Discussion

While the literature search did not identify publications that directly compared the performance of clinical prediction algorithms to history and physical exam, it did identify forty-eight validation studies that compared the accuracy of clinical tools, questionnaires or prediction algorithms against PSG or HSAT. Our recommendations are therefore based on these validation studies, which are described. Relevant outcomes data from these validation studies are summarized in the supplemental material, Table S2 through Table S36. Due to the uncertainty regarding clinical outcomes for patients misclassified by the prediction rules, the TF was unable to establish a cutoff for number of misclassified patients that would be considered acceptable. Nevertheless, all the clinical prediction models evaluated resulted in upper ranges of predicted false negatives per 1,000 patients that exceeded 100, a number that was determined by the TF to be clearly excessive for a stand-alone diagnostic test for OSA. In summary, the diagnostic performance of clinical questionnaires, morphometric models, and clinical prediction rules that consider multiple variables including symptoms, exam findings and subject characteristics, have all been evaluated against PSG and/ or HSAT. Overall, while sensitivity appears to be in the higher range it is not sufficient to adequately exclude the possibility of OSA. Specificity tends to be lower, resulting in a higher number of false positives that further limit the utility of these clinical or morphometric rules and models in the diagnosis of OSA. It should also be noted that some of these studies were conducted in focused populations (e.g., commercial drivers, elderly, bariatric surgery patients, etc.), thus limiting generalizability. The following discussion has been organized to review the data by questionnaire or clinical prediction rule.

BERLIN QUESTIONNAIRE: The Berlin Questionnaire consists of eleven questions divided into three categories to classify the patient as high or low risk for OSA.³³ Our review identified nineteen studies that evaluated the performance of the Berlin Questionnaire against PSG in the identification of patients with OSA.^34–52 The studies were conducted in a wide variety of geographic locations including Brazil,³⁸ Canada,^34,42 Greece,³⁷ Iran,³⁶ Korea,⁴⁰ Turkey,⁴³ and the United States.^41,44 Various patient populations were considered, including those in primary care clinics, sleep clinics, the veteran population, and patients with cardiac disease. The patients included in these studies were mostly men (50% or greater in the majority of studies) with suspected OSA; they were overweight or obese, and middle-aged. Overall, the Berlin Questionnaire produced a large number of false negative results, thereby limiting its utility as an instrument to diagnose patients with OSA. Specifically, when assessing the performance of the Berlin Questionnaire in identification of subjects with an AHI cutoff of ≥ 5, the pooled sensitivity was 0.76 (95% CI: 0.72 to 0.80), while the pooled specificity was 0.45 (95% CI: 0.34 to 0.56) (see supplemental material, Figure S1). Assuming a prevalence of 87% in a high-risk population, the result was an unacceptably high number of false negative results of 209 per 1,000 patients (95% CI: 174 to 244) (see supplemental material, Table S2). Furthermore, the questionnaire had suboptimal accuracy, ranging from 56% to 70%; accuracy became progressively more compromised with consideration of higher OSA severity thresholds (see supplemental material, Table S2 through Table S4 and Figure S1 through Figure S6).

Five studies evaluated the performance of the Berlin Questionnaire against HSATs.^53–57 When using an AHI cutoff of ≥ 15, the pooled estimate for sensitivity was 0.76 (95% CI: 0.64 to 0.85); specificity was 0.44 (95% CI: 0.30 to 0.58) and accuracy was 67%. When using an AHI cutoff of ≥ 5 and assuming a prevalence of 87% in a high-risk population, the number of false negative results was 531 per 1,000 patients (95% CI: 357 to 679) (see supplemental material, Figure S5 through Figure S7, Table S5 and Table S6).

The quality of evidence for the use of the Berlin Questionnaire was low after being downgraded due to either heterogeneity, indirectness, or imprecision.

EPWORTH SLEEPINESS SCALE: The Epworth Sleepiness Scale (ESS) is a self-reported questionnaire involving eight questions to assess the propensity for daytime sleepiness or dozing.⁵⁸ Our review identified seven studies that evaluated the performance of the ESS against PSG in the identification of patients with OSA. These studies were conducted in China, Brazil, Croatia, Turkey and the United States, thus reflecting a wide geographic sampling.^{38,43,50,51,59–61} Participants were those suspected of OSA and included mainly male, middle-aged and overweight or obese individuals. The overall results indicate that the ESS had a large number of false negative results limiting its utility for the diagnosis of OSA. When considering an AHI of ≥ 5, the ESS revealed a range of sensitivity of 0.27–0.72 and specificity of 0.50–0.76 (see supplemental material, Table S8). The ESS demonstrated an accuracy ranging from 51% to 59% for the AHI ≥ 5 cutoff. Therefore, the ESS had a high number of false negative results (range of 244 to 635 per 1,000 patients; assuming a prevalence of 87%). When considering a cutoff of AHI ≥ 15 and assuming a prevalence of 64% in high-risk patients, the number of false negative results increased and ranged from 269 to 506 per 1,000 patients (see supplemental material, Table S9). Findings from one study, comparing the performance of the ESS against HSATs for identification of patients with OSA, showed low sensitivity of 0.36 (95% CI: 0.19 to 0.57) and higher specificity of 0.77 (95% CI: 0.66 to 0.86)⁵⁷ (see supplemental material, Table S11).

The quality of evidence for the use of the ESS ranged from low to high across different AHI cutoffs after being downgraded to due to heterogeneity, indirectness, or imprecision The TF determined that the overall quality of evidence across AHI cutoffs was low.

STOP-BANG QUESTIONNAIRE: The STOP-BANG questionnaire is an OSA screening tool consisting of four yes/no questions and four clinical attributes.⁶² We identified ten studies, involving primarily middle-aged, obese males suspected of OSA that evaluated the performance of STOP-BANG questionnaire against PSG in the identification patients with OSA.^{42,49–52,63–67} The overall findings reveal that the STOP-BANG questionnaire had high sensitivity, but low specificity for the detection of OSA. These findings became more pronounced when higher levels of OSA category cutoffs were considered. The number of potential false negative diagnostic results limits use of the STOP-BANG as an instrument to diagnose individual patients with OSA. Specifically, when considering an AHI ≥ 5 and assuming a prevalence of 87% in high-risk patients, the sensitivity in the studies was 0.93 (95% CI: 0.90 to 0.95), but specificity was 0.36 (95% CI: 0.29 to 0.44) with a range of accuracy of 52 to 53%. The number of false negatives when compared against PSG was 61 per 1,000 patients (95% CI: 43 to 87), assuming a prevalence of 87% (see supplemental material, Figure S8 and Table S12). The sensitivity further improved and specificity was further compromised when progressively higher level of AHI cutoffs were considered (see supplemental material, Figure S10 through Figure S12, Table S13 and Table S14). The sensitivity and specificity of the STOP-BANG was similar when compared against HSAT^55,68 (see supplemental material, Table S15 through Table S17), or against PSG or HSAT^62,68 (see supplemental material, Tables S18 through Table S20).

