probability of SOREMPs increases with increasing sleepiness as measured by the MSLT (4).

From a clinical standpoint, false-positive MSLT results that needlessly bring narcolepsy into a differential diagnosis otherwise centered on OSA provide a reason to exercise some discretion in the decision to obtain an MSLT. The MSLT may still be useful in a patient suspected to have OSA when an objective assessment of sleepiness is particularly important. In such circumstances, the results of the present study should be helpful in the interpretation of 2omSOREMPs: The diagnosis of narcolepsy should be considered, but among other evidence that may explain 2omSOREMPs in a non-narcoleptic, particularly useful features appear to be male sex, objectively measured excessive sleepiness, short nocturnal REM sleep latency, and apnea severity. Although 2omSOREMPs provide important evidence in the evaluation for narcolepsy, their predictive value is much higher when the patient's history also suggests this diagnosis (4).