Validation of self-applied unattended polysomnography using Somte V2 PSG (Somte) for diagnosis of obstructive sleep apnoea (OSA) in pregnant women in early to mid-gestation

Our report follows Standards for Reporting Diagnostic Accuracy (STARD) guidelines for reporting diagnostic studies [8] (Online supplement 1).

A total of 410 participants were screened for eligibility and eighty-four participants consented for participation. Thirty-two consented participants were ineligible to complete Somte as they were recruited prior to Somte addition to the study protocol. Of the fifty-two consented participants recruited after addition of Somte to the study protocol, thirty-four participants underwent a Somte, and twenty-nine participants completed PSG. Both Somte and PSG were attempted by twenty-nine participants and of these, one was excluded due to an extended duration of time between tests (17 days) and four excluded due to Somte failure to start. Subsequently, twenty-four participants are included for data analysis. The study flow chart is shown in Fig. 2. Baseline demographics data of study participants (n = 24) is shown in Table 1.

OSA (AHI ≥ 5) was identified in 5 participants (20.8%) on PSG and 5 participants (20.8%) on Somte. RDI ≥ 5 was identified in 10 participants (41.7%) on PSG and 7 participants (29.2%) on Somte. Percentage of REM achieved during sleep did not differ between Somte (21.3%) and PSG (20.2%) (p = 0.179). The median (IQR) and range (minimum, maximum) of Somte and PSG (AHI and RDI), and sensitivity, specificity, PPV, NPV and area under ROC curve (AHI ≥ 5 and RDI ≥ 5) is shown in Table 2. Likelihood ratio for AHI ≥ 5 for Somte compared to PSG was 15.20 (positive) (95% CI 2.15 to 107.63) and 0.21 (negative) (95% CI 0.04 to 1.22), Likelihood ratio for RDI ≥ 5 for Somte compared to PSG was 8.40 (positive) (95% CI 1.19 to 59.36) and 0.43 (negative) (95% CI 0.20 to 0.93).

PSG were completed at around 14.1 weeks’ gestation (IQR 13.4, 15.7). The time interval between Somte and PSG was a median of 4 days (IQR 2, 7; range 1–12). The median TAS for Somte and PSG was 461.8 min (IQR 386.6, 506.5) and 526.3 (IQR 499.1, 535.4) (p =  < 0.001) respectively. The Median TST for Somte and PSG was 391.3 (IQR 320.4, 429.1) and 438.2 (IQR 345.4, 475.9) (p = 0.10), respectively. Sleep study Median (IQR) of AHI and RDI of Somte and PSG, and additional sleep variable data are shown in Table 3.

ROC curves are shown in Fig. 3a, b. Positive and negative predictive values were calculated using cross tabulation in SPSS and are shown at Fig. 4. The Kappa coefficient for AHI ≥ 5 comparing PSG and Somte is 0.747 ± 0.17, and for RDI ≥ 5 comparing PSG and Somte is 0.552 ± 0.17. Bland–Altman plots for AHI ≥ 5 and RDI ≥ 5 are shown at Fig. 5a, b. No adverse events were reported by participants undertaking Somte and/or PSG.

One participant displayed increased muscle tone on chin EMG during rapid eye movement (REM) on Somte and PSG. Additionally, the participant demonstrated frequent motor activity (subtle head movements and one episode of body jerking) during each REM period, as evidenced on video PSG, and was subsequently diagnosed with REM sleep without atonia (RSWA).

In addition to the four Somte studies that failed to start, technical issues were seen in both Somte (n = 13, 54.2%) and PSG (n = 6, 25.0%), p = 0.41 (Fisher’s exact test). Somte reported signal quality loss in flow (n = 3, 12.5%), SpO₂ (n = 3, 12.5%), EEG (n = 3, 12.5%), EMG (n = 6, 25.0%), and ECG (n = 2, 8.3%). PSG studies reported signal quality loss in airflow (n = 1,4.2%), SpO₂ (n = 2, 8.3%), EEG (n = 2, 8.3%) and EMG (n = 3, 12.5%). None of the signal quality losses, in either PSG or Somte, resulted in a non-diagnostic study. However, 2 Somte studies used surrogate imputations for scoring the study, as permitted in the AASM scoring guidelines.