The quality of evidence for the use of the STOP-BANG questionnaire ranged from low to high across different AHI cutoffs was after being downgraded due to either indirectness, inconsistency, or imprecision. The TF determined that the overall quality of evidence across AHI cutoffs was moderate.

STOP QUESTIONNAIRE: Our review identified five studies that evaluated the diagnostic performance of the STOP questionnaire against PSG.^{49–51,67,69} The STOP questionnaire showed moderate to high sensitivity, low specificity, and moderate accuracy (see supplemental material, Figure S14 and Table S21 through Table S23). When considering an AHI ≥ 5, the sensitivity was 0.88 (95% CI: 0.77 to 0.94), the specificity was 0.33 (95% CI: 0.18 to 0.52), and the accuracy in a high-risk population ranged from 74% to 86%. Assuming a prevalence of 87%, the number of false negatives was 104 per 1,000 patients (95% CI: 52 to 200) (see supplemental material, Table S21). When considering an AHI cutoff of ≥ 15, the sensitivity ranged from 0.62–0.98, the specificity ranged from 0.10–0.63, and the accuracy in a high-risk population ranged from 60% to 79%. Assuming a prevalence of 64% in a high-risk population, the number of false negatives ranged from 13 to 243 per 1,000 patients (see supplemental material, Table S22). When considering an AHI cutoff of ≥ 30, the sensitivity ranged from 0.91–0.97, the specificity ranged from 0.11–0.36, and the accuracy in a high-risk population ranged from 48% to 49%. Assuming a prevalence of 36% in a high-risk population, the number of false negatives ranged from 11 to 32 per 1,000 patients (see supplemental material, Table S23).

The quality of evidence for the use of the STOP questionnaire ranged from low to moderate across different diagnostic cutoffs and risk groups after being downgraded due to heterogeneity or imprecision. The TF determined that the overall quality of evidence across AHI cutoffs was low.

MORPHOMETRIC MODELS: Our review identified two studies that used morphometric models to predict OSA that was confirmed using sleep study data.^70,71 In a group of hypertensive patients, a multivariable apnea prediction score that combined symptoms, body mass index, age and sex was used to assess OSA risk.⁷⁰ In another study involving primarily middle-aged males, those with OSA were compared to those without OSA by using a morphometric clinical prediction formula incorporating measures of craniofacial anatomy (e.g., palatal height, maxillary and mandibular intermolar distances).⁷¹ While these studies demonstrate relatively high sensitivity (range of 0.88– 0.98) to predict AHI ≥ 5, the specificity was quite low (range of 0.11–0.31) (see supplemental material, Table S24). When considering adjusted neck circumference in both the hypertensive and chronic kidney disease populations, there are similar findings of relatively high sensitivity, but poor specificity, with improvements in specificity using higher AHI cutoffs^55,70 (see supplemental material, Table S25 and Table S26).

The quality of evidence for the use of morphometric models and adjusted neck circumference was moderate after being downgraded due to imprecision.

MULTIVARIABLE APNEA PREDICTION QUESTIONNAIRE: The performance of the Multivariable Apnea Prediction (MAP) questionnaire has been evaluated against PSG in those with suspected OSA,^35,72–74 a sample of hypertensive patients,⁷⁰ and also a sample of older adults⁷⁵ with findings of lower levels of specificity and high numbers of false positive results (see supplemental material, Table S27 and Table S28).

The quality of evidence for the use of the MAP questionnaire was judged moderate; it was downgraded due to imprecision.

CLINICAL PREDICTION MODELS: Four studies evaluated the performance of clinical prediction models against PSG,^61,76–78 and three studies^75,79,80 evaluated these models against HSAT. Two of the studies compared respiratory parameters against PSG: a study involving a Chinese cohort that evaluated snoring while sitting⁷⁶ and another single study assessing respiratory conductance and oximetry.⁷⁸ Results demonstrated a sensitivity ranging from 0.33–0.90, and a specificity ranging from 0.50– 1.00 using an AHI cutoff of ≥ 5. Other studies compared clinical prediction rules including age, waist circumference, ESS score and minimum oxygen saturation, and another evaluated gender, nocturnal choking, snoring and body mass index against PSG; these reported reasonably high sensitivity (range of 0.72–0.94) and specificity (range of 0.75–0.91) considering different AHI thresholds.^61,77 Clinical prediction rules have been evaluated against HSAT in select populations, i.e., the elderly,⁷⁵ bariatric surgery candidates,⁷⁹ and commercial drivers.⁸⁰ These studies reported sensitivities ranging from 0.76–0.97 and specificities ranging from 0.19–0.75 using an AHI cutoff of ≥ 30^75,79,80 (see supplemental material, Table S29 through Table S31).

The quality of evidence for the use of clinical prediction models ranged from moderate to high across the different AHI cutoffs after being downgraded due to imprecision. The TF determined that the overall quality of evidence across AHI cutoffs was moderate.

OTHER OSA PREDICTION TOOLS: Our literature review identified other OSA prediction tools, including the OSA50, the clinical decision support system, the OSAS score, and the Kushida Index. The OSA50 questionnaire involves four components including age ≥ 50, snoring, witnessed apneas and waist circumference.⁸¹ A study involving Turkish bus drivers⁸² and a validation study for the OSA50 in the primary care setting⁸¹ showed a sensitivity ranging from 0.49–0.98 and a specificity of 0.82 in both studies (see supplemental material, Table S32 and Table S33).

The performance of a hand-held clinical decision support system (assessing sleep behavior, breathing during sleep and daytime functioning) against PSG was evaluated in a study of veterans with ischemic heart disease. The system showed a high sensitivity of 0.98 (95% CI: 0.92 to 1.00) and a high specificity of 0.87 (95% CI: 0.66 to 0.97)⁴¹ (see supplemental material, Table S34).

The OSAS score involves assessment of the Friedman tongue position, tonsil size, and body mass index. In a sample of individuals suspected to have OSA, the sensitivity of the OSAS score was 0.86 (95% CI: 0.80 to 0.91) against PSG at an AHI > 5 cut = off, however; specificity was lower at 0.47 (95% CI: 0.34 to 0.56) with a high number of false positives in the low-risk group³⁹ (see supplemental material, Table S35).

One study evaluating the performance of the Kushida Index against PSG showed a high sensitivity of 0.98 (95% CI: 0.95 to 0.99) and high specificity of 1.00 (95% CI: 0.92 to 1.00) to detect AHI ≥ 5 (see supplemental material, Table S36).

