A multicentre randomised controlled trial and economic evaluation of continuous positive airway pressure for the treatment of obstructive sleep apnoea syndrome in older people: PREDICT

Article history

The research reported in this issue of the journal was funded by the HTA programme as project number 08/56/02. The contractual start date was in August 2009. The draft report began editorial review in November 2013 and was accepted for publication in September 2014. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the reviewers for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

Professor John R Stradling was an acting consultant for ResMed (UK) Ltd from April 2013. Professor Mary J Morrell received research funding from ResMed (UK) Ltd as an unrestricted education grant that was awarded to collaborator Professor AS Simonds, Royal Brompton Hospital and Harefield NHS Foundation Trust.

Copyright statement

© Queen’s Printer and Controller of HMSO 2015. This work was produced by McMillan et al. under the terms of a commissioning contract issued by the Secretary of State for Health. This issue may be freely reproduced for the purposes of private research and study and extracts (or indeed, the full report) may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NIHR Journals Library, National Institute for Health Research, Evaluation, Trials and Studies Coordinating Centre, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

Overview of sleep apnoea

Obstructive sleep apnoea (OSA) is caused by occlusion of the pharyngeal airway during sleep that results in a pause in breathing (apnoea). Each apnoea event or partial occlusion (hypopnoea) is associated with hypoxaemia, and is usually terminated by a brief arousal from sleep and an acute surge in blood pressure (BP). 1 The subsequent sleep disruption leads to symptoms of excessive daytime sleepiness in some,2 but not all, people with OSA. 3 When OSA occurs with symptoms of excessive daytime sleepiness it is termed obstructive sleep apnoea syndrome (OSAS).

The long-term implications of severe4 OSAS are considerable in middle-aged people. Daytime sleepiness impairs function and increases accident risk,5,6 with OSAS patients being two to four times more likely to have road traffic accidents (RTAs) as a result of reduced alertness while driving. 7 OSAS patients are also more likely to experience mood changes8,9 and reduced quality of life,10,11 which is often attributed to reduced social functioning and vitality. 12 In addition, there is some evidence of reduced cognitive function,13–16 although the extent of the neurocognitive deficits in patients with OSA is currently debated. 17

The cardiovascular impact of OSAS has been established using epidemiological data to show that people with OSA have a threefold increased likelihood of developing hypertension over 4 years, independent of other risk factors. 18,19 In addition, treatment trials in patients with severe OSAS have produced a 2 mmHg to 3 mmHg reduction in BP. 20–22 Untreated severe OSAS may be associated with an increased risk of stroke,23,24 cardiovascular disease25–27 and death. 28–30 However, the close association between OSAS and obesity,31 as well as other disorders that predispose to vascular disease, makes it difficult to determine the risk factors associated with OSAS. 32 This is especially true in older people, who are more likely to have comorbidities.

Obstructive sleep apnoea syndrome in older people

The prevalence of OSAS was reported to be approximately 4% in males and 2% in females in a US cohort of 602 employed men and women (30–60 years). 33 However, more recent estimates from the same cohort predict that up to 14% of males and 5% of females have OSAS. 34,35 This represents a substantial increase since 1990, in part because of the increasing prevalence of obesity36 and the ageing population. Specifically, the prevalence of OSA (in the absence of daytime sleepiness) appears to increase with age, although there is some evidence to suggest that it plateaus or decreases in the population over the age of 65 years. 37 In a study that used similar criteria to define sleep apnoea in younger and older people, prevalence was eight times higher in community-dwelling older men (65–100 years) compared with 3% in a younger population (20–44 years). 38 Table 1 reviews in detail the prevalence of OSA and OSAS in older people. The wide variation in estimates is likely to reflect the definitions used to quantify the OSA or OSAS and the different health status of the older populations studied, for example relatively healthy community-dwelling individuals or nursing home residents with comorbidity.

Aetiology of obstructive sleep apnoea syndrome in older people

The high prevalence of OSA in older people has led to debate regarding its causes and the consequences of the disease in this population. 57–59 In middle-aged people, pharyngeal occlusion occurs as a result of a reduction of pharyngeal dilator muscle tone during sleep60 coupled with excessive extraluminal pressure around the airway, produced by excessive adipose tissue. 61,62 In susceptible individuals these factors lead to airway collapse during sleep. Therefore, it is perhaps not surprising that neck circumference is a significant risk factor for OSA. 63,64 However, in older people, additional factors, such as an age-related reduction in pharyngeal muscle function65,66 and structural changes to the upper airway, increase the vulnerability to collapse. 67 Specifically, a decrease in the size of the upper airway lumen in older people,68 associated with an age-related lengthening of the pharyngeal airway in women69 and a descent of the hyoid bone,70 creates a predisposition to airway collapse.

Symptoms of obstructive sleep apnoea in older people

Older people report different levels of sleepiness and, compared with younger populations, rate their health differently for the same level of OSA severity. 71 This may be because older people have become habituated to the reduction in sleep quality that occurs as part of the normal ageing process. 72–74 Hence, older people may not suffer symptoms of daytime sleepiness as a result of the further sleep disruption caused by OSA. Alternatively, increased daytime sleepiness may be less debilitating in older people, who have different family and work demands and may have more time for daytime naps. In addition, older populations are more likely to have comorbidities which may cause sleep disruption75 and polypharmacy contributing to excessive daytime sleepiness. 76 Specifically, nocturia may disturb sleep, and there is some suggestion that OSA exacerbates nocturia. 77,78 Taken together, these factors could modify daytime sleepiness and obscure the symptoms of OSA. Therefore, although excessive sleepiness (regardless of its cause) is associated with increased all-cause mortality in older people,79 the proportion of sleepiness that is a result of OSA in older people, and hence could be modified by treatment, is unknown.

Both the ageing process80 and OSA14,15,81 are associated with a reduction in cognitive function. However, few studies have investigated the impact of OSA on cognitive function in older people. In those studies that have measured cognitive function in older people, cognitive impairment appears to be independently related to both OSA severity and increasing age, but the coexistence of these factors does not further increase dysfunction. 57,82,83 One explanation for the preservation of cognitive function in OSA patients is that neural compensation can overcome the cognitive deficits that are associated with the effects of intermittent hypoxia and/or sleep deprivation on the brain. 84 Whether or not the capacity for neural compensation is decreased in older people, who have less neural reserve, is unknown. 57 Recent data have shown that poorer sleep quality is associated with factors that may accelerate cognitive decline in older people and this finding requires further investigation. 85

With respect to the cardiovascular impact of OSA in older people, there are limited studies on the long-term consequences. Prospective observational data over 8 years86 showed that severe OSA in older people is associated with cardiovascular mortality, as it is in middle-aged people. Specifically, the cardiovascular risk in older people with untreated OSAS resulted from increased stroke and heart failure deaths. 23,25,86 However, a potential survival bias in people who have survived into older age means that they may be different in some way from younger people with OSAS. Alternatively, studies in older people with OSA may be selecting those who have developed OSA later in life.

Treatment of obstructive sleep apnoea syndrome

The evidence-based treatment of choice for moderate to severe OSAS in middle-aged patients is continuous positive airway pressure (CPAP) therapy, which modifies the cardinal symptom of excessive daytime sleepiness20 and is cost-effective. 20,87

A literature search of the PubMed and The Cochrane Library databases to September 2013 (discussed in more detail in Chapter 4, Reviews for external evidence), without language restrictions, for full articles reporting randomised controlled trials (RCTs) assessing the efficacy of CPAP treatment in OSAS, in a population with an average age of 60 years or older and the capacity to give informed consent, identified only three studies which included patients with cardiovascular conditions and compared CPAP with sham CPAP88 or no CPAP. 89,90 None of the studies was conducted in the UK or in a secondary care setting. Furthermore, they did not collect generic measures of health utility. These studies were not generalisable to the overall patient population; consequently, these studies were not used to inform the cost-effectiveness estimates in the health economic model, and which were derived solely from the results of Positive Airway Pressure in Older People: a randomised controlled trial (PREDICT).

Multiple systematic reviews and meta-analyses have assessed the efficacy of CPAP therapy; the most recent and relevant to date being the National Institute for Health Research (NIHR) Health Technology Assessment (HTA) systematic review and economic analysis of CPAP devices for the treatment of OSAS. 20 This review concluded that CPAP is a clinically effective and cost-efficient treatment for moderate to severe OSA in well-defined middle-aged populations. It found that the majority of studies investigating the effect of CPAP treatment had enrolled patients between 44 and 58 years of age. However, it highlighted evidence gaps, with a need for trials in other patient groups, one such group being older people. It concluded that ‘clinical trials to define treatment effects at the extremes of age particularly in the elderly where cardiovascular comorbidity complicates assessment would be beneficial’. 20 Therefore, despite the high prevalence of OSA in older people, there is a paucity of evidence on the relative benefits or risks of CPAP treatment in older people. In addition, it cannot be assumed the benefits of CPAP treatment in younger populations will be replicated in older people.

Establishing the efficacy and cost-effectiveness of treatments for all common disease in older populations is a priority for health-care planners. PREDICT was an investigator-initiated project, funded by the HTA programme of the UK NIHR to address the evidence gap and enable the formulation of good-quality guidance on care for older people with OSAS.

Trial design

Positive Airway Pressure in Older People: a randomised controlled trial (ISRCTN90464927) was a pragmatic, single-blinded (investigator-blinded), parallel-group, multicentre RCT of 12 months’ duration (Figure 1).

FIGURE 1.

Trial design.

All patients were randomised to receive CPAP plus best supportive care (BSC) or BSC only. The coprimary outcomes were the clinical effectiveness of CPAP in improving subjective sleepiness at 3 months and the cost-effectiveness of CPAP over the 12-month period.

Recruiting centres

Recruitment took place at secondary and tertiary care referral centres in England, Scotland and Wales, serving a variety of ethnic and social groups, and including both urban and rural areas.

At the start of the trial, patients were recruited through six secondary/tertiary care referral centres: Churchill Hospital (Oxford), Musgrove Park Hospital (Taunton), Royal Brompton Hospital (London), Royal Infirmary Edinburgh (Edinburgh), St James’s University Hospital (Leeds) and St Woolos Hospital (Newport). As the trial progressed, a further 18 centres requested to join via NIHR portfolio database; these centres were sent a feasibility questionnaire and subsequently nine further centres were opened, one of which was later closed because of recruitment difficulties. This left eight additional secondary care referral centres: Aintree Hospital (Liverpool), Blackpool Victoria Hospital (Blackpool), City General Hospital (Stoke-on-Trent), Freeman Hospital (Newcastle upon Tyne), Great Western Hospital (Swindon), Heartlands Hospital (Birmingham), New Cross Hospital (Wolverhampton) and Royal Berkshire Hospital (Reading). All centres had established sleep services where patients with OSAS are diagnosed and treated with CPAP therapy.