This work demonstrates Somte can be self-applied by pregnant women, and accurately diagnoses OSA in early to mid-pregnancy. At the clinically significant threshold of AHI ≥ 5, compared to gold standard PSG, Somte demonstrated excellent specificity, NPV, and area under the ROC curve, but had a lower sensitivity and PPV. At RDI ≥ 5, Somte demonstrated a lower specificity, NPV and sensitivity compared to PSG using AHI, though reported a higher area under the ROC curve and PPV. Proportional bias was noted on the Bland–Altman plots, with Somte underestimating AHI and RDI scores at higher severity scores. Technical issues occurred in Somte and PSG, although, except for the Somte which failed to start, these did not preclude the usefulness of the test.

Other studies investigating agreement in AHI scores between Somte and PSG have reported differing findings to our study. Ferré et al. reported similar area under the ROC curve to our study at AHI ≥ 5 of 0.90 vs 0.94 though differences in sensitivity (90.5% vs 80.0) specificity (83.5% vs 94.7%), PPV (96.0% vs 80.0%) and NPV (63.5% vs 94.7%) which may reflect the differences in age, gender, and severity of OSA between the two studies [16]. Cunnington et al. did not compare results at cut off AHI ≥ 5 but at cut off AHI ≥ 10, reported area under the ROC curve of 89.4, sensitivity 96.7%, specificity of 40.0%, PPV 91.2% and NPV 65.4%. This represents a higher sensitivity than our study (96.7% vs 80.0%) but lower specificity (40.0% vs 94.7%) and NPV (65.4% vs 76.4%) [17]. Our study reports lower sensitivity than both Ferré and Cunnington, though higher specificity and this may reflect the different patient population in which the tests are conducted. Further, both Ferré and Cunnington conducted simultaneous tests and use different EEG positions than our study (central vs frontal), which may have resulted in different results. A systematic review investigating portable sleep diagnostic tests report overall support of the use of level II devices, with specificity increasing with increased OSA severity [18]. The low specificity of our study findings may reflect the low AHI scores in our study.

The major strength of this study design is the completion of the index and reference tests during early to mid-gestation. This may improve early access to PSG in the early to mid-gestation period as recommended for at risk women [3]. By using a self-application method, we have been able to demonstrate accuracy compared to attended PSG, for pregnant women requiring sleep diagnostic testing.

Further strengths of this study include use of blinding of the reference test result to minimise bias, the use of a single software package for the scoring of Somte and PSG data, the use of a single scorer across both the reference and index tests and inclusion of pregnant women with and without traditional signs or symptoms of OSA. By including pregnant women with personal or family history of GDM and/or preeclampsia, this makes our data relevant for real life clinical applications, as our participants broadly reflect the population recommended for OSA screening in pregnancy [3].

Four Somte tests failed to start following completion of the set-up process by the participant and the cause was not identified. Subsequent discussions with the manufacturer point to the possibility the device was draining the battery whilst in standby mode. Additional technical difficulties that resulted in signal loss or dropout were reported in 54.2% of our Somte studies although these did not result in test failure. Failure of Somte has been reported previously by Light et.al with 14/221 (6%) of studies failing due to poor oximetry trace (Sp0₂) when conducted in the home, in non-pregnant adults, following set-up by trained technicians [19]. We note the sleep scientist involved in scoring all Somte and PSG in our study has significant experience in scoring sleep studies and we would caution the use of inexperienced scorers to score unattended PSG of pregnant women who have self-applied the device, due to the high rate of technical issues encountered in our participants, despite extensive education prior to testing, and on-call phone support available during the test.