The quality of the evidence for other prediction tools ranged from low to high across different tools, diagnostic cutoffs, and risk groups after being downgraded due to imprecision and indirectness.

OVERALL QUALITY OF EVIDENCE: The quality of evidence for specific clinical tools, questionnaires and prediction algorithms ranged from very low to high after being downgraded due to imprecision, indirectness, and heterogeneity. However, due to the heterogeneity of the tools, questionnaires and prediction algorithms described above combined with the low likelihood that future research would result in a change of the accuracy of these tools, the TF determined that the overall quality of evidence for the recommendation against using clinical tools, questionnaires or predictive tools was moderate.

BENEFITS VERSUS HARMS: These clinical tools, questionnaires and prediction algorithms carry the risk of not capturing the diagnosis of OSA when indeed OSA is present. Given the downstream effects of false negative diagnostic results, this would translate into high levels of OSA-related decrements in QOL, morbidity, and mortality due to undiagnosed and untreated OSA. On the other hand, false positive results would result in unnecessary testing and treatment for sleep apnea. Therefore, the TF determined that the potential harms outweigh the potential benefits of using clinical tools, questionnaires and prediction algorithms alone to diagnose OSA.

PATIENTS' VALUES AND PREFERENCES: Evaluation with clinical tools, questionnaires or prediction algorithms may be less burdensome to the patient and physician, when compared to HSAT or PSG. However, this must be weighed against their low levels of accuracy and the likelihood of misdiagnosis. In contrast, PSG and HSAT require more resources and create more burden for the patient and physician; however, they provide greater value in terms of higher diagnostic accuracy and therefore a higher likelihood that patients will receive appropriate treatment. Based on its clinical judgment, the TF determined that the vast majority of patients would not favor the use of clinical questionnaires or prediction tools alone for the diagnosis of OSA.

Discussion

The formulation of these recommendation statements was guided by evidence from twenty-six validation studies that evaluated the diagnostic accuracy of HSAT against PSG,^{35,53,62,67,81,86–106} as well as seven RCTs that compared clinical outcomes from management pathways.^{83–85,107–110} Four of these RCTs were determined to be most relevant to clinical practice, as they did not require oximetry testing as a criterion for inclusion and used conventional methods for determination of PAP pressures (i.e., APAP or attended titration).^83–85,110 This subset of studies will be referred to as “RCTs most generalizable to clinical practice” for the remainder of this discussion section.

ACCURACY: The following paragraphs are organized by type of HSAT device and components or combinations of components, as described in the literature.

A total of twenty-six validation studies were identified that reported accuracy outcomes. The data from these validation studies are summarized in the supplemental material, Table S37 through Table S58. In two studies that evaluated the performance of Type 2 HSAT devices against PSG,^67,86 when using an AHI ≥ 5 cutoff, accuracy in a high-risk population (assuming a prevalence of 87%) ranged from 84% to 91%. Using a cutoff of AHI ≥ 15, the accuracy of these devices was 88% in a high-risk group (see supplemental material, Table S37 and Table S38).^67,86 Seven studies evaluated the performance of Type 3 HSAT devices against PSG, but the AHI cutoffs employed varied across studies, resulting in sub-grouping by AHI cutoffs for our analyses.^87–93 When using an AHI ≥ 5 cutoff, accuracy in a high-risk population (assuming a prevalence of 87%) ranged from 84% to 91%, whereas in a low-risk population (assuming a prevalence of 55%) accuracy ranged from 70% to 78% based on the seven studies (see supplemental material, Table S39). Using a cutoff of AHI ≥ 15, the accuracy of these devices in a high-risk population ranged from 65% to 91%, based on six studies^{87,89–92,94} (see supplemental material, Table S40). Using a cutoff of AHI ≥ 30, the accuracy of the devices in the high-risk population was 88% (95% CI: 81% to 94%), based on five studies (see supplemental material, Table S41).

Five studies evaluated the performance of 2–3 channel HSAT devices against PSG. In a high-risk population using cutoffs of AHI ≥ 5,^95–97 AHI ≥ 15,^95–99 and AHI ≥ 30,^96,97 accuracy ranged from 81% to 93%, 72% to 87%, and 71% to 90%, respectively. Using the same cutoffs in a low-risk population, accuracy ranged from 77% to 88%, 68% to 95%, and 88% to 91%, respectively (see supplemental material, Table S42 through Table S44). When the performance of 2–3 channel HSAT was evaluated against unattended in-home PSG, using a cutoff of AHI ≥ 15, accuracy in a high-risk population was 86% (95% CI: 76% to 93%);⁵³ using a cutoff of AHI ≥ 30, accuracy ranged from 83% to 91% (see supplemental material, Table S45 and Table S46).^53,81

Six studies evaluated the performance of single channel HSAT against attended or unattended PSG (see supplemental material, Table S47 through Table S50, and Table S51 through Table S53, respectively).^{73,100–103,111}

A single study evaluated the performance of oximetry against unattended in-home PSG.¹¹¹ Using a cutoff of AHI ≥ 5, accuracy was 73% (95% CI: 68 to 78%) in a high-risk population, and 79% (95% CI: 74 to 84%) in a low-risk population. Using oximetry to identify OSA at an AHI ≥ 5 cutoff, and assuming a prevalence of 87% in a high-risk population, the findings of the study¹¹¹ would result in an estimated average of 274 misdiagnosed patients out of 1,000 tested, and 210 misdiagnosed patients out of 1,000 tested in a low-risk group (assuming a prevalence of 55%). Using a cutoff of AHI ≥ 15 and AHI ≥ 30, oximetry has an accuracy of 86% (95% CI: 83 to 91%) and 74% (95% CI: 71 to 76%) in a high-risk population, and an accuracy of 80% (95% CI: 75 to 84%) and 63% (95% CI: 59 to 67%) in a low-risk population, respectively (see supplemental material, Table S51 through Table S53).