Ethical consideration

The trial was approved via the Integrated Research Application System (National Research Ethics Service/NHS/Health and Social Care Committees) (reference number 09/H0708/33). The trial was also approved by the local NHS Research and Development Office at each site.

Patients

Interventions

Patients were randomised to receive CPAP plus BSC or BSC alone.

Assessment

Both groups had identical visit schedules. Structured clinical assessments were performed at baseline and at 3 months and 12 months. Assessment visits were carried out at each local centre. Occasionally, research nurses agreed to see a patient in the patient’s own home if he/she was unable to attend the hospital. All patients received a telephone call from their centres at 1 week, 1 month and 6 months to record symptoms and side effects and to optimise CPAP adherence. Additionally, all patients completed monthly diaries recording their ESS score, functionality, quality of life, health-care usage, change in medication, caffeine and alcohol intake, frequency of exercise and any side effects.

All patients enrolled in the trial underwent a domiciliary overnight respiratory polygraphy (Embletta^® GOLD™, Embla^®, Amsterdam, the Netherlands) prior to treatment allocation, which was scored centrally. Domiciliary overnight pulse oximetry (Pulsox^®–300i, Konica-Minolta Inc., Osaka, Japan) was performed at 3 and 12 months. Table 2 summarises the assessments completed at each time point.

Outcome measures

Data collection and monitoring

Data generated by all centres were collected on CRFs, which were posted to the Oxford Respiratory Trials Unit (ORTU), the trial data co-ordinating centre, where they were entered on to a database that was created and maintained by the Medical Research Council Clinical Trials Unit (MRC CTU). The staff entering the data into the database had no part in the data collection, analysis or interpretation. All patients’ trial consent forms were reviewed and a 100% automated check was conducted for the ESS inclusion criteria. Automated data checks for consistency and date were completed for all CRFs. Data were also checked for inconsistencies in range and missing data. Missing or ambiguous data were queried with individual research nurses and resolved whenever possible. Quality control of CRF data entry was completed on a regular basis throughout the duration of the trial. Site initiation visits were organised for all sites prior to commencing recruitment and were conducted by the chief investigator, trial manager and clinical research fellow. Interim monitoring visits were completed for five centres and source data verification was completed during those visits. Eleven close-out visits were completed remotely and four centres were visited. All adverse events were reviewed by the Independent Data Monitoring Committee (IDMC).

Randomisation

Patients were randomised using a telephone computerised service provided by the MRC CTU. Allocation was physically carried out during working hours from Monday to Friday. The allocation group was indicated to the unblinded research nurse once the baseline data collection was completed.

The randomisation programme was created by the MRC CTU in accordance with its standard operating procedure and held on a secure server, access to which was confined to the CTU data manager. Allocation was on a 1 : 1 basis with a random element of 80% and stratified by disease severity (enrolment ESS score of > 13 vs. ≤ 13), functionality using the TDS score of > 1 versus ≤ 1 and recruitment centre. In the analysis, baseline ESS scores and TDS scores were entered into models as fixed-effects continuous variables. Recruiting centre was adjusted for using random effects in order to avoid dropping centres that may recruit only a single patient.

Sample size

The primary analysis was the difference between the two treatment groups in the mean change of ESS score from baseline to 3 months. The ESS is a scale from 0 to 24 and a 1-point change on the ESS is indicative of a shift in the symptom state in one domain which was considered to be the minimally clinically important difference. In the recent NICE/HTA appraisal of CPAP for OSAS in middle-aged patients,20 the effect of CPAP treatment on the difference in ESS score in middle-aged patients with mild OSA was –1.07 [standard deviation (SD) 2.4]. The inclusion criterion for this trial was in the range of moderate OSAS severity, but, since sleepiness may be less pronounced in older people, the power calculations were performed assuming a treatment response similar to that seen in mild OSAS in middle-aged patients. To detect a 1-point change in ESS score (SD of change 2.4) required 244 patients randomised in a 1 : 1 ratio with a 90% power at the two-sided 5% significance level. In shorter (less than 6 months’ duration) randomised trials with a similar design, the loss to follow-up rate was approximately 5%. Since PREDICT was of longer duration and undertaken in older people with comorbidity, it was conceivable that the loss to follow-up rate could be up to 10%. Therefore, the sample size for the trial was 270 patients (135 in each group).

Statistical methods

The statistical analysis plan was finalised and approved by the Trial Management Group (TMG). Statistical significance was tested at the 5% level for all analyses. All analyses were adjusted for the minimisation factors (enrolment ESS score of > 13 vs. ≤ 13, functionality using the TDS of > 1 vs. ≤ 1 and recruitment centre) to optimise power and reduce bias. In addition to the minimisation factors, age, sex, ODI and body mass index (BMI) were also adjusted for in an additional analysis of the primary outcome. All analyses were by intention to treat, incorporating all randomised patients who had complete data on the outcome of interest (complete-case analysis). No adjustments for multiple testing were made, but the statistical significance of the secondary outcomes was interpreted cautiously because of the large number of secondary analyses performed.

A secondary sensitivity analysis of the primary outcome was performed in order to establish proof of principle whereby patients who swapped from the BSC group to CPAP were excluded from the analysis. The effect of baseline ESS score, age, ODI and BMI and the effect of CPAP use on the primary outcome were also investigated.

All analyses and modelling were undertaken in Stata version 12.0 (StataCorp LP, College Station, TX, USA).

Summary of changes to the protocol

The changes to the trial documents following National Research Ethics Committee (REC) approval in October 2009 are summarised below; a copy of the Statistical and health economic analysis plans is given in Appendix 1.

1. Substantial amendment SA01 (approved by the REC on 4 November 2009). Changed the version number of the PIS mentioned in the consent form to match the PIS version 2.0 already approved.

2. Substantial amendment SA02 (approved by the REC on 2 December 2009). Changed contact details, updated staff details, minor editing and formatting of the sleep diaries and added information regarding data transfer in the PIS. Clarified which ESS score measurements would be used in the analysis, quantified what was meant by a clinical diagnosis of OSAS, corrected a mistake in one of the exclusion criteria, added sections explaining the blinding in more detail and the delivery of CPAP/service provision and clarified the procedure for returning the driving questionnaire.

3. Substantial amendment SA03 (approved by the REC on 10 May 2010). Updated staff/committee details, minor editing and administrative changes. Further information added to the PIS. Standard letters inviting patients to attend their 3- and 12-month visits were introduced at the request of the participating centres. The sleep diaries were updated and information regarding Sibutramine was removed from the BSC booklet. Clarifications were required for the blinding procedure, one of the exclusion criterion, the minimisation criteria, the trial treatment, loss to follow-up and the procedure for assessing safety, quality control and adverse events section.

4. Non-substantial amendment NA04 (acknowledged by the REC on 14 May 2010). One of the minimisation criterion had been changed in error and was corrected.

5. Substantial amendment SA05 (approved by the REC on 24 February 2011). Updated staff/committee membership and contact details. Clarified that the results of the Embletta test done prior to trial enrolment were acceptable as long as they were done not more than 3 months before randomisation. Amended the coenrolment guidelines and listed the blood tests. Updated the monitoring, amendments and safety-reporting section so that it referred to a device trial rather than an investigational medicinal product trial.

6. Substantial amendment SA06 (approved by the REC on 20 June 2011). Clarified the primary and secondary outcomes, selection of centres and patients and treatment data collection. Updated the follow-up section. Corrected the sample size calculation and added information regarding the role of the IDMC and TMG.

7. Substantial amendment SA07 (approved by the REC on 8 June 2012). Clarification of how the cardiovascular risk is measured, the ESS score calculations and the analysis plan. Corrections of the statistical calculations, update of the trial manager’s details and administrative corrections.

All amendments were implemented prior to breaking of the treatment allocation code and prior to finalising the analysis plan.

Trial conduct

Recruitment

Recruitment took place between February 2010 and May 2012. The overall recruitment rate is shown in Figure 2. All the 12-month visits and trial exit were completed by May 2013. Although the trial was powered for 270 patients, 278 were recruited. This occurred because when approaching the target number a randomisation stop date was announced to coincidence with the end of a calendar month. Eight additional patients had completed their enrolment visit and randomisation prior to the official stop date. The TMG agreed the additional patients should be included.

FIGURE 2.

Cumulative recruitment.

The Consolidated Standards of Reporting Trials diagram shows the flow of patients through the trial (Figure 3). ‘Withdrew consent’ implies the patients withdrew from the treatment and trial, and ‘discontinued treatment’ implies the patient stopped their allocated treatment but remained in the trial. In total, 1614 individuals were screened as potential patients: of these 541 (34%) were eligible and subsequently 278 (51%) were randomised. 245 (88%) completed their 3-month follow-up and 231 (83%) completed their 12-month follow-up and the trial.

FIGURE 3.

Trial screening, enrolment, randomisation and follow-up.

Data collected on the screening logs enabled the 1073 ineligible patients to be grouped into the following categories:

• not meeting inclusion ODI or ESS criteria, n = 442 (41%)

• previous exposure to CPAP, n = 79 (7%)

• awake oxygen saturations < 90% on air or FEV₁/FVC < 60%, n = 171 (16%)

• being a professional driver, reporting sleepiness while driving, shift work or any severe symptom of OSAS for which the referring physician felt CPAP was mandatory, n = 216 (20%)

• no information or incomplete data, n = 165 (15%).

Baseline data

In total, 278 patients were randomised; 140 to the CPAP with BSC and 138 to BSC alone. All 278 patients completed the baseline enrolment visit. The majority of patients were male (82%), with a mean age of 70 years, ranging from 65 to 89 years, and had on average moderate OSAS: ESS mean score (SD) of 11.6 (SD 3.7) and ODI 28.7 (SD 19.1) events/hour. The majority of patients were white (96%) and obese and had on average 11 years of education and normal MMSE. A total of 228 (82%) were current drivers, and 146 (53%) slept alone. Baselines demographic are shown in Table 3, clinical characteristics in Table 4, sleep characteristics in Table 5 and sleep measurements in Table 6. None of the baseline data between groups were considered different to any important degree.

Coprimary outcomes

Subjective sleepiness

The primary outcome, the change in subjective sleepiness between groups at 3 months, is shown in Table 7. CPAP resulted in a mean change [standard error (SE)] of –3.8 (0.4) from an average (SD) of 11.5 (3.3) at baseline to 7.7 (4.0) at 3 months. BSC showed a mean change (SE) of –1.6 (0.3) from a baseline average (SD) of 11.4 (4.2) to 9.8 (4.3) at 3 months. The adjusted treatment effect at 3 months was –2.1 (95% CI –3.0 to –1.3) in favour of CPAP, which is statistically significant (p < 0.001). An additional analysis adjusting for age, sex, BMI and baseline ODI did not alter this result.

