Heading

INTRODUCTION:   Auto-scored in-home polysomnography (PSG) accuracy can be impacted by the signal quality associated with self-application.  This study evaluated criteria to identify when the auto-scored measures of sleep architecture and sleep-disordered breathing will be equivalent to edited results.  

METHODS::  This retrospective analysis was conducted with 218 consecutive, complete Sleep Profiler PSG2TM studies (Advanced Brain Monitoring, Carlsbad, CA) that included 54% males, mean age 45+13.4 years, and BMI 31+7.0 kg/m2.  Patients self-applied a forehead-worn device to affix three frontopolar electroencephalography (EEG) sensors and a nasal cannula, a wireless wrist oximeter, and thorax and abdomen RIP belts.  After watching a 5-min video, patients practiced affixing the device.  During the night, voice messages alerted the patient if the SpO2 sensor probe was not affixed, and up to four times/night when the cannula was misplaced.

Auto‐staging was performed using previously described techniques that relied on the ratios of the power spectral densities and auto‐detection of cortical and micro‐arousals, sleep spindles, and ocular activity obtained from Af7-Af8 (EEG), Af7-Fpz (LEOG), and Af8-Fpz (REOG) signals.  Auto-detected OSA severity was based on a 10-sec cessation in airflow for apneas and a 30% reduction in airflow plus a 4% desaturation to emulate the AASM2007 criteria (AHI2007), or a 3% desaturation or a cortical arousal to confirm hypopneas for the  AASM2012 criteria (AHI2012).  The same technician performed a focused review to confirm the sleep staging accuracy, insert apnea or hypopnea events during periods with airflow or SpO2 loss, or remove events due to artifact/movement.

Failed studies were characterized by thresholds applied to airflow, SpO2 and/or EEG signal loss as defined in Table 1.  The balance was further stratified into High Quality, Marginal EEG, and Airflow/SpO2 groups to assess the impact of editing on auto-scored results.  Bonferroni-corrected t-tests were reported with p<0.001.  Classification accuracies were derived from the comparisons of auto-scored AHIs to the edited AHIs (i.e., represents the “presumed” true OSA severity).

RESULTS:  A PSG2 failure rate of 6% was attributed to nine records with extensive airflow/SpO2 loss and four records with excessive EEG loss.  From the 205 acceptable records, minimum recording times of 6, 7, and 8 h were obtained in 91%, 71%, and 44% of the records, respectively.  Auto-staged (Auto) and edited recording times were significantly less in the studies with excessive airflow/SpO2 loss vs. the other two groups.  Table 2 presents changes in the proportion of records with minimum sleep times, stratified by signal quality, before and after editing (shaded values identify important differences).

Table 2: Percentage of records in each signal quality group with auto vs. edited minimum sleep times > 4, 5 and 6 h

World Sleep Congress October 2017, Prague Czech Republic The median time to edit the records was 10-min (Inter-quartile range 10-15 min) with increased data loss having no impact on the time taken to edit the records.  Records with Airflow/ SpO2 Marginal had substantially less auto-staged sleep time due to the assignment of invalid to periods with poor signal quality.  In the High Quality group, editing significantly decreased wake after sleep onset (WASO) and stage N1 and increased REM time.   After editing, the High Quality studies had significantly greater sleep time vs. the two groups with marginal signal quality.  The significantly greater WASO observed in the Marginal EEG records (vs. the other two groups) may have resulted from poor signal quality.  Longer Auto and edited REM times in the High Quality group (as compared to the other two groups) may be explained by increased sleep times.

Table 3: Mean + SD of significant sleep architecture measures, stratified by auto vs. edited and signal quality group

Differences in the distributions of OSA severities before and after editing are presented in Table 4 with proportional differences > 5% highlighted.   Differences between Auto and edited AH²⁰⁰⁷ and AHI²⁰¹² when signal quality was high were negligible.  Editing increased the proportion of patients with more severe AHI²⁰¹² in records with marginal signal quality.

Table 4: Distributions of records stratified by OSA severity for edited vs. auto-scored, AHI criteria and signal quality group

Table 5 presents classification accuracies between the edited vs. auto-scored AHI2007 and AHI2012, stratified by clinical cutoffs of 5, 10, and 15 events/h and signal quality groups.  Sensitivity, specificity, positive percent agreement (PPV) and negative percent agreement (NPV) less than 0.85 are highlighted.

Table 5: Agreement between edited vs. auto-scored AHI values

CONCLUSIONS:  Over 90% of the studies included a minimum of 6 h of study time and ~80% had 5 h of edited sleep time.   Editing impacted sleep architecture by converting epochs staged N1 to REM, and reducing WASO.  Using the AHI2007 criteria, the edited and auto-scored OSA severities were equivalent across all clinical cut-off and data quality groups.   The auto-scored AHI2012 provided a reliable estimate of the OSA severity at clinical cut-offs >10 or >15 events/h when signal quality was high, and at >15 events/h with Marginal EEG.  Results obtained when the airflow/ SpO2 signal quality thresholds were exceeded must be edited prior to interpretation.

Get in touch
Contact Us