A single study evaluated the performance of PAT, oximetry, and actigraphy against simultaneous unattended in-home PSG and reported a sensitivity of 0.88 (95% CI: 0.47 to 1.00), specificity of 0.87 (95% CI: 0.66 to 0.97) and accuracy of 88% (95% CI: 50 to 100%) in high-risk patients using a cutoff of AHI ≥ 5.¹⁰⁴ These findings would result in 121 misdiagnosed patients out of 1,000 tested in a high-risk population (based on a prevalence of 87%), and 125 misdiagnosed patients out of 1,000 tested in a low-risk population (based on a prevalence of 55%) (see supplemental material, Table S54).¹⁰⁴ Two cross-over studies randomized patients to home-based PAT, and in-laboratory simultaneous PSG and PAT.^105,106 For comparison to in-laboratory PSG, only the home-based PAT data were used for this recommendation. A single study that evaluated the performance of the PAT device in the home against in-laboratory PSG using a cutoff of AHI ≥ 5,¹⁰⁶ reported a specificity of 0.43 (95% CI: 0.22 to 0.66). When two studies evaluated the home-based PAT device against in-laboratory PSG at an AHI cutoff of ≥ 15, specificity ranged from 0.77 to 1.00 and sensitivity ranged from 0.92 to 0.96.^105,106 A single study evaluated the PAT device at an AHI cutoff of ≥ 30, and reported a specificity of 0.82 (95% CI: 0.57 to 0.96) and sensitivity of 0.92 (95% CI: 0.62 to 1.00)¹⁰⁵ (see supplemental material, Table S55 through Table S57).

The quality of evidence for diagnostic accuracy was downgraded due to indirectness, imprecision, or inconsistency. The quality ranged from low to high based on different tools and algorithms, diagnostic cutoffs, and risk groups.

The potential consequences for patients classified in true and false positive or negative categories are summarized in the supplemental material, Table S1. The TF concluded that the numbers of patients potentially misclassified by HSAT was high enough to be of clinical concern, particularly when tests were inconclusive or negative. In a population that has increased risk of moderate to severe OSA, both the increased likelihood of false negatives and the significant impact of missed diagnoses on patient outcomes can cause significant harm. This reasoning supports required use of a diagnostic test with higher sensitivity (PSG) in this population if HSAT provides a negative or non-diagnostic result.

CLINICAL OUTCOMES ASSESSMENT: The TF concluded that evaluating the impact of diagnostic accuracy on clinical outcomes is complicated by a number of factors that can cause discordance between tests, including night-to-night variability and inconsistent definitions of respiratory events (e.g., hypopneas) between HSAT and PSG. In addition, there is uncertainty regarding clinical outcomes for patients misclassified by HSAT.

For these reasons, studies that compared clinical outcomes in patients randomized to management pathways based on PSG or HSAT diagnostic assessment, within the same research protocol, provide the best opportunity to assess the acceptability of clinical outcomes using HSAT.

SUBJECTIVE SLEEPINESS: A meta-analysis of seven RCTs compared changes in patient-reported sleepiness, using the ESS, in patients diagnosed by HSAT or PSG, followed by PAP titration (see supplemental material, Figure S18).^{83–85,107–110} The meta-analysis showed a clinically and statistically insignificant difference of 0.38 points (95% CI: −1.07 to 0.32 points) greater improvement in patients randomized to the HSAT pathway versus the attended PSG pathway. This difference indicates that subjective sleepiness is similarly improved in patients who initiate PAP treatment based on diagnosis using either HSAT or PSG. The quality of evidence for subjective sleepiness was high.

QUALITY OF LIFE: Six RCTs, using various validated instruments (i.e., FOSQ, SAQLI, and SF-36), compared QOL in patients diagnosed by HSAT or PSG, followed by PAP titration.^{84,85,107–110} Meta-analysis demonstrated differences in pooled effects between pathways that were not significant (see supplemental material, Figure S19 through Figure S23, and Table S58). The quality of evidence ranged from moderate to high based on the measure used to assess QOL. The quality of evidence for the SF-36 physical and mental summary scores was downgraded due to imprecision. The TF considered the overall quality of evidence for QOL to be high as FOSQ and SAQLI measures of QOL were considered more critical for decision-making than the SF-36 measures.

CPAP ADHERENCE: Six RCTs evaluated CPAP adherence (mean hours of use per night); meta-analysis found no significant difference between the two assessment pathways (see supplemental material, Figure S24).^{83–85,108–110} When determining adherence by number of nights with greater than 4 hours of use, meta-analysis of five RCTs found a clinically insignificant trend towards increased CPAP adherence in the HSAT arm versus the PSG arm (see supplemental material, Figure S25).^{83–85,107,110} The quality of evidence for CPAP adherence was moderate to high across different AHI cutoffs after being downgraded due to imprecision. The TF determined that the overall quality of evidence across AHI cutoffs was high.

FAILURE TO COMPLETE DIAGNOSTIC ALGORITHM: Among the four RCTs most generalizable to clinical practice, three studies^83–85 required use of PSG if HSAT was inconclusive (did not provide adequate data or showed a low AHI after 1 or 2 unsuccessful attempts) and after 1 or 2 failed APAP trials (e.g., insufficient use, elevated residual AHI, persistent large leak). Based on data reported by a multicenter RCT there was concern regarding risk of non-completion of diagnostic testing when initial HSAT did not provide a definitive result. Rosen et al. 2012⁸⁴ reported that 30% (10/33) of subjects with technically inadequate HSATs and 16% (14/88) of subjects with low AHI on HSAT failed to proceed per protocol to PSG. There was also evidence indicating reduced effectiveness of repeated HSAT attempts for technical failures: 82% (147/180) of initial HSAT attempts were technically acceptable, whereas only 60% (12/20) of second attempts resulted in a technically acceptable study. Although failure to complete the diagnostic algorithm was not originally considered a critical outcome, the TF ultimately determined that it was critical for decisions regarding follow-up for inconclusive HSAT attempts. The quality of evidence regarding performance of PSG after a single inconclusive HSAT (versus multiple attempts) was low.

OVERALL QUALITY OF EVIDENCE: The TF determined that the critical outcome for diagnostic accuracy assessment was the number of false negative results. The quality of evidence for accuracy was downgraded to moderate due to imprecision, inconsistency, or indirectness. The quality of evidence for the clinical outcomes of sleepiness, quality of life, and CPAP adherence was high. Depression and cardiovascular outcomes were also considered critical outcomes; however, evidence for these outcomes was not available. Therefore, the overall quality of evidence for recommendation 2 is moderate.

In addition to accuracy and clinical outcomes, the TF determined that failure to complete the diagnostic algorithm was a critical outcome for repeat testing after a negative, inconclusive or technically inadequate HSAT. The quality of evidence for performing PSG after a single inconclusive HSAT was determined to be low, as only one study addressed this outcome. Therefore, the overall quality of evidence for recommendation 3 is low.