TABLE 7 - Change in ESS score at 3 months

SE, standard error.

Time assessed	BSC	CPAP
n randomised	138	140
n analysed	124	124
Baseline (at randomisation), mean (SD)	11.4 (4.2)	11.5 (3.3)
Month 3, mean (SD)	9.8 (4.7)	7.7 (4.1)
Month 4, mean (SD)	9.7 (4.2)	7.7 (4.3)
Mean of months 3 and 4 (SD)	9.8 (4.3)	7.7 (4.0)
Mean change from baseline (SE)	–1.6 (0.3)	–3.8 (0.4)
Treatment effect (95% CI), p-value	–2.1 (–3.0 to –1.3), p < 0.001

Sensitivity analysis

Sensitivity analyses were performed (1) excluding two patients who swapped from BSC to CPAP prior to the 3-month assessment and (2) including all randomised patients by replacing missing values using multiple imputation. Excluding the two patients who swapped from BSC to CPAP prior to the 3-month visit resulted in a treatment effect of –2.1 (95% CI –3.0 to –1.3), p < 0.001, in favour of CPAP, identical to the primary analysis.

Results from the imputation analyses, calculating the effect of the incomplete ESS score data reported by 14 patients, estimated a change of –2.0 (95% CI –2.8 to –1.2) in favour of CPAP (p < 0.001). Once again this showed the primary result to be robust. The imputation analysis assumes missing outcomes are similar to the observed outcomes in patients with similar characteristics, but this may not be true, as missing outcomes may be better or worse than those observed. The assumption can be varied to see how sensitive the observed results are to the missing data. Figure 4 shows that observed results are not sensitive and that extreme assumptions about the missing data are needed to make any significant change to the primary analysis (i.e. a 5-point difference between missing and observed values) so any sensible assumptions about the missing data do not change the results.

FIGURE 4.

Sensitivity analysis of the primary outcome to missing data. Missing outcomes are imputed under the assumption that the average of the missing outcomes differ to the mean of the observed outcomes by an amount corresponding to each point of the x-axis. Imputation is performed for each treatment group separately.

Planned exploratory analysis

Exploratory analysis were planned to investigate the effect of CPAP use and age, BMI, ESS score and ODI at baseline on the primary ESS score outcome. Patients were split into tertiles by their average CPAP use in the last month of follow-up prior to their 3-month visit. The analysis by CPAP use is shown in Table 8. The change in ESS score between baseline and 3 months in those who used CPAP the most (third tertile) was 11.4 at baseline and 6.1 at 3 months. This resulted in a treatment effect of –3.7 (95% CI –4.8 to –2.6), p < 0.001, compared with BSC.

TABLE 8 - The effect of the CPAP use on the ESS score over the month prior to 3-month assessment

Time assessed	BSC	CPAP
		First tertile	Second tertile	Third tertile
n	124	38	37	41
Mean usage (hours/night), (minimum–maximum)	–	0 (0–0)	1.9 (0.001–4.6)	6.4 (4.6–8.6)
Baseline ESS score, mean (SD)	11.4 (4.2)	10.6 (3.0)	12.0 (3.9)	11.4 (2.7)
ESS score month 3, mean (SD)	9.8 (4.3)	8.0 (3.9)	9.0 (4.5)	6.1 (2.7)
Change, mean (SD)	–1.6 (2.9)	–2.6 (3.9)	–3.1 (4.3)	–5.3 (3.4)
Treatment effect (95% CI)	–	–1.3 (–2.4 to 0.1)	–1.3 (–2.4 to –0.1)	–3.7 (–4.8 to –2.6)
p-value	–	0.032	0.034	< 0.001

The effect of CPAP therapy compared with BSC on the primary ESS score outcome was also assessed separately over age, BMI, ESS score and ODI at baseline using FPs and is shown in Figure 5. The treatment effect was larger in patients with higher ESS score at baseline (p < 0.001).

FIGURE 5.

Interactions between primary treatment effect on ESS score and age, BMI, baseline ESS score and ODI. (a) Treatment effect over age; (b) treatment effect over BMI; (c) treatment effect over baseline ESS score; and (d) treatment effect over ODI.

Secondary outcomes

Subjective sleepiness

The change in subjective sleepiness, as measured by the mean ESS score of months 10, 11 and 12, is shown in Table 10. CPAP resulted in a mean change (SD) of –4.2 (SD 4.1) in ESS score, from an average of 11.4 (SD 3.4) at baseline to 7.2 (SD 3.6) at 12 months. BSC showed a change of –2.1 (SD 3.6), from a baseline of 11.3 (SD 4.0) to 9.2 (SD 4.0) at 12 months. The difference between the two groups at 12 months was –2.0 (95% CI –2.8 to –1.2) in favour of CPAP, which was statistically significant (p < 0.001). A sensitivity analysis excluding eight patients who swapped from BSC to CPAP was performed but this did not alter the conclusion; the difference between the two groups was –2.1 (95% CI –3.0 to –1.3; p < 0.001), in favour of CPAP.

TABLE 10 - Change in ESS score at 12 months

Timed assessed	BSC	CPAP
n randomised	138	140
n analysed	122	116
ESS score baseline (at randomisation), mean (SD)	11.3 (4.0)	11.4 (3.4)
ESS score month 10, mean (SD)	9.3 (4.3)	7.3 (4.1)
ESS score month 11, mean (SD)	9.6 (4.4)	7.2 (4.1)
ESS score month 12, mean (SD)	9.0 (4.1)	7.0 (3.8)
Mean of months 10, 11 and 12, mean (SD)	9.2 (4.0)	7.2 (3.6)
Mean change from baseline (SD)	–2.1 (3.6)	–4.2 (4.1)
Treatment effect (95% CI), p-value	–2.0 (–2.8 to –1.2), p < 0.001

Continuous positive airway pressure reduced subjective sleepiness at 3 months; the effect was maintained at 12 months and was statistically significant (p < 0.001). This is shown graphically in Figure 6. Similarly, the effect was larger in patients with greater CPAP use. The analysis by CPAP use is given in Table 11.

FIGURE 6.

The treatment effect of CPAP compared with BSC care on subjective sleepiness. Adjusted treatment effects of CPAP and BSC and their 95% CI on the mean ESS score of months 3 and 4 (coprimary outcome) and of months 10, 11 and 12 (secondary outcome). Lower scores indicate an improvement.

TABLE 11 - The effect of the CPAP use on the ESS score over the 3 months prior to 12-month assessment

Descriptor	BSC	CPAP
		First tertile	Second tertile	Third tertile
n	122	52	30	30
Mean usage (hours/night) (minimum–maximum)	–	0 (0–0)	2.3 (0.002–4.4)	6.3 (4.5–8.9)
Baseline ESS score, mean (SD)	11.3 (4.0)	11.2 (3.5)	11.4 (4.0)	11.8 (2.6)
ESS score months 10,11 and 12, mean (SD)	9.2 (4.0)	8.1 (3.9)	7.3 (3.5)	5.6 (2.6)
Change, mean (SD)	–2.1 (3.6)	–3.0 (4.4)	–4.2 (3.4)	–6.2 (3.3)
Treatment effect (95% CI)	–	–1.0 (–2.0 to 0.1)	–2.0 (–3.2 to –0.7)	–3.6 (–4.9 to –2.4)
p-value	–	0.063	0.002	< 0.001

Objective sleepiness

Sleepiness was also measured objectively using the OSLER test at 3 and 12 months. The mean time to fall asleep is shown in Tables 12 and 13 (3 and 12 months, respectively). The difference between groups was statistically significant at 3 months (p = 0.024) in favour of CPAP but less so at 12 months (p = 0.058). The mean time for patients to fall asleep is also shown in Kaplan–Meier plots in Figure 7.

TABLE 12 - Oxford Sleep Resistance test at 3 months

Time assessed	BSC	CPAP	Treatment effect (minutes), (95% CI)	p-value
n	121	116	2.8 (0.4 to 5.2)	0.024
Baseline, mean time to sleep (minutes) (SD)	21.5 (13.4)	23.6 (12.7)
Month 3, mean time to sleep (minutes) (SD)	22.8 (13.9)	27.3 (12.4)
Mean change from baseline 3 months (SD)	1.3 (10.8)	3.6 (10.6)

TABLE 13 - Oxford Sleep Resistance test at 12 months

Time assessed	BSC	CPAP	Treatment effect (minutes), (95% CI)	p-value
n	115	110	2.6 (–0.1 to 5.3)	0.058
Baseline, mean time to sleep (minutes) (SD)	21.4 (13.3)	24.5 (12.7)
Month 12, mean time to sleep (minutes) (SD)	23.8 (13.4)	27.8 (11.6)
Mean change from baseline at 12 months (SD)	2.4 (11.9)	3.3 (13.2)

FIGURE 7.

Kaplan–Meier plot of average time taken to fall asleep for each patient at baseline, 3 months and 12 months. (a) Baseline; (b) 3-month visit; and (c) 12-month visit.

Quality of life and mood

Generic quality of life was assessed using the SF-36 (version 1) at 3 and 12 months. Raw scores were analysed with factor loadings obtained from Jenkinson et al. 102 The difference between groups in the energy/vitality domain was statistically significant at 3 months (p = 0.001) and 12 months (p = 0.004) in favour of CPAP. The MCS score was also statistically significant at 3 months (p = 0.046) but not at 12 months (p = 0.22). The physical functioning score was also statistically significant at 12 months (p = 0.033) in favour of CPAP but not at 3 months (p = 0.16). The difference between the two groups on each summary score at the 3- and 12-month visits is shown in Figure 8.

FIGURE 8.

Short Form questionnaire-36 items treatment effects at 3 and 12 months. Adjusted treatment effects and their 95% CI, CPAP versus BSC, on the MCS, the PCS and the eight individual components at 3 and 12 months. Higher score indicate an improvement.

Disease-specific quality of life was measured using the SAQLI, a sleep apnoea-specific questionnaire which also incorporates side effects associated with CPAP. Both groups showed an improvement but the effect was greater in the CPAP group at 3 months (p = 0.005) and 12 months (p = 0.001).

Mood was assessed using the HADS, which was summarised into an anxiety score and a depression score. Both groups showed a reduction in their score at 3 and 12 months but the difference between groups at either time point was not statistically significant.

The SF-36, SAQLI and HADS scores are shown in Tables 14 and 15 (3 and 12 months, respectively).

