Our study found excellent corelation between Philips Respironics Sleepwere G3 version 3.9.1 automated scoring and manual scoring of HSAT recorded with by Alice NightOne, a type III portable polygraphy device. The two-stage model for OSA screening, in which a positive SBQ is followed by automated scoring, has good diagnostic properties, and agrees with manual scoring of HSAT.

In our study, automated HSAT scoring yielded slightly higher REI (1.6/hour) than manual scoring but showed excellent correlation (Pearson’s r = 0.93). Kristiansen et al. found similar results (1.1/hour higher, Pearson’s r = 0.96), Cachada et al. (Pearson’s r = 0.91) [18], and Labarca (Pearson’s r = 0.95). However, Cohen’s kappa showed moderate agreement (0.58) for OSA diagnosis and weak agreement (0.33) for severity [19]. In our study, interrater reliability was excellent for OSA presence (Cohen’s kappa = 0.80) and substantial for severity categorization (Cohen’s kappa = 0.77).

The slightly higher REI from automated scoring could misclassify some borderline cases as mild OSA. However, this is of limited clinical relevance, as asymptomatic mild OSA patients are usually not started on CPAP therapy unless they have other conditions [8], furthermore CPAP shows no cardiovascular benefits [20] and has a high treatment failure rate in this group [21]. Importantly, no significant cases were missed.

Automated HSAT scoring had a sensitivity of 0.97 and specificity of 0.82 for any OSA (REI ≥ 5), slightly higher than Kristiansen’s findings for the Nox T3 (sensitivity 0.93, specificity 0.71) [22]. Our statistics were affected by an outlier classified as severe OSA by automated scoring but normal by manual scoring due to a missing thoracic belt signal. This underscores the need for high-quality signals in all channels for reliable automated scoring in clinical practice.

The excellent characteristics of automated HSAT in our study, along with strong correlations in other studies, underscores the potential of automated scoring. Our findings and Peñacoba’s views suggest that automated HSAT scoring could be a viable solution in primary care, where time constraints and limited scoring experience exist [23].

Provided an adequate file sharing platform and the willingness of sleep laboratories and clinics to cooperate type 3 PG recordings made in primary practice for screening could be used for definite diagnosis of OSA. Over time, automated scoring of HSAT may enable primary care physicians to make diagnoses independently.

To the best of our knowledge this is the first time a two-stage screening for OSA in the family practice setting, using type 3 PG, was performed. Our two-stage model showed good agreement with manual scoring (Cohen’s kappa of 0.62) for any OSA (REI ≥ 5). It had a sensitivity of 0.64, specificity of 0.97, and accuracy of 0.81, outperforming the Slovenian SBQ questionnaire alone (sensitivity 0.65, specificity 0.87). For moderate to severe OSA (REI ≥ 15), crucial for identifying those who benefit most from treatment, it achieved a kappa of 0.73, sensitivity of 0.73, specificity of 0.97, and accuracy of 0.92.

Chai-Coetzar’s two-stage screening using a questionnaire and type 4 polygraphy (pulse oximetry with desaturation detection) showed 88% sensitivity and 82% specificity for severe OSA (AHI ≥ 30) [24]. Gurubhagavatula et al. tested various two-stage models in patients with arterial hypertension, finding them effective for severe OSA but less so for milder forms [25]. Unlike type 3 PG that we used, type 4 PG, used in these studies, does not monitor airflow and chest movements, potentially underestimating REI as it cannot detect apneas without desaturation or arousals and is therefore less accurate [26].

If we had referred patients based on our two-stage model instead of the SBQ alone, referrals would have decreased by 15.3%. Whilst not directly comparable Peñacoba found that type 4 PG reduced referrals by 55.1% compared to a four-question screening questionnaire [27]. Our model missed one OSA patient (1.7%) compared to the SBQ alone. This patient was positive on the SBQ, negative with automated scoring, and had an REI of 5.5 on manual scoring of HSAT.

Whilst opinions on whether to screen, whom to screen and how to screen remain divided, the authors agree with Miller [28] who believes that family medicine clinics are the ideal place for OSA screening, where it could take place as part of regular screening of high-risk patients.

Our study had several limitations. First, we restricted participant age to 70 years to enhance engagement and comprehension of instructions, potentially affecting the sample’s representativeness. Compliance with CPAP therapy decreases in patients over 65[29] so these patients would potentially benefit less in practice.

The COVID-19 pandemic extended our recruitment period from one to four years, including a two-year hiatus, slowing recruitment upon resumption. Limited funding for patient recruitment also posed a challenge.

Using type 3 PG with manual scoring instead of type 1 PSG was another limitation. However, this aligned with standard care and available resources, as type 3 PG is common in clinical practice. We excluded patients taking sedatives, opioids, or tranquilizers, and those with heart failure, neuromuscular disease, or COPD stage D, as these conditions are better evaluated with type 1 PSG according to AASM guidelines.

Using type 3 PG with manual scoring instead of type 1 PSG was a limitation. One reason is that HSAT cannot distinguish between central and obstructive sleep apnea. However, this approach aligns with standard care and available resources, as type 3 PG is common in clinical practice. Given that central sleep apnea is present in only 0.4% of the general population [30] and we excluded patients taking sedatives, opioids, or tranquilizers, as well as those with heart failure, neuromuscular disease, or COPD stage D, where central apnea is more common, we can be confident in the validity of the method used.