RESOURCE USE: Though a single night of HSAT is less resource-intensive than a single night of PSG, the relative cost-effectiveness of management pathways that incorporate each of these diagnostic strategies is unclear. Economic analyses have compared the cost-effectiveness of management pathways that incorporate diagnostic strategies using HSAT or PSG.^112–114 All have concluded that PSG is the preferred diagnostic strategy from an economic perspective for adults suspected to have moderate to severe OSA. An important factor in these analyses is the favorable cost-effectiveness of OSA treatment in patients with moderate to severe OSA, particularly when longer time horizons are considered. As a result, diagnostic strategies that lead to increased false negatives, and leave patients untreated, or increase false positives, and unnecessarily treat patients, have less favorable cost-effectiveness. It is important to note that these economic analyses are susceptible to error because of imprecision in modelling of management pathways and limitations in the quality of data available to estimate parameters. The impact of errors can be magnified when extrapolated over long time horizons.

Relative cost-effectiveness of management pathways that use HSAT or PSG for diagnosis can be assessed in the context of a RCT, if resource utilization is measured. Among the four RCTs most generalizable to clinical practice,^83–85,110 only one provided this information.⁸⁴ The study reported that in-trial costs were 25% less in the home-arm than the in-laboratory-arm.⁸⁴ These estimates were based on the Medicare Fee Schedule for the various study procedures, including office visits and diagnostic testing, and take into account the need to repeat studies.⁸⁴ A subsequent cost minimization analysis of this RCT also considered costs from a provider perspective.¹¹⁵ While provider costs (capital, labor, overhead) were generally less for the home program, this was not true for all modelled scenarios. The provider perspective highlighted the large number of cost components necessary to ensure high quality home-based OSA management, which narrowed the cost difference relative to lab management.

The available studies indicate that the potential cost advantages of HSAT over PSG are not as high as reflected by the cost difference of a single night of testing. Even when HSAT is used in appropriate populations and conditions, additional HSAT and PSG are needed for patients with technically inadequate or inconclusive studies, in order to achieve an accurate diagnosis. In addition, if a home management pathway is used in a manner that results in reduced effectiveness relative to PSG, use of HSAT could in fact be less cost effective than using PSG. Examples of this include use in patient populations with predominantly mild OSA in which there are a higher proportion of negative or indeterminate HSAT results that require follow-up PSG, or use in patients at risk for non-obstructive sleep-related breathing disorders that may not be accurately diagnosed with HSAT. The TF determined that if HSAT is used in the recommended context and management pathway, it would be more cost-effective than if it is used outside this framework.

BENEFITS VERSUS HARMS: Use of HSAT may provide potential benefits to patients with suspected OSA. Such benefits could include convenience, comfort, increased access to testing, and decreased cost. HSAT can be performed in the home environment with fewer attached sensors during sleep. The availability of HSAT for diagnosis may improve access to diagnostic testing in resource-limited settings, or when the patient is unable to leave the home or healthcare setting for testing. In addition, HSAT may be less costly when used appropriately. These benefits must be weighed against the potential for harm. Harms could result from the need for additional diagnostic testing among patients with technically inadequate or inconclusive HSAT findings, or from misdiagnosis and subsequent inappropriate therapy or lack of therapy. As summarized above, the use of HSAT has not been demonstrated to provide inferior clinical benefit, compared to PSG when used in the appropriate context. Therefore, the TF determined that if HSAT is used in the context described in the recommendations and remarks, the risk of harm is minimized and the probability of potential economic benefits increased.

The TF was concerned that, in clinical practice (in contrast to the RCT setting) there would be higher levels of drop out from diagnostic testing, among patients with initial study attempts that did not result in diagnoses of OSA. In particular, there was concern that patients with false negative HSAT results may not complete additional testing after learning of a negative result, despite the presence of symptoms of OSA. In addition, as described above, HSAT is less accurate than PSG and more likely to result in false negative results. For these reasons, the TF recommends that if the initial HSAT shows a negative or inconclusive result, PSG, rather than a second HSAT, should be performed. There are similar concerns that, following a technically inadequate HSAT, repeat HSAT may be associated with a higher rate of technical failure on the second study, and with increased risk of drop out from the diagnostic process. Therefore, the TF also recommends that if the initial HSAT is technically inadequate, PSG rather than a second HSAT should be performed. On the other hand, the TF recognizes that there may be specific circumstances in which repeat HSAT is appropriate after an initial failed HSAT. These circumstances would include cases in which both of the following are present: the clinician determines that there is a high likelihood of successful recording on a second attempt, and the patient expresses a preference for this approach.

The TF recognizes that HSAT may have value to patients in some contexts beyond what is covered by these recommendations, but has limited the recommendations to apply to situations where there is sufficient evidence to guide evaluation of benefits versus harms.

PATIENTS' VALUES AND PREFERENCES: Individual patient preference for PSG or HSAT will differ depending on circumstances and values. In one of the four RCTs most generalizable to clinical practice, both HSAT and PSG were performed for each patient, and 76% preferred HSAT.¹¹⁰ This means that a significant percentage (24%) still preferred PSG. Unfortunately, there is insufficient data about diagnostic testing preferences in clinical practice, where preferences may differ from what is seen in the RCT setting. The availability of different options for diagnosis may increase satisfaction, if patient preferences are included in the process of choosing the diagnostic test type. If HSAT is used, the TF determined that patients would value accurate diagnosis, good clinical outcomes, and increased convenience. Based on their clinical judgment, the TF also determined that patients would prefer not having a repeat HSAT if the initial test result is negative, as repeated HSAT would be less likely to produce a definitive result and would unnecessarily inconvenience the patient. In this situation, proceeding directly to PSG, which has greater sensitivity to detect OSA, would be preferred by most patients. The TF also determined that most patients would prefer not to have a repeat HSAT if the initial test was technically inadequate, to avoid inconvenience, but that some patients may desire this option, in specific cases in which there was high likelihood of an adequate result with repeat testing.

SPECIAL CONSIDERATIONS: The following sections describe special considerations when using HSAT for the diagnosis of OSA. They provide additional support for, and explanation of the Remarks, and are based on specifications used by studies that support the recommendation statements.

CLINICAL POPULATION: A review of RCTs that met inclusion criteria indicated that the following criteria should be used to establish the presence of increased risk of moderate to severe OSA and to determine if HSAT use is reasonable: excessive daytime sleepiness occurring on most days, AND the presence of at least two of the following three criteria: habitual loud snoring; witnessed apnea or gasping or choking; or diagnosed hypertension. Among the four RCTs most generalizable to clinical practice, two of the four studies^83,84 required ESS > 12 as an entry criterion: One¹¹⁰ required at least two out of three criteria (i.e., sleepiness (ESS > 10), witnessed apnea, snoring) for participation; and one, which was performed in a Veteran's Administration population, did not specify any specific entry criteria besides suspected OSA (though the average ESS for participants was elevated at > 12 and 95% were men).⁸³ In the latter study, 9.9% of individuals in the PSG arm were found to have AHI < 5.⁸³ In addition to sleepiness, at least two studies in this subset had specific inclusion criteria such as snoring, witnessed apnea, gasping or choking at night, or hypertension.^83,85 One study incorporated neck circumference in the determination of high risk of OSA.⁸⁴

EXCLUDED PATIENT POPULATIONS: Three of the four RCTs most generalizable to clinical practice excluded patients with significant cardiopulmonary disease and other significant sleep disorders.^83,84,110 Two studies excluded patients taking opioids, having uncontrolled psychiatric disorder, neuromuscular disease, and patients with significant safety-related issues related to driving or work. Other notable exclusion criteria, specified by at least one of the studies, included lack of an appropriate living situation, pregnancy, and alcohol abuse. The single study that did not mention exclusion criteria noted that 3 of 148 individuals in the HSAT arm were diagnosed with CSA and 4 of 148 individuals required supplemental oxygen or bi-level PAP and exited the study.⁸⁵ In the PSG arm of the study, 6 of 148 individuals were diagnosed with CSA and 12 of 148 required supplemental oxygen or bi-level PAP. Studies outside the four RCTs most generalizable to clinical practice had similar inclusion/exclusion criteria.

Therefore, based on information from three of the four RCTs most generalizable to clinical practice that specified exclusion criteria, and for the reasons discussed above in Resource Use, Benefits and Harms, and Patients' Values and Preferences sections, the TF determined that HSAT should be used in an uncomplicated clinical population. This is defined as the absence of significant cardiopulmonary disease (e.g., heart failure, chronic obstructive pulmonary disease [COPD]), potential respiratory muscle weakness due to neuromuscular conditions, chronic opiate medication use, history of stroke, concern for a significant sleep disorder other than OSA (e.g., CSA, parasomnia, narcolepsy, severe insomnia), and environmental or personal factors that preclude the adequate acquisition and interpretation of data from HSAT.^83,84,110

FOLLOW-UP: Based on information from the four RCTs most generalizable to clinical practice,^83–85,110 the TF determined that HSAT should be used in the context of an OSA management pathway that incorporates a PAP therapy initiation protocol for APAP or PSG titration, early follow-up after initiation of therapy, and PSG titration studies for patients failing APAP therapy. All RCTs incorporated early follow-up of APAP titration (within 2–7 days after HSAT) by skilled technical staff.^83–85,110 As described above, the recommendation for using HSAT to diagnose OSA is based on clinically significant improvements in clinical outcomes. Therefore, the TF determined that HSAT should be used in the context of an OSA management pathway that incorporates a PAP therapy initiation protocol and early follow-up after initiation of therapy.

CLINICAL EXPERTISE: All four RCTs that were most generalizable to clinical practice administered HSAT at academic or tertiary sleep centers with highly skilled sleep medicine providers and technical staff.^83–85,110 HSAT recordings were reviewed by a sleep medicine specialist. One RCT that was not included in this subset (because an overnight oximetry was used as entry criteria) used a simplified nurse-led model of care involving nurse specialists experienced in management of sleep disorders (mean of 8.3 years of experience with CPAP therapy). Therefore, the TF determined that HSAT should be administered by an accredited sleep center under the supervision of a board-certified sleep medicine physician, or a physician who has completed a sleep fellowship, but is awaiting the next opportunity to take the board examination.

HOME SLEEP APNEA TESTING DEVICE: Among the four RCTs that were most generalizable to clinical practice, three used conventional Type 3 devices (nasal pressure, thoracic and abdominal excursion using RIP technology, oxygen saturation, EKG, body position, and oral thermistor in some cases),^84,85,110 and one used a 4-channel device⁸³ based on PAT with three additional channels (heart rate, pulse oximetry, and actigraphy). The TF determined that testing should be performed using these types of HSAT devices that have been demonstrated to be technically adequate. Additional guidance on technical specifications regarding HSAT is provided in The AASM Manual for the Scoring of Sleep and Associated Events.²⁴

RECORDING TIME: In the four RCTs most generalizable to clinical practice, the minimum requirement for an acceptable study was 4 hours of adequate flow and oximetry signals.^83–85,110 Whereas one HSAT study⁸³ used PAT as a surrogate of flow, two studies recorded nasal pressure flow^85,110 and one study recorded thermistor in addition to nasal pressure flow.⁸⁴ The latter three studies also recorded thoracic and abdominal movements.^84,85,110 All of these studies showed at least equivalence of adherence to PAP therapy and functional improvement in the home versus in-laboratory management pathways.^84,85,110 Therefore, the TF determined that a protocol requirement of a minimum of 4 hours of good quality data from HSAT recording, during the habitual sleep period, is warranted to diagnose OSA.

Additionally, nine non-RCT validation studies reported minimum requirements for duration of acceptable signal quality.35,53,54,81,86,88,93,96,116 The required signals and minimum durations included nasal pressure flow and oximetry for at least 3 hours^88,93,116 or 4 hours^53,81,86,96 and single-channel nasal airflow recording for a minimum of 3 hours³⁵ or only 2 hours.⁵⁴ The diagnostic accuracy of the cardiorespiratory devices compared against PSG for the detection of OSA at different AHI cutoff points was relatively high. One study reported a sensitivity and specificity of 0.88 and 0.84, respectively, for a HSAT AHI cutoff point of ≥ 9 events/h.⁵³ In a separate study, the sensitivity and specificity for unattended in-home PSG was 0.91 and 0.89 for an AHI cutoff of > 10 events/h, but 0.88 and 0.55, respectively for an AHI cutoff of > 5 events/h.⁸⁶ In another study, at an AHI cutoff of > 10 events/h, HSAT had a sensitivity of 0.87, and a specificity of 0.86.⁸⁸

Overall, the body of evidence investigating the minimum number of hours of adequate data on HSAT required to accurately diagnose OSA is very limited. There are no data to suggest that fewer than 4 hours of technically adequate recording compromises the accuracy of test results, and there is no direct evidence on the impact of a minimum number of recording hours of HSAT on clinical outcomes. Based on available indirect evidence, the TF weighed the “risk” of undergoing less than the required duration of good quality HSAT with resultant false negative (or false positive) results, against the “benefit” of potentially increasing the accuracy by performing PSG. Performing PSG in the scenario of a “positive” diagnosis of OSA is less likely to alter clinical decision-making and may, in fact result in unnecessary delays in care with increased cost. Conversely, a “negative” HSAT, in the scenario of a high pretest probability of OSA, will justify PSG even when the test is of adequate quality and duration. The TF believes that the goals of establishing an accurate diagnosis, while minimizing patient inconvenience and cost, align with patient preferences.

NIGHTS OF RECORDING TIME: The adequacy of a single night HSAT performed for the diagnosis of OSA in the context of an appropriate clinical population and management pathway is supported by published evidence. Our literature review only identified two studies relevant to the question of whether multiple nights of recording is superior to a single night.^35,73 These studies evaluated the performance of multiple nights (3) of single channel HSAT device (i.e., nasal pressure transducer or oximetry) to the first night of recording. Utilizing PSG as the reference, the studies found that recording over three consecutive nights may decrease the probability of insufficient data and marginally improve accuracy when compared against a single night of recording. However, the TF considered this evidence insufficient to establish the superiority of multiple-night HSAT protocol over a single-night HSAT protocol, as the studies only included a single channel recording and did not evaluate clinically meaningful outcomes or efficiency of care.

A single HSAT recording encompassing multiple nights may have potential advantages or drawbacks relative to only a single night of recording. For example, if multiple-night HSAT improved accuracy or resulted in fewer inconclusive or inadequate studies, patient outcomes or costs might improve. On the other hand, the potential for multiple-night recordings to increase cost and patient inconvenience must be considered. Insufficient evidence exists to support routine performance of more than a single night's recording for HSAT.

Discussion

Four studies examining the validity of HSAT for the diagnosis of OSA in patient populations with significant cardiorespiratory morbidity met our inclusion criteria.^117–120 No RCTs were identified that randomized patients with significant co-morbidities, as outlined above, to management pathways using either PSG or HSAT for diagnosis.

PATIENTS WITH COMORBID HEART FAILURE: Our review identified three studies that included patients with heart failure.^117,119,120 A study of 50 patients with stable heart failure (Class 2–4; left ventricular ejection fraction < 40%) evaluated the performance of home oximetry against PSG.¹¹⁷ Home oximetry was considered positive if the 2% ODI ≥ 10, and the PSG was considered positive if AHI ≥ 15 using a hypopnea criteria that did not require oxygen desaturation or arousal. Home oximetry data was not obtained in 3 patients and oximetry had to be repeated in 2 patients. This study found oximetry to have a sensitivity of 0.85 and specificity of 0.93 in identifying sleep-disordered breathing.¹¹⁷ The specificity was poor for identifying CSA, based on desaturation/ resaturation patterns (specificity of 0.17; sensitivity of 1.0) with 10 of 12 patients with OSA identified as having CSA. A study of 50 patients with heart failure (Class 3; LVEF ≤ 35%) evaluating the performance of an HSAT device that included ECG (2 leads), oximetry, and respiratory impedance sensors against PSG, was able to obtain valid data in 44 patients in the home setting. Sensitivity, specificity and accuracy at AHI ≥ 5 and AHI ≥ 15 cutoffs were 0.92, 0.52, and 0.73 and 0.67, 0.78, and 0.75 respectively.¹¹⁹ Unfortunately, the performance of the device in distinguishing central from obstructive events was not evaluated.

A study of 100 patients with stable heart failure (mean LVEF ± SD: 34.6% ± 11) evaluated the performance of simultaneous 2-channel HSAT device (nasal pressure flow and oximetry) against unattended in-home PSG.¹²⁰ In the 90 patients with valid HSAT recordings, the sensitivity and specificity was 0.98 and 0.60, respectively, using an AHI ≥ 5 cut off (hypopneas required 4% oxygen desaturation for both HSAT and PSG), and 0.93 and 0.92% using an AHI ≥ 15 cutoff. Among these patients, 29% had CSA, 19% had OSA, and 13% had both, based on PSG. The type of sleep apnea could not be determined using the HSAT device. Meta-analysis of these studies (see supplemental material, Table S61) found that in a population of 1,000 patients at high risk of moderate to severe OSA (64% prevalence), 45 to 230 more false negative and 18 to 79 more false positives would result from the use of HSAT.^117,119,120 The quality of evidence for was downgraded to low due to imprecision and indirectness.

PATIENTS WITH COMORBID COPD: Only one study addressed the validity of HSAT (nasal pressure, respiratory excursion (piezoelectric sensor), body position and pulse oximetry) in patients with COPD.¹¹⁸ Of 72 patients with stable COPD (GOLD stage II and III) and symptoms of OSA, only 26 patients (36%) had HSAT studies of reasonable quality.¹¹⁸ When comparing HSAT to PSG, the intraclass correlation coefficient was 0.47 (accuracy not provided).¹¹⁸ Data regarding detection of hypoventilation was not provided. Evidence was downgraded to very low based on imprecision, indirectness, and risk of bias due to significant data loss.

PATIENTS WITH OTHER COMORBIDITIES: No studies were identified that met our inclusion criteria that specifically evaluated the use of HSAT for diagnosis of OSA in patients with history of stroke, chronic opioid medication use, neuromuscular disease with respiratory muscle impairment, high risk of hypoventilation, or severe insomnia. Therefore, the TF concluded that HSAT has not been adequately validated or demonstrated to provide favorable clinical outcomes and efficient care in these patient populations.

OVERALL QUALITY OF EVIDENCE: The evidence for the use of HSAT in diagnosis of OSA among patients with comorbid heart failure was based on three studies, and this evidence was downgraded to low because of imprecision and indirectness.^117,119,120 The evidence for the use of HSAT in diagnosis of OSA among patients with COPD was based on a single, small study in which the majority of subjects had technically inadequate HSAT data due to recording failure. There was no direct evidence regarding suitability of HSAT for the diagnosis of OSA in patients with neuromuscular disease with respiratory impairment, hypoventilation, chronic opioid medication use, history of stroke, or severe insomnia. The overall quality of evidence for HSAT in patients with comorbid conditions was downgraded to very low due to imprecision, indirectness, and risk of bias.

BENEFITS VERSUS HARMS: Certain patient populations are at increased risk of having forms of SDB other than OSA (e.g., CSA, hypoventilation, and hypoxemia). These forms of SDB can cause significant morbidity and mortality if left untreated. HSAT has not been validated to diagnose some of these types of SDB (CSA, hypoventilation); therefore, the use of HSAT in populations at increased risk for SDB other than OSA increases the likelihood of not detecting these breathing disorders, which could lead to inadequate treatments, increased long-term healthcare costs, morbidity and mortality. In addition, the accuracy of HSAT has not been validated in patients with severe insomnia where it may be compromised leading to similar outcomes. Though the cost of diagnostic PSG is higher than HSAT, the TF determined that the benefits of increased accuracy, use of appropriate therapy, and improved clinical outcomes outweigh this factor. There are, however, instances where PSG cannot be performed for practical reasons (hospitalization, inability of patient to leave home setting or participate in PSG), and use of HSAT may be reasonable, as the alternative is to not addressing SDB at all.

PATIENTS' VALUES AND PREFERENCES: Based on clinical judgment, the TF determined that patients at increased risk for non-OSA SDB would want these breathing disorders to be adequately diagnosed and treated, as therapy of these disorders can result in significant improvement in health and well-being, and would therefore prefer PSG. Similarly, patients with severe insomnia needing evaluation of OSA would prefer PSG. If the optimal diagnostic test (PSG) was not feasible, then they would desire to have other diagnostic tests (i.e., HSAT) available that may aid their clinical provider in providing care for SDB.

Discussion

Our literature search yielded nine studies that met inclusion criteria.^{112,121–128} Three focused on the diagnostic accuracy of the initial portion of the PSG recording, against the accuracy of using the same full-night recording,^121,122,127 and a fourth study compared the accuracy of the diagnostic portion of a split-night study against a separate full-night study.¹²³ Three studies compared success of CPAP titration in those undergoing a split-night study against those undergoing a full-night sleep recording.^125,126,128 One study compared CPAP adherence in those who underwent split-night studies against those who had full-night studies.¹²⁴ A study that did not provide data suitable for inclusion in a meta-analysis examined cost-effectiveness of the split-night study versus the full-night study. Data from this study was considered in the evaluation of resource use.¹¹²

DIAGNOSTIC ACCURACY: Four studies that examined diagnostic accuracy and performance characteristics of a split-night protocol used the initial truncated PSG to serve as a representative surrogate of the initial diagnostic portion of a split-night study; the first 2–3 hours of the recording were compared to the full night of sleep recording.^{121–123,127} One study found that the 2-hour AHI and 3-hour AHI strongly correlated with the full-night AHI (concordance correlation coefficient = 0.93 and 0.97, respectively).¹²¹ This study reported a sensitivity of 0.80 (95% CI: 0.67 to 0.90) and specificity of 0.93 (95% CI: 0.83 to 0.98) using a cutoff of AHI ≥ 5, and a sensitivity of 0.77 (95% CI: 0.56 to 0.91) and specificity of 0.98 (95% CI: 0.92 to 1.00) using a cutoff of AHI ≥ 15 (see supplemental material, Table S62 and Table S63). When comparing 3 hours of recording versus the full-night recording, excellent consistency of the AHI was observed; there was no significant difference in the AHI derived from the first 3 hours of total sleep time versus the total sleep time (concordance correlation coefficient adjusted for REM and supine sleep of 0.96 and an accuracy of 93%),¹²¹ even in those with a milder degree of OSA (accuracy for AHI cutoffs of ≥ 5, ≥ 10 and ≥ 15 were 95, 97 and 99.5% respectively). One study assessed the diagnostic validity of a 2-hour recording and identified an optimal AHI cutoff of ≥ 30 events/h as providing the highest accuracy (90.9%).¹²² This study reported a specificity of 0.90 and a sensitivity of 0.92 (see supplemental material, Table S64). Another study showed an AHI Pearson correlation coefficient between a full-night study and the diagnostic portion of the split-night study of 0.63 when the split-night study recording time was ≥ 90 minutes.¹²³ Finally, a study that compared sleep and respiratory parameters during the first 3 hours of the night against the values recorded during the entire night did not find a significant difference in AHI.¹²⁷ Given the lack of definitive data, the TF elected not to designate a specific AHI threshold to inform the decision to initiate PAP titration during a split-night study protocol. The quality of evidence for diagnostic accuracy was downgraded to low due to indirectness, imprecision, and risk of bias.

CPAP OUTCOMES: Our literature review identified three studies that examined CPAP success in the split-night versus full-night CPAP titration recordings. One study, focused on upper airway resistance syndrome, found no difference in the success rates of CPAP titration, defined as a respiratory effort-related arousal (RERA) index < 5 on the final CPAP setting.¹²⁵ A cross-over study involving comparisons of split-night CPAP recordings versus full-night CPAP titration recordings in patients with OSA, showed no significant difference of the AHI, arousal index and the percentage sleep time with oxygen saturation below 90% while on CPAP, though the final CPAP pressure was lower at the end of the split-night titration (8.8 versus 10.3 cm H₂O).¹²⁶ One study reported no clinically significant difference in adherence to CPAP treatment in patients undergoing a split-night study (78.7%) versus a full-night study with follow-up titration (77.5%)¹²⁴ (see supplemental material, Figure S26, Figure S27, and Table S65). A meta-analysis of two studies (performed by the TF) comparing reduction of AHI after CPAP treatment with split-night PSG against full-night PSG found no clinically significant difference. The quality of evidence for CPAP outcomes was downgraded to low, due to imprecision associated with a limited number of studies and small sample size.

RESOURCE USE: A single cost-effectiveness analysis demonstrated that split-night studies were less costly than full-night studies based on cost per quality of life year (QALY) gained ($1,979 versus $2,092) and would be considered more cost-effective than full-night studies when third-party willingness to pay fell below $11,500 per QALY gained (a level of cost per QALY that would still be considered a good value for payers).¹¹² However, the TF had low confidence in the certainty of resource use, given the lack of high quality evidence to inform cost effectiveness.

OVERALL QUALITY OF EVIDENCE: The available studies were methodologically limited due to a number of issues: use of suboptimal study designs (not RCTs), use of the initial portion of a full-night PSG recording as a surrogate for the baseline portion of a split-night study,^121,127 and a lack of consistent use of standard monitoring (e.g., nasal pressure transducer).¹²¹ The overall quality of evidence was determined to be low due to a combination of imprecision, indirectness, and the risk of bias.

BENEFITS VERSUS HARMS: The split-night protocol, in comparison to a full-night baseline assessment followed by a separate PAP titration, has the potential to provide the needed diagnostic information and effective CPAP settings within the same recording. Potential disadvantages of the split-night study include insufficient diagnostic sampling (e.g., limited REM sleep time and limited supine time in those with difficulty initiating sleep), and insufficient time to ascertain appropriate CPAP treatment settings. Based on clinical judgment, the TF determined that there is low certainty that the benefits of a split-night study in comparison to full-night studies exceed the harms.

PATIENTS' VALUES AND PREFERENCES: When comparing the split-night study to the full-night study, existing data are consistent and demonstrate a high level of reproducibility of the standard AHI metric and effective identification of the optimal CPAP pressure. These data also suggest that the two approaches lead to similar follow-up CPAP adherence. Based on their clinical judgment, the TF members determined that the majority of well-informed patients would prefer a split-night protocol over a full-night protocol, when clinically appropriate and feasible (Figure 2), due to the lower cost, and the convenience of potentially completing a diagnostic and titration study during one sleep study. However, electing to use a split-night protocol still leaves the possibility that a patient will need to return for a second sleep study, if the diagnostic or titration portions of the split-night study are inconclusive.