TABLE 14 - Quality of life and mood questionnaires at 3 months

Outcome	BSC			CPAP			Treatment effect (95% CI)	p-value
	n	Baseline, mean (SD)	Month 3, mean (SD)	n	Baseline, mean (SD)	Month 3, mean (SD)
SF-36
Bodily pain	125	59.9 (26.8)	60.5 (26.4)	123	61.9 (28.4)	61.4 (26.9)	–0.7 (–5.6 to 4.2)	0.78
Energy/vitality	123	45.8 (22.0)	47.0 (22.5)	121	49.9 (20.5)	56.6 (20.9)	6.4 (2.7 to 10.2)	0.001
General health	124	55.9 (21.8)	55.3 (22.1)	123	56.5 (23.4)	57.7 (22.1)	1.8 (–1.5 to 5.0)	0.29
Mental health	125	76.7 (14.7)	77.7 (16.8)	123	76.2 (17.2)	80.4 (15.4)	2.8 (–0.1 to 5.6)	0.062
Physical functioning	124	54.9 (29.0)	55.0 (29.5)	121	58.2 (26.3)	60.6 (27.5)	2.6 (–1.1 to 6.3)	0.16
Role emotional	125	72.3 (39.2)	72.8 (37.0)	122	76.8 (38.3)	78.7 (34.8)	3.0 (–4.3 to 10.3)	0.41
Role physical	122	40.4 (42.2)	44.5 (40.8)	122	53.1 (40.9)	53.5 (41.8)	2.0 (–6.4 to 10.4)	0.64
Social functioning	125	73.7 (27.8)	72.0 (29.0)	123	76.5 (25.8)	80.1 (25.1)	6.0 (1.1 to 11.0)	0.017
MCS	118	51.2 (9.9)	51.5 (10.0)	119	51.9 (10.1)	54.1 (8.9)	1.9 (0.0 to 3.8)	0.046
PCS	118	31.0 (13.6)	31.7 (15.1)	119	34.2 (13.8)	34.2 (14.3)	–0.2 (–2.5 to 2.1)	0.84
SAQLI
–	119	4.7 (1.2)	5.0 (1.1)	121	4.8 (1.2)	5.3 (1.1)	0.3 (0.1 to 0.5)	0.005
HADS
Anxiety	125	5.5 (3.7)	4.9 (3.5)	123	5.3 (4.0)	4.2 (3.4)	–0.5 (–1.1 to 0)	0.064
Depression	124	4.4 (3.0)	4.3 (2.9)	123	4.5 (2.8)	4.0 (2.9)	–0.4 (–0.9 to 0.1)	0.17

TABLE 15 - Quality of life and mood questionnaires at 12 months

Outcome	BSC			CPAP			Treatment effect (95% CI)	p-value
	n	Baseline, mean (SD)	Month 12, mean (SD)	n	Baseline, mean (SD)	Month 12, mean (SD)
SF-36
Bodily pain	117	60.2 (26.3)	59.2 (27.2)	114	61.2 (28.4)	60.5 (26.9)	0.1 (–5.4 to 5.5)	0.98
Energy/vitality	116	46.6 (21.8)	48.4 (22.6)	112	49.4 (20.4)	56.7 (21.4)	6.1 (1.9 to 10.3)	0.004
General health	116	56.0 (21.4)	54.8 (22.0)	114	55.9 (23.6)	57.8 (21.8)	2.6 (–1.1 to 6.4)	0.17
Mental health	117	76.7 (14.5)	78.0 (18.0)	113	76.3 (17.9)	79.7 (17.2)	1.9 (–1.7 to 5.4)	0.30
Physical functioning	117	55.5 (28.5)	54.3 (29.1)	113	57.2 (26.3)	60.7 (29.1)	4.7 (0.4 to 9.1)	0.033
Role emotional	117	72.6 (38.8)	72.9 (38.9)	113	76.4 (38.8)	79.6 (33.5)	5.1 (–3.4 to 13.7)	0.24
Role physical	116	41.8 (41.7)	42.2 (42.0)	112	52.5 (41.2)	50.4 (42.8)	3.0 (–6.6 to 12.6)	0.54
Social functioning	117	73.1 (26.9)	74.3 (29.2)	114	76.2 (26.2)	78.9 (25.3)	2.6 (–2.9 to 8.1)	0.35
MCS	114	51.1 (9.8)	52.0 (10.4)	108	52.1 (10.2)	53.9 (9.4)	1.4 (–0.8 to 3.6)	0.22
PCS	114	31.3 (13.2)	30.9 (15.3)	108	33.8 (13.9)	33.7 (14.9)	0.6 (–2.2 to 3.4)	0.68
SAQLI
–	114	4.7 (1.2)	5.1 (1.1)	113	4.8 (1.2)	5.5 (1.1)	0.4 (0.2 to 0.6)	0.001
HADS
Anxiety	117	5.5 (3.6)	4.5 (3.5)	114	5.2 (3.9)	4.1 (3.5)	–0.2 (–0.9 to 0.5)	0.58
Depression	116	4.4 (3.0)	4.2 (3.2)	114	4.6 (2.9)	3.9 (3.1)	–0.4 (–1.0 to 0.3)	0.23

Tertiary outcome

Of the 140 patients randomised to CPAP treatment, 120 (86%) reported that they were still using CPAP at 3 months and 99 (71%) at 12 months. Actual CPAP data were obtained in 117 patients at 3 months, with median usage of 1.52 hours/night [interquartile range (IQR) 0.19 to 5.12 hours], and 102 patients at 12 months, with median usage of 2.22 hours/night (IQR 0.10 to 5.09 hours). Assuming zero usage in those patients with missing data and who stopped treatment during follow-up gave a more conservative estimate of median CPAP use of 1.33 hours/night (IQR 0.13 to 5.0 hours) at 3 months and 1.26 hours/night (IQR 0.04 to 4.45) at 12 months. CPAP usage data are shown in Table 24.

Serious adverse events

There were 37 serious adverse events, of which 15 (in 12 patients) occurred in the CPAP group and 22 (in 13 patients) in the BSC group. They included two deaths, one in the CPAP group and one in the BSC group. All events were independently classified as unrelated to the trial. There was no suggestion of clinically important harm from CPAP use.

Self-reported side effects

This trial involved the use of an approved medical device which is the mainstay of treatment for OSAS in middle-aged populations. Therefore, the TSC did not expect any serious adverse events or adverse reactions of relevance to the device. Nonetheless, CPAP is associated with common side effects which were reported by the patients. The side effects were independently classified into categories, as suggested by the IDMC and TSC, and presented in Table 25. Treatment side effects are also incorporated in the SAQLI questionnaire.

Systematic review of existing cost-effectiveness evidence on continuous positive airway pressure

This systematic review provides an overview of the existing cost-effectiveness evidence, as well as an assessment of the quality and relevance of the data from the perspective of the UK NHS, on whether or not CPAP is a cost-effective treatment for patients aged ≥ 65 years. Appendix 3 reports the methods and detailed results of the systematic review, including summary data and extraction tables of the relevant studies. An overall summary of the cost-effectiveness evidence and key areas of uncertainty is provided below. The findings from the review provide the basis for the development of a new decision-analytic model reported in Economic model.

The systematic review on the existing cost-effectiveness evidence on CPAP for the treatment of OSAS found 10 relevant studies. Most studies used a Markov model to synthesise the available evidence;20,111–115 health states for cardiovascular events and RTAs were typically included. The characterised health effects of CPAP included lower risk of RTAs,20,111–117 work accidents,116 cardiovascular events20,112–114,116,117 and diabetes,116 and direct improvements in HRQoL from reduced sleepiness (all 10 studies). The improvement in HRQoL was estimated by converting other measures of quality of life118 or daytime sleepiness20,114 into health utilities, obtained directly from patients in before-and-after studies111–113,115,119 or from assumptions. 116,117 All studies concluded that CPAP is a cost-effective treatment for patients with OSAS. In general, the cost-effectiveness results were robust to alternative assumptions on parameter inputs with the exception of the health utility gain from treatment with CPAP.

The studies share two key limitations for the current decision problem:

1. None examined the cost-effectiveness of CPAP in patients aged ≥ 65 years.

2. All relied on indirect evidence to estimate the health utility benefit from treatment.

Although CPAP is likely to be beneficial in older people (as discussed in Chapter 1), the magnitude of such benefits cannot be inferred from the estimates obtained from younger populations. CPAP may be at least as cost-effective in older people, given their greater baseline risk of cerebrovascular events and the greater prevalence of age-related cognitive dysfunction, which CPAP could improve. On the other hand, CPAP may be less cost-effective, since older people generally suffer from other conditions which may affect sleep, such as Parkinson’s disease, which are not affected by CPAP. In addition, CPAP may be less effective in reducing BP in older people given their reduced acute BP response to each arousal from sleep. Finally, older people are likely to drive less often than working-age people (or shorter distances) and therefore have a lower risk of RTAs.

The limitations of the studies discussed above support a de novo analysis of the cost-effectiveness of CPAP, particularly focused on older people, and the integration of health utility evidence reported directly from patients in PREDICT. The analysis has two components, a within-trial analysis based on PREDICT which examines the cost-effectiveness of CPAP over 1 year and a decision model extrapolating to the patients’ lifetime and integrating external evidence.

Within-trial economic evaluation

Results

Health-related quality of life

The EQ-5D health utility values for each treatment group are shown in Table 29. The proportion of patients who answered the EQ-5D questionnaire at each month varied between 100% (CPAP group at baseline) and 66% (CPAP group at months 10 and 11). The proportion of missing data was similar across treatment groups. Figure 9 shows the observed EQ-5D values over the trial for both patient groups. No clear pattern emerged. The CPAP group had greater mean EQ-5D scores at baseline and at months 2, 4–7, 9 and 12 (and lower scores in the other months); however, the differences were not statistically significant. Therefore, no clear treatment effect emerges from the comparison of EQ-5D scores between groups.

TABLE 29 - European Quality of Life-5 Dimensions health utility values over the trial

Month	BSC				CPAP
	n	%	Mean	SD	n	%	Mean	SD
0	136	98.55	0.680	0.242	140	100.00	0.693	0.249
1	116	84.06	0.687	0.246	112	80.00	0.684	0.280
2	100	72.46	0.685	0.258	98	70.00	0.700	0.292
3	121	87.68	0.704	0.251	121	86.43	0.672	0.301
4	101	73.19	0.679	0.259	97	69.29	0.692	0.284
5	101	73.19	0.660	0.267	97	69.29	0.671	0.328
6	109	78.99	0.663	0.255	103	73.57	0.677	0.295
7	105	76.09	0.652	0.271	97	69.29	0.687	0.276
8	104	75.36	0.683	0.268	88	62.86	0.668	0.311
9	109	78.99	0.650	0.275	95	67.86	0.682	0.287
10	102	73.91	0.694	0.254	92	65.71	0.647	0.318
11	99	71.74	0.647	0.286	92	65.71	0.656	0.310
12	119	86.23	0.680	0.264	115	82.14	0.689	0.301

FIGURE 9.

European Quality of Life-5 Dimensions health utility values over the trial. Figure shows mean (as a square or lozenge) and the 95% CI (as lines).

The SF-6D health utility values are shown in Table 30 and Figure 10. The proportion of patients who answered the SF-36 questionnaire was higher than for the EQ-5D; this may have been related to the questionnaire being administered at the 3-month and 12-month clinic visit, while the EQ-5D was returned monthly by post. There was also less variability in the SF-6D health utility values than that observed in the EQ-5D. This may also have been related to the administration of the questionnaire or to the differences in the scoring algorithm. The difference between treatment groups was non-statistically significant for the SF-6D.

TABLE 30 - Short Form questionnaire-6 Dimensions health utility values over the trial

Month	BSC				CPAP
	n	%	Mean	SD	n	%	Mean	SD
0	137	99.28	0.659	0.092	138	98.57	0.661	0.088
3	125	90.58	0.661	0.088	123	87.86	0.681	0.087
12	118	85.51	0.653	0.096	113	80.71	0.679	0.111

FIGURE 10.

Short Form questionnaire-6 Dimensions health utility values over the trial. Figure shows mean (as a square or lozenge) and the 95% CI (as lines).

Health-care resource use and costs

The number and proportion of returned questionnaires on health-care resource are shown in Table 31. Baseline refers to the resource use in the month prior to enrolment. The proportion of returned questionnaires was similar across treatment groups and varied between 100% and 66%.

TABLE 31 - Number and proportion of returned questionnaires on health-care resource use

Month	BSC		CPAP
	n	%	n	%
Baseline	138	100.00	140	100.00
1	118	85.51	116	82.86
2	101	73.19	99	70.71
3	122	88.41	120	85.71
4	102	73.91	96	68.57
5	101	73.19	98	70.00
6	109	78.99	105	75.00
7	106	76.81	96	68.57
8	106	76.81	92	65.71
9	109	78.99	95	67.86
10	102	73.91	93	66.43
11	99	71.74	93	66.43
12	120	86.96	116	82.86

The resources used by each patient group [average number of times (SD) the resource was used] over the 12-month follow-up period are shown in Table 32. The estimates were obtained prior to multiple imputation of missing data. The group allocated to BSC alone had more contacts with the NHS but the differences are non-statistically significant. The most frequent contact was visits to GP (CPAP 6.93 vs. BSC 7.27 visits per patient). The second most frequent NHS contact were visits to the nurse (CPAP 3.36 vs. BSC 4.78 visits per patient), followed by outpatient appointments (CPAP 2.83 vs. BSC 3.65 per patient). Hospital overnight admissions were rare (CPAP 0.42 vs. BSC 0.55 per patient). There were 93 patients in the CPAP group who initiated medication (386 items) and 62 who discontinued medication (197 items), whereas in the BSC group 92 patients initiated medication (444 items) and 57 discontinued medication (192 items).

TABLE 32 - Health-care resource use over the trial

Type	BSC		CPAP
	Average per patient	SD	Average per patient	SD
Any contact with NHS	7.30	1.69	6.70	1.72
Visit to GP	7.27	3.20	6.93	2.94
GP home visit	0.32	0.67	0.18	0.47
Visit to nurse	4.78	3.23	3.36	2.16
Nurse home visit	0.27	0.80	0.14	0.51
GP telephone call	0.92	1.07	0.54	0.78
NHS direct telephone call	0.21	0.50	0.06	0.27
Ambulance	0.22	0.47	0.17	0.46
Accident and emergency	0.42	0.68	0.32	0.55
Outpatient clinic	3.65	2.48	2.83	1.99
Patients who stayed in hospital overnight at least once	0.42	0.64	0.33	0.57
Hospital overnight admissions	0.55	1.30	0.42	1.01
Emergency admissions	0.23	0.47	0.19	0.44
Nights in hospital	2.15	4.85	1.11	2.62

The health-care resource costs over the trial are shown in Table 33. These costs refer to the resource use data presented in Table 32 and multiplied by the relevant unit costs (see Table 26). On average, patients randomised to CPAP incurred less cost, but the difference was not statistically significant. The highest costs were those related to outpatient appointments, hospital overnight admissions and visits to the GP. The SD shows how much dispersion there was around the average cost; in some cost categories the SD was greater than the mean (e.g. GP home visit, nurse home visit), indicating that there was a large variability in the costs incurred by each patient.

TABLE 33 - Health-care resource costs over the trial

Type	BSC alone		CPAP
	Average (£)	SD (£)	Average (£)	SD (£)
Visit to GP	312.62	137.71	298.03	126.56
GP home visit	35.57	73.41	20.26	51.98
Visit to nurse	69.27	46.78	48.79	31.39
Nurse home visit	9.35	27.85	5.04	17.69
GP telephone call	24.05	27.72	14.12	20.38
NHS Direct telephone call	5.95	13.94	1.55	7.51
Ambulance	64.79	136.53	50.76	133.72
Accident and emergency	45.40	73.27	34.23	59.88
Outpatient clinic	386.84	263.18	300.25	210.43
Hospital overnight admissions	321.17	757.68	246.78	590.14
Emergency admissions	35.82	74.40	30.57	68.88
Total costs associated with health-care resource use	1311	1009.44	1050	830.21

Costs of CPAP treatment

The average costs of CPAP treatment per patient (including the respective components) are shown in Table 34. All patients allocated to the intervention received a standardised CPAP machine (S9 Autoset^TM ResMed (UK) Ltd, Abingdon, Oxfordshire, UK) and a mask. Eighty-two patients (59%) also received a humidifier. Patients who discontinued CPAP were assumed to return the machine (together with the humidifier, if relevant). Only eight patients allocated to the BSC group switched to CPAP treatment. Since the analysis follows intention to treat, these patients are assumed not to incur the costs of CPAP treatment. Given the small number of patients switching to CPAP treatment, this is unlikely to affect the results. The average cost of CPAP treatment was estimated at £201.14 per patient per year.

TABLE 34 - Costs of CPAP treatment

Item	Cost element	Number	Average cost per patient (£)
A	Annual equivalent cost of CPAP machine	–	70.32
B	Annual equivalent cost of humidifier	–	26.98
C	Number (and proportion) of patients who received a humidifier	82 (59%)	–
D	Average annual equivalent cost of humidifier per patient ( = B × C)	–	15.81
E	Average annual equivalent cost per patient ( = A + D)	–	86.13
F	Average cost of masks	–	104
G	Average cost of masks assuming (10% of patients received 2) ( = 1.1 × F)	–	114
H	Average cost per filter	–	0.58
I	Average cost of filters per patient per year (2 filters per year) ( = 2 × H)	–	1.16
	Average cost of CPAP treatment per patient ( = E + G + I)	–	201.14

Cost-effectiveness analysis

The cost-effectiveness results for PREDICT are presented in Table 35. The analysis was conducted post multiple imputation and included the costs of CPAP treatment (£201.14) for those patients allocated to the CPAP group. Detailed costs and HRQoL post imputation are given in Appendix 3. The average cost per patient was £1363 (95% CI £1121 to £1606) for those allocated to CPAP and £1389 (95% CI £1116 to £1662) for those allocated to BSC alone. The accrued cost was, on average, –£35 (95% CI –£390 to £321) lower for those allocated to CPAP. The average QALYs obtained from EQ-5D health utilities were 0.680 (95% CI 0.638 to 0.722 QALYs) for the CPAP group and 0.666 (95% CI 0.627 to 0.705 QALYs) for those allocated to BSC alone. The average QALYs obtained from SF-6D health utilities for were 0.678 (95% CI 0.664 to 0.691 QALYs) for the CPAP group and 0.658 (95% CI 0.643 to 0.673 QALYs) for those allocated to BSC alone. The CPAP group experienced more EQ-5D QALYs [0.005 (95% CI –0.034 to 0.044 QALYs)] and more SF-6D QALYs [0.018 (95% CI 0.003 to 0.034 QALYs)]. The improvements in QALYs were small, albeit statistically significant, for the SF-6D. The improvement of 0.005 QALYs is equivalent to 2 days in full health and the improvement of 0.018 QALYs is equivalent to 7 days. Overall, CPAP appeared to have improved health outcomes as well as reduced overall costs to the NHS. Therefore, CPAP with BSC dominated BSC alone. In these situations, it is not appropriate to present ICERs. 132

TABLE 35 - Cost-effectiveness results (post multiple imputation) over the trial

Treatment group	Costs		EQ-5D QALYs		SF-6D QALYs
	Average (£)	SE	Average	SE	Average	SE
CPAP	1363	£123	0.680	0.021	0.678	0.007
BSC	1389	£139	0.666	0.020	0.658	0.008
Incremental costs and QALYs
CPAP with BSC – BSC alone	–35	£180	0.005	0.020	0.018	0.008

The cost-effectiveness planes for the base case with EQ-5D QALYs and the alternative scenario with SF-6D QALYs are shown in Figure 11a and b, respectively. The lines represent the cost-effectiveness thresholds conventionally used in the NHS (£20,000 per additional QALY gained). The simulations for the cost-effectiveness results using EQ-5D QALYs were evenly spread across the four quadrants while the simulations using SF-6D QALYs were mostly concentrated on the eastern quadrants, indicating that there was considerable uncertainty as to CPAPs health benefits, particularly those captured by EQ-5D, and whether or not it was cost-saving. The small improvement in health outcomes was more certain with SF-6D QALYs.

FIGURE 11.

Cost-effectiveness plane for the overall population. (a) EQ-5D QALYs; and (b) SF-6D QALYs.

The cost-effectiveness acceptability curve for the base case (EQ-5D QALYs) and its alternative scenario with SF-6D QALYs is shown in Figure 12. The probability that the intervention was cost-effective at the thresholds conventionally used by NHS is 0.61 per QALY gained for the base case and 0.96 for the scenario with SF-6D QALYs. In the base case, the probability that CPAP was cost-effective plateaus at 0.60 as the cost-effectiveness threshold increased. This occurred because of the uncertainty around the improvement in EQ-5D QALYs observed during the trial, reflected in the 95% CI of –0.034 to 0.044. In contrast, the probability that CPAP was cost-effective was above 0.9 across the full range of cost-effectiveness thresholds with SF-6D QALYs. This is consistent with the scatter pattern in the cost-effectiveness plane (see Figure 11).

FIGURE 12.

Cost-effectiveness acceptability curve.

Subgroup analysis

The results for the subgroups defined according to ESS score at baseline are shown in Table 36. There were 184 patients with ESS score of < 13 (CPAP, n = 88; BSC, n = 96) and 94 patients with ESS score of ≥ 13 (CPAP, n = 52; BSC, n = 42). For the less severe OSAS subgroup, CPAP appears to be less costly and provide fewer QALYs than BSC; therefore, the ICER for BSC alone versus CPAP was £1118 per QALY gained for the base case with EQ-5D QALYs. CPAP dominated in the SF-6D QALYs scenario because it was associated with lower costs and greater QALYs; therefore, an ICER was not calculated. For the more severe OSAS subgroup, CPAP dominated BSC alone because it was associated with lower costs and better health outcomes in both the base case with EQ-5D QALYs and the scenario with SF-6D QALYs.

TABLE 36 - Cost-effectiveness results for subgroup populations (post multiple imputation)

Treatment group	Costs		EQ-5D QALYs		SF-6D QALYs
	Average (£)	SE (£)	Average	SE	Average	SE
Subgroup ESS score of < 13
CPAP	1393	162	0.677	0.027	0.675	0.009
BSC	1440	179	0.684	0.023	0.660	0.10
Incremental costs and QALYs
CPAP with BSC – BSC alone	–19	231	–0.017	0.025	0.018	0.010
Subgroup ESS score of ≥ 13
CPAP	1313	188	0.686	0.035	0.682	0.012
BSC	1274	212	0.624	0.037	0.654	0.012
Incremental costs and QALYs
CPAP with BSC – BSC alone	17	270	0.049	0.032	0.018	0.012

There was considerable uncertainty in these results because in both subgroups, similarly to the overall population, the differences in costs were small and not statistically significant, –£19 (95% CI –£475 to £438) for the less severe OSAS subgroup and £17 (95% CI –£520 to £555) for the more severe OSAS subgroup. The differences in QALYs were also small and not statistically significant. The less severe OSAS subgroup experienced a difference in EQ-5D QALYs of –0.017 (95% CI –0.066 to 0.033 QALYs) and in SF-6D QALYs of 0.018 (95% CI –0.002 to 0.038 QALYs). The more severe OSAS subgroup experienced positive differences for both EQ-5D and SF-6D QALYs, although not statistically significant, at 0.049 (95% CI –0.014 to 0.111) for EQ-5D QALYs and 0.018 (95% CI –0.006 to 0.043) for SF-6D QALYs. Thus, it is difficult to draw definite conclusions from this analysis given the level of uncertainty around the results. The EQ-5D QALYs appeared to follow the improvement in ESS score, which is more pronounced in the more severe OSAS subgroups (see Table 8). The improvement in SF-6D QALYs was similar in the two subgroups, although the differences in QALYs between CPAP and BSC alone were small across subgroups.

The cost-effectiveness planes for more severe (ESS score at baseline of ≥ 13) and less severe OSAS (ESS score at baseline of < 13) are shown in Figure 13a and c, and b and d, respectively. As with Figure 11, results for EQ-5D QALYs are shown in Figure 13a and c and for SF-6D QALYs are shown in Figure 13b and d. The axis scales are the same to facilitate comparison. The results differed depending on ESS score at baseline and HRQoL instrument used to obtain QALYs. In the less severe OSAS population, the simulations for EQ-5D QALYs were mostly spread across the western quadrants, indicating decrements in health outcomes. The simulations for SF-6D QALYs were concentrated in the eastern quadrants, indicating positive gains in health. In the more severe OSAS population, most of the simulations were located in the eastern quadrants for both EQ-5D and SF-6D QALYs. In other words, we are more confident that CPAP was cost-effective in the more severe OSAS subgroup.

FIGURE 13.

Cost-effectiveness plane for population subgroups. (a) ESS score of < 13 EQ-5D; (b) ESS score of < 13 SF-6D; (c) ESS score of ≥ 13 EQ-5D; and (d) ESS score of ≥ 13 SF-6D.

The cost-effectiveness acceptability curves for the subgroups are shown in Figure 14. The certainty around the cost-effectiveness of the intervention depended on the HRQoL instrument and the subgroup considered. In the less severe OSAS subgroup, the probability that CPAP was cost-effective decreased as the cost-effectiveness threshold increased for EQ-5D QALYs but increased for SF-6D QALYs. This reflected the distribution of the differences in QALYs presented in Figure 13. Using EQ-5D QALYs, the mean difference was close to zero but the majority of the simulations returned a decrement in QALYs between the treatment groups. In the cost-effectiveness acceptability curve, that decrement was compensated by the decrease in costs. However, as the threshold increased, the value placed on QALYs increased in relation to the costs savings. Therefore, for higher threshold values, CPAP was less likely to be cost-effective. Using SF-6D QALYs, the majority of simulations returned a gain in QALYs between treatment groups. This was because, as the threshold increased, the gains were valued, and the probability that CPAP was cost-effective increases. In the more severe OSAS group, the probability that CPAP was cost-effective increased as the threshold increased. This reflected the cost-effectiveness plane in Figure 13, where most of the simulations, for both EQ-5D and SF-6D QALYs, returned gains in health.

FIGURE 14.

Cost-effectiveness acceptability curves for less severe OSAS population. (a) ESS score of < 13; and (b) ESS score of ≥ 13.

Sensitivity analysis

A selection of the results for the sensitivity analysis is presented in Table 37 (see Appendix 3 for additional information). The cost differences and QALY differences were small and close to zero in the majority of analyses, and this resulted in variable ICERs. Therefore, while the results were relatively consistent in terms of costs and QALYS, the cost-effectiveness conclusions were mixed. In two scenarios with EQ-5D QALYs and three with SF-6D, QALYs CPAP dominated BSC alone. In four scenarios with EQ-5D QALYs and three with SF-6D QALYs, CPAP had a positive ICER. In two of the scenarios with EQ-5D QALYs, the ICER was above the conventional thresholds of cost-effectiveness used by NICE of £20,000 and £30,000 per QALY gained.

TABLE 37 - Summary results of the sensitivity analysis

NA, not applicable.

D indicates that CPAP dominates because it is associated with both lower costs and QALY gains.

Scenario	Difference in costs		Difference in EQ-5D QALYs		ICER/EQ-5D QALYs (£)	SF-6D QALYs		ICER/SF-6D QALYs (£)	Probability that CPAP is cost-effective at £20,000/QALY gained
	Average (£)	SE (£)	Average	SE		Average	SE		EQ-5D QALYs	SF-6D QALYs
Base case	–35	180	0.005	0.020	D	0.018	0.008	D	0.61	0.96
1: Frequent replacement scenario (CPAP costs = £608.95)	373	180	0.005	0.020	74,600	0.018	0.008	20,722	0.25	0.46
2: CPAP used for 1 year (CPAP costs = £710.16)	474	180	0.005	0.020	94,800	0.018	0.008	26,333	0.20	0.30
3A: Complete-case analysis (EQ-5D QALYs)	–258	258	0.010	0.029	D	NA	NA	NA	0.75	NA
3B: Complete-case analysis (SF-6D QALYs)	–201	247	NA	NA	NA	0.015	0.013	D	NA	0.91
4: Mean interpolation	43	201	0.006	0.018	7167	0.014	0.007	3071	0.55	0.81
5A: Missing not at random (costs)	–27	201	0.005	0.020	D	0.018	0.008	D	0.62	0.96
5B: Missing not at random (QALYs)	–35	180	–0.006	0.19	5833	0.012	0.009	D	0.42	0.85

Economic model

The ICER for CPAP was above the conventional thresholds of cost-effectiveness in scenarios 1 and 2, which assumed a higher treatment cost for CPAP. Scenario 1 assumed a greater frequency in the replacement of the machine and in the use of the consumables and scenario 2 assumed that the CPAP machine was used for only 1 year then discarded. In scenario 1, the CPAP and humidifier were assumed to be replaced every 3 years, the masks every 3 months and the filters monthly. The average cost of CPAP per patient per year tripled in scenario 1 from £201.14 in the base case to £608.94 and also more than tripled in scenario 2 to £710.16. This increase in the costs of CPAP resulted in an increase in the average difference in total costs, from –£35 (95% CI –£390 to £321) in the base case to £373 (95% CI £17 to £729) in scenario 1 and to £474 (95% CI £119 to £830) in scenario 2. The ICER for scenario 1 using EQ-5D QALYs was £74,600 and for scenario 2 was £47,800; for the same scenarios using SF-6D QALYs, it was £94,800 and £26,333, respectively. These results suggested that the cost-effectiveness of CPAP was highly sensitive to the costs of CPAP treatment.

Scenario 3 used only the data from patients who completed all questionnaires on HRQoL and costs. It assumed that these data were missing completely at random; that is, that the probability that data were missing was independent of both the observed and unobserved data. CPAP dominated BSC alone.

In scenario 4, missing data were imputed with the average of the observed data for each patient. The mean difference in costs changed slightly to £43 (95% CI –£353 to £440) while the differences in QALYs remained unchanged. Nevertheless, the change in the mean difference in costs between treatment groups resulted in a positive ICER of £7167 per EQ-5D QALY gained and £3071 per SF-6D QALY gained. This showed how small differences in costs or QALYs had a large impact on the results. Note that this scenario should be considered with caution, as it assumed that patients’ costs and HRQoL were expected to remain constant over the trial, which may not be the case.

The ‘missing not at random’ scenarios explored how the results changed if missing data on HRQoL and costs were assumed to be systematically different from that of patients with similar characteristics but who returned the questionnaires. The difference in costs changed very slightly, to –£26 (95% CI –£422 to £369), with little impact on the results. The difference in EQ-5D QALYs changed more noticeably from 0.005 (95% CI –0.034 to 0.044) to –0.006 (95% CI –0.044 to 0.031) whereas the difference in SF-6D QALYs changed slightly from 0.018 (95% CI 0.003 to 0.034) to 0.012 (95% CI –0.006 to 0.03). As a result, assuming that patients with missing data experienced lower HRQoL reduced the probability that CPAP was cost-effective from 0.61 to 0.42 for EQ-5D QALYs and from 0.96 to 0.85 for SF-6D QALYs.

Final model

A cohort Markov model was developed to evaluate the cost-effectiveness of CPAP compared with BSC alone. The model tracked the health outcomes (health utility values, deaths) and costs over the cohort’s lifetime. In the base case, the model consisted of two health states: OSAS (treated with CPAP or treated with BSC alone) and death (Figure 15). Cycle length was 1 year. CPAP was assumed to improve HRQoL and reduce NHS costs as observed in PREDICT (for more details see Within-trial economic evaluation). Adherence in the first year was obtained from PREDICT and assumed to deteriorate over time, as reported in McArdle et al. 148 The estimated of adherence used in the economic model is presented in Table 39. Adherence in McArdle et al. reduced by 10% in year 2, by 1% in year 3 and by 5% in year 4. 148 Applying these reductions to the adherence at 1 year observed in PREDICT yielded an adherence of 63.9% at year 2, 63.3% at year 3 and 60.1% at year 4. After year 4, adherence was assumed to remain constant.

TABLE 39 - Adherence with CPAP therapy modelled from previous studies

Adherence	PREDICT	McArdle et al.148	Reduction	Decision model
1 year	71%	84%	–	71%
2 years	–	74%	84% – 74% = 10%	71% × 90% = 63.9%
3 years	–	73%	74% – 73% = 1%	63.9% × 99% = 63.3%
4 years	–	68%	73% – 68% = 5%	63.3% × 95% = 60.1%

In a scenario analysis, the model included four health states (OSAS, OSAS post CHD, OSAS post stroke and death) and two events (CHD and stroke). In this scenario, CPAP not only improved HRQoL and reduced NHS costs but also changed the risk for cardiovascular events. The patient cohort started in the OSAS state. At each 1-year cycle, patients were at risk of CHD and stroke. Following CHD or stroke, patients moved to the state post CHD or post stroke, respectively. The states of post CHD and post stroke had an elevated risk of death. The Framingham risk equation was used to link the effect of CPAP on intermediate outcomes (systolic BP, cholesterol) to cardiovascular events.

FIGURE 15.

Diagram of model structure for the scenario analysis.

Health-related quality of life and quality-adjusted life-years

Health-related quality of life was expressed in terms of QALYs by quality adjusting the period of time for which the average patient was alive within the model using an appropriate health utility value. Table 40 shows the health utility values used in the model. Health utility in the OSAS state corresponded to the average baseline EQ-5D in the patients that took part in PREDICT. The effect of CPAP during the first year corresponded to the improvement in EQ-5D QALYs observed in the trial at 1 year, adjusted for baseline, as presented in Table 35. The effect in subsequent years was adjusted for adherence assuming that reductions in adherence had a proportional impact on treatment effect. For example, a reduction in compliance of 10% from 31% to 27.9% reduced the effect of CPAP in health utility by 10% from 0.005 to 0.0045. The health utility decrements associated with age, CHD and stroke were obtained from the catalogue of EQ-5D scores developed by Sullivan et al. 150

TABLE 40 - Health utility values used in the model

NA, not applicable.

Standard errors of CPAP effects years 2 to 4 assumed the same as in year 1.

Health state or event	Mean		SD or SE		Distribution	Source
	EQ-5D	SF-6D	EQ-5D	SF-6D
OSAS untreated	0.687	0.660	0.245	0.090	Beta	PREDICT
CPAP effect in year 1	0.005	0.018	0.020	0.008	Normal	PREDICT
CPAP effect in year 2	0.0045	0.0162	0.020a	0.008a	Normal	Calculated (effect × 90%)
CPAP effect in year 3	0.0045	0.0160	0.020a	0.008a	Normal	Calculated (effect × 89%)
CPAP effect in year 4	0.0042	0.0152	0.020a	0.008a	Normal	Calculated (effect × 85%)
Age decrement	0.0003	NA	0.0002	NA	Normal	Sullivan et al.150
For the scenario with cardiovascular events
Stroke decrement	0.1009		0.0123		Normal	Acute cerebrovascular disease150
CHD decrement	0.0557		0.0112		Normal	Acute MI150

Cardiovascular events

In the scenario including cardiovascular events, the Framingham risk equations were employed to estimate the risk of events from the intermediate outcomes recorded in PREDICT. 20,143 The characteristics of the patient cohort for the scenario with cardiovascular events are presented in Table 41. These correspond to the average patient characteristics at baseline in PREDICT. Table 42 presents the probability of a cardiovascular event based on the characteristics of the cohort at baseline. The probability was calculated for subgroups of patients by sex, smoking status and diabetes using only those patients with complete data for these variables at baseline. The risk for the PREDICT patient population was a weighted average of the risks for each subgroup, weighted by their relative proportions in the population.

TABLE 41 - Characteristics of the patient cohort for the scenario with cardiovascular events

HDL, high-density lipoprotein.

Variables	Mean	SD
Age (years)	70.59	4.66
Systolic BP (mmHg)	139.01	18.92
Total cholesterol, mmol/l (mg/dl)	4.58 (177.02)	1.04 (40.35)
HDL cholesterol, mmol/l (mg/dl)	1.24 (31.00)	0.34 (13.11)

TABLE 42 - Probability of cardiovascular events at baseline (using characteristics presented in Table 41)

NA, not applicable.

Sex	Male				Female				Overall
Smoking status	Yes	Yes	No	No	Yes	Yes	No	No	–
Diabetes	Yes	No	No	Yes	Yes	No	No	Yes	–
Number of patients	5	8	102	56	0	0	15	5	–
1-year probability of cardiovascular events
CHD	0.053	0.041	0.027	0.036	NA	NA	0.012	0.024	0.030
Stroke	0.010	0.006	0.003	0.005	NA	NA	0.002	0.006	0.004
Death from CHD	0.013	0.011	0.006	0.007	NA	NA	0.001	0.007	0.006
Death from cardiovascular disease	0.014	0.011	0.006	0.008	NA	NA	0.003	0.008	0.007

All-cause death

In the model, patients are at risk of death at every cycle. Patients in the OSAS state experience an age-dependent risk of death, obtained from interim lifetables for England and Wales for the years 2009–11. 151 For the scenario including cardiovascular events, the age-dependent risk of death from causes other than cardiovascular was estimated using a cause elimination approach. Patients in the post-stroke or post-CHD OSAS group (who survived the cardiovascular event) were then at an elevated risk of death as in the McDaid et al. analysis, with a relative risk of 3.2 (95% CI 2.67 to 3.83)152 for the post-CHD state and a relative risk of 2.3 (95% CI 2.0 to 2.7) for the post-stroke state. 153

Resource use and costs

Resource use and costs can be split into two components: (1) those related to CPAP therapy and (2) those related to cardiovascular events, only applicable in the scenario including cardiovascular events. The within-trial analysis estimated that patients allocated to CPAP accrued on average £35 (95% CI –£390 to £321) lower costs (Table 43). This cost difference included the costs of CPAP and the costs associated with any health-care resource use. Therefore, this difference in costs was applied to the hypothetical patient cohort in the OSAS state. As with the improvement in health utility, the cost difference was adjusted for reduced adherence over time. For the scenario including cardiovascular events, costs associated with cardiovascular events were obtained from the McDaid et al. report and inflated to 2011–12 prices. 20,124,154

TABLE 43 - Costs used in the model

NA, not applicable; PSSRU, Personal Social Services Research Unit.

Parameter	Cost		Distribution	Source
	Mean	SE
Costs related to CPAP
Costs in the OSAS state untreated	£1389	£139	Gamma	PREDICT, within-trial analysis (see Table 10)
Effect of CPAP on costs	–£35	£180	Normal	PREDICT, within-trial analysis (see Table 10)
Costs related to cardiovascular events
Cost of fatal CHD event
Price year 2004–05	3021	367	Normal	McDaid et al.,20 Briggs et al.154
Inflated to 2011–12	3716	451
Cost of non-fatal CHD event
Price year 2004–05	9997	429	Normal	McDaid et al.,20 Briggs et al.154
Inflated to 2011–12	12,296	528
Ongoing cost of CHD
Price year 2004–05	751	117	Normal	McDaid et al.,20 Briggs et al.154
Inflated to 2011–12	924	144
Inflation index from 2004–05 to 2011–12	285.7/232.3 = 1.23	NA	NA	PSSRU124
Acute cost of stroke (year 1)
Price year 2004–05	9067	294	Normal	McDaid et al.,20 Vergel et al.155
Inflated to 2011–12	11,152	362
Ongoing cost of stroke (year 2 and beyond)
Price year 2004–05	2392	282	Normal	McDaid et al.,20 Vergel et al.155
Inflated to 2011–12	2942	347

Patient population

The model followed a hypothetical patient cohort that corresponded to the patients in PREDICT, that is patients aged 65 years and over with OSAS (see Table 41 for the patients’ characteristics at baseline).

Sensitivity analyses

The sensitivity analyses conducted in the model-based analysis are summarised in Table 44. All analyses were fully probabilistic. The sensitivity analyses aimed to explore the robustness of the results to the main assumptions and parameter inputs. Therefore, scenario analyses were performed for the effect of CPAP on cardiovascular risk outcomes and the cost of CPAP therapy. Univariate sensitivity analysis was conducted to the cost of the CPAP machine.

TABLE 44 - Details of the key elements of the base-case analysis and variation used in the scenario analysis

Scenario	Element	Position in base-case analysis	Variation in the sensitivity analysis
1	CPAP used for 1 year	The costs of the CPAP machine and the humidifier are annuitised over 7 years. Yearly replacement for masks. Filters replaced every 6 months	CPAP is assumed to be used for 1 year and discarded after that; therefore, the cost of the machine is not annuitised. CPAP therapy costs £710.16 per patient
2	Cardiovascular effects with CPAP direct effects	Changes in the risk of cardiovascular events are not included in the base-case model	CPAP changes the risk of cardiovascular events as predicted by the Framingham risk equation through its effect on BP and cholesterol as observed in the PREDICT clinical trial. CPAP increases health outcomes and reduces costs are reduced in PREDICT
3	Cardiovascular effects only	Changes in the risk of cardiovascular events are not included in the base-case model	CPAP changes the risk of cardiovascular events as predicted by the Framingham risk equation through its effect on BP and cholesterol as observed in the PREDICT clinical trial. The direct effect of CPAP on costs and QALYs is not considered

Results

Base case

The results for the base case are presented in Table 45. CPAP reduced the average costs per patient by –£369 and improved health outcomes by 0.051 EQ-5D QALYs. The improvement for SF-6D QALYs was 0.182. Since CPAP reduced costs and improved health outcomes, it dominated BSC alone and an ICER is not calculated.

TABLE 45 - Cost-effectiveness results for the base case

Treatment	Average costs	Average EQ-5D QALYs	Average SF-6D QALYs
CPAP	£15,887	8.046	7.862
BSC	£16,216	7.994	7.680
Incremental costs and QALYs
CPAP with BSC – BSC alone	–£329	0.051	0.182

The distribution of average costs and average QALYs over the 10,000 simulations conducted for the probabilistic sensitivity analysis are shown in Figure 16. Similarly to the within-trial analysis, there was considerable uncertainty around the results using EQ-5D QALYs. In the analysis with SF-6D QALYs, most simulations were located in the eastern quadrants; the large majority were below the cost-effectiveness threshold, represented by the diagonal line.

FIGURE 16.

Cost-effectiveness plane for the model-based base-case analysis. (a) EQ-5D; and (b) SF-6D.

The probability that CPAP was cost-effective over a range of cost-effectiveness thresholds is shown in Figure 17. The uncertainty around the results observed in the cost-effectiveness planes above (see Figure 16) is reflected in the probability that the CPAP was cost-effective. In the EQ-5D analysis, the scatter of the simulations translated into a curve plateauing at 0.6. The SF-6D analysis indicated a greater probability that the CPAP was cost-effective across the range of thresholds. The probability that CPAP was cost-effective at the conventional thresholds used by NICE of £20,000 and £30,000 per QALY gained was 0.62 for EQ-5D QALYs and 0.95 and 0.97, respectively, for SF-6F QALYs. These results were consistent with those of the within-trial analysis.

FIGURE 17.

Cost-effectiveness acceptability curves for model-based base-case analysis.

Subgroup analyses

The cost-effectiveness results for the subgroup populations defined according to ESS score at baseline are shown in Table 46. In the less severe OSAS subgroup (ESS score of < 13 at baseline), the use of CPAP treatment reduced overall costs by £201. The effect on QALYs was dependent on the measure used: EQ-5D QALYs were reduced by 0.169 but SF-6D QALYs were increased by 0.181. Therefore, the ICER for the less severe OSAS subgroup using EQ-5D QALYs was £1189 per QALY gained. CPAP dominated in the SF-6D QALYs scenario. In the more severe OSAS subgroup, the use of CPAP treatment increased overall costs by £176 and both EQ-5D and SF-6D QALYs were increased. The ICERs were £360 per EQ-5D QALY gained and £967 per SF-6D QALY gained. These results were similar but in greater order of magnitude to the results of the within-trial analysis.

TABLE 46 - Cost-effectiveness results for subgroup populations

Treatment	Average costs	Average EQ-5D QALYs	Average SF-6D QALYs
Subgroup ESS score of < 13
CPAP	£16,019	7.823	7.861
BSC	£16,221	7.992	7.679
Incremental costs and QALYs
CPAP with BSC – BSC alone	–£201	–0.169	0.181
Subgroup ESS score of ≥ 13
CPAP	£16,396	8.483	7.860
BSC	£16,216	7.994	7.678
Incremental costs and QALYs
CPAP with BSC – BSC alone	£176	0.489	0.182

The cost-effectiveness planes for both population subgroups and by instrument used for obtaining QALYs, EQ-5D or SF-6D are shown in Figure 18. The same scale was used across the four plots to facilitate comparisons. In analysis of the less severe OSAS subgroup (ESS score of < 13 at baseline), using EQ-5D, there was considerable uncertainty on how CPAP affected costs and QALYs. In the analysis using SF-6D, some uncertainty around the effect on costs remained, but CPAP appeared to improve health outcomes across most of the simulations. In the more severe OSAS subgroup (ESS score of ≥ 13 at baseline), there was some degree of certainty that CPAP improved health outcomes, in terms of both EQ-5D and SF-6D. The uncertainty around the impact on costs remained.

FIGURE 18.

Cost-effectiveness plane for subgroup populations. (a) ESS score of < 13 EQ-5D; (b) ESS score of < 13 SF-6D; (c) ESS score of ≥ 13 EQ-5D; and (d) ESS score of ≥ 13 SF-6D.

The cost-effectiveness acceptability curve over a range of cost-effectiveness thresholds for both subgroups is shown in Figure 19. The curves were similar to the within-trial analysis. In the less severe OSAS subgroup, the probability that CPAP was cost-effective at £20,000 per QALY gained was 0.28 for the analysis with EQ-5D QALYs and 0.89 for the analysis with SF-6D QALYs. In the more severe OSAS subgroup, the probability that CPAP was cost-effective was 0.91 for the analysis with EQ-5D QALYs and 0.83 for the analysis with SF-6D QALYs.

FIGURE 19.

Cost-effectiveness acceptability curve for the subgroup (a) ESS score of < 13; and (b) ESS score of ≥ 13.

Sensitivity analysis

The results of the scenario analysis are shown in Table 47. See Appendix 3 for more detailed results including average costs and QALYs for each treatment, cost-effectiveness planes and cost-effectiveness acceptability curves. In scenario 1, the costs of CPAP treatment were increased to £710.16 because of more frequent replacement of the machine and consumables. Therefore, the difference in costs in the first year increased from –£35 to £474 (see Table 37). This cost difference was extrapolated over the patients’ lifetime in the model to £4785 (from –£329) in the base case. The probability that CPAP was cost-effective reduced from 0.62 to 0.20 when using EQ-5D QALYs and from 0.95 to 0.31 when using SF-6D QALYs. This was consistent with the findings of the within-trial analysis.

TABLE 47 - Cost-effectiveness results for the scenario analysis

Probability that CPAP is cost-effective presented for the most common subgroup (male patients who do not smoke and are not diabetic; 102 patients, 53.4% of patient population in the trial).

D indicates that CPAP dominates because it is associated with both lower costs and QALY gains.

Scenario	Difference in average costs	Difference in average EQ-5D QALYs	ICER/EQ-5D QALYs	Difference in average SF-6D QALYs	ICER/SF-6D QALYs	Probability that CPAP is cost-effective at £20,000/QALY gained
						EQ-5D QALYs	SF-6D QALYs
Base case	–£329	0.051	D	0.182	D	0.62	0.95
1. CPAP used for 1 year ( = £710.16)	£4785	0.051	£94,404	0.182	£26,599	0.20	0.31
2. Cardiovascular effects with effect of CPAP on costs and QALYs	–£327	0.022	D	0.139	D	0.58a	0.92a
3. Cardiovascular effects only	–£10	–0.024	£401	–0.023	£427	0.27a	0.28a

Scenario 2 included the effect of CPAP on cardiovascular risk predicted by the Framingham risk equations. The impact of including cardiovascular outcomes was very small. The difference in QALYs was reduced from 0.051 to 0.022 when using EQ-5D QALYs and from 0.182 to 0.139 when using SF-6D QALYs.

In scenario 3, only cardiovascular effects were considered; the cost and QALY difference observed at 12 months in PREDICT were not included. Over the patients’ lifetime, the differences in costs and QALYs were very small and uncertain (–£10; EQ-5D QALYs, –0.024; SF-6D QALYs, –0.023). This reflects the small effect of CPAP in diastolic BP and cholesterol observed in the trial.

The impact of the cost of the CPAP machine on the EQ-5D ICERs and in the probability that CPAP is cost-effective is shown in Figure 20. Note that the cost of the CPAP machine was annuitised over 7 years using a discount rate of 3.5%. Therefore, if the cost of the CPAP machine was doubled from £430 to £860, the annuitised cost of the machine increased from £70.32 to £140.65. This cost should be added to the annuitised cost of the humidifier (£26.98 multiplied by the proportion of people who received humidifier, 0.59 = £15.81) and the cost of consumables (£115.16) to give a final therapy cost of £271.62. In this analysis, the ICER for CPAP was above £20,000 per QALY gained and the probability that CPAP was cost-effective was below 0.50 for a machine cost of £1290 (three times the base-case cost of £430). These results indicated that a key driver of cost-effectiveness was whether or not patients returned the machine rather than the cost of the machine itself.

FIGURE 20.

Impact of the unit cost of the CPAP machine on the probability that the CPAP therapy is cost-effective.

Discussion of the assessment of the cost-effectiveness

Continuous positive airway pressure treatment appeared to be a cost-effective alternative to BSC alone. CPAP decreased costs by a small amount and improved health outcomes. However, the differences in costs and health outcomes between the treatment groups were small and uncertain. The subgroup analysis by level of sleepiness suggested that CPAP was more likely to be cost-effective in the more severe OSAS subgroup of patients. These results were consistent across the within-trial and model-based analyses.

The key drivers of cost-effectiveness were the costs of CPAP treatment and the benefits of CPAP on HRQoL. Some patients may fail to return the machine following treatment discontinuation. In these patients, the CPAP machine is a sunk cost, as it does not benefit the patient and cannot be reissued to another patient. If CPAP machines are not fully utilised over the lifetime of the machine, on average, the cost per patient of CPAP treatment increases. In the extreme scenario that all patients discontinued treatment after 1 year and did not return the machine, the probability that CPAP was cost-effective at £20,000 per QALY gained was 0.20 for EQ-5D QALYs and 0.30 for SF-6D QALYS. The implication for clinical practice is that patients should be encouraged to return the CPAP machine if they cease to use it.

The benefits in HRQoL were small when measured by either EQ-5D or SF-6D, but the SF-6D results may seem more pronounced as the difference was statistically significant. The smaller amount of variability with the SF-6D compared with the EQ-5D may be related to the data collection process and/or the instrument itself. The EQ-5D was collected every month through a sleep diary, which was filled in by the patient at home and sent by post. The SF-36, from which SF-6D is derived, was collected less often during a clinic visit at baseline and at 3 months and 12 months. This may have influenced the reporting of HRQoL by patients, although this would have been the case for both the CPAP and the BSC groups. In addition, the EQ-5D and SF-6D questionnaires have some differences. The SF-6D uses 11 questions from the SF-36 health status measure divided over six health domains: pain (six levels), mental health (five levels), physical functioning (six levels), social functioning (five levels), role limitations (four levels) and vitality (five levels). 121 The EQ-5D comprises five questions, with three levels each, on mobility, self-care, usual activities, pain, and anxiety and depression. 156 The dimensions in SF-6D, particularly vitality, may render the SF-6D more sensitive to changes in sleepiness and sleep quality. PREDICT is the first trial to collect SF-6D and EQ-5D following treatment of OSAS; however, some studies have compared SF-36 with EQ-5D and found that SF-36 was more sensitive to the impact of CPAP on HRQoL. 153,155

The impact of including cardiovascular outcomes on the cost-effectiveness of CPAP appeared to be negligible. This reflected the small difference in BP and cholesterol between patient groups observed in PREDICT. The increase in systolic BP would increase the risk of cardiovascular outcomes whereas the change in cholesterol would decrease it. Overall, and on balance, the change in these intermediate outcomes resulted in a small decrease in costs and QALYs. A limitation of this assessment was the incorporation of the impact of CPAP on cardiovascular outcomes. The Framingham risk equations have not been validated in a population older than 74 years of age. Therefore, the risk of cardiovascular outcomes may not have been correctly estimated. The direction of the bias is unclear and would have depended on whether the risk was under- or overestimated and on the effect of BP versus cholesterol on overall cardiovascular risk. CPAP would have been favoured if the combined effect of CPAP on BP and cholesterol decreased overall cardiovascular risk and the risk was overestimated or if the combined effect increased overall risk and risk was underestimated. Nonetheless, the impact of this bias was likely to be small, given the small change in costs and QALYs observed for