Our study limited Somte EEG to F3/4 only which was repositioned to the forehead. Use of repositioned F3/4 (with M1/2 referencing) was chosen for ease of application and has been investigated in other studies involving non-pregnant adults. Ferré et.al investigated agreement between Somte with F3/4 EEG only, and simultaneous PSG, demonstrating good agreement in sleep efficiency (68%), specificity (84%), and area under the ROC curve (0.86) at AHI cut off ≥ 5, and high sensitivity (91%), though did report a statistically significant difference in AHI scores [16]. Further, Light et.al investigated scoring of sleep vs wake in non-pregnant adults, using F3/4 only EEG, repositioned to the forehead, similarly to our study. Lights’ study scored a single night of full PSG study twice, one with access to all collected EEG signals, and one with access to only F3/4 (other signals hidden) [19]. The study demonstrated high agreement (94%) in sleep vs wake epochs in studies using only F3/4 compared to those with full EEG. Our study did not complete the tests simultaneously and this may have contributed to the differences in reporting of maximum AHI and RDI scores in Somte studies (26.9, 29.8) compared to PSG (18.1, 19.5). Whilst the overall AHI and RDI scores did not differ between tests, higher RDI scores in both Somte and PSG is consistent with other studies reporting higher RERA's in pregnant women [20]. Given the implications of RERA’s in endothelial dysfunction [21, 22], Somte represents an opportunity to detect increased upper airway burden in pregnant women, in early to mid-gestation, without requiring an overnight stay in the sleep laboratory.

Whilst our study aim was not to assess movement disorders of sleep during pregnancy, one participant in this study was diagnosed with RSWA based on PSG findings. Additionally, this participant demonstrated increased chin EMG during REM in the Somte study, suggesting this incidental finding was replicated in the home environment. The participant reported no history to suggest a presence of a movement disorder and we did not use extended EMG (SINBAR) protocol during the study collection, however the reporting physician determined sufficient evidence on video and polysomnography for a clinical determination of RSWA. The significance of this incidental polysomnographic finding is unclear.

The main limitation of this study was a small sample size. Low study participation rates and high rates of failure to attend after consent, resulted in a lower-than-expected number of participants. This study was completed during a period of political and social instability due to the SARS COV2 pandemic and access to patients was restricted several times through the collection period, so it was unsurprising that we encountered difficulty. In other sleep diagnostic validation studies in pregnant women, conducted in later gestation, similar sample sizes are generally reported. A study investigating home PSG and simultaneous wrist worn Watch-PAT 200 in pregnant women with gestation ≥ 28 weeks, achieved a sample size of 31 [23]. A further study investigating attended PSG and simultaneous ARES Unicorder, in pregnant women 28.6 ± 6.3 weeks gestation achieved a sample size of 16 participants [24]. A study which compared limited channel Apnealink to PSG in late gestation (≥ 28 weeks) within a 14 day window between tests, achieved a sample size of 30 participants [5]. The largest validation study in late gestation is a multicentre study investigating level III Alice PDX compared to attended PSG, in the third trimester, which achieved a sample size of 149 participants, though demonstrated poor agreement in AHI scores [25].

In early to mid-gestation, the largest study investigating agreement with attended PSG, is reported by our group. Our study, which validated Level III Apnealink Air compared to attended PSG in pregnant women around 14 weeks gestation, achieved a sample size of 49 participants and demonstrated good agreement in AHI compared to PSG at AHI ≥ 5 and included participants with severe OSA on PSG (range 0.0–119.7) [4]. Together, our validation studies present two diagnostic test options for use in the diagnosis of OSA in pregnant women in the early to-mid gestation period. We were unable to confirm the validity of the Somte in severe OSA cases in pregnancy and note the overall poorer performance of Somte to detect OSA using RDI. Larger studies to validate Somte in severe OSA, and to verify the performance of the RDI using the Somte device would be beneficial. Due to the study limitations including small sample size, technical issues, and absence of severe OSA amongst study participants, further studies verifying these findings would be beneficial prior to translating these findings into clinical practice.

Self-applied, unattended Somte was an acceptable substitute to attended PSG in the identification of OSA in pregnant women in early to mid-gestation in this small sample though may fail to detect OSA in some women, particularly when using RDI scores. This provides an opportunity to improve early access to diagnostic testing in the early to mid-pregnancy as recommended for at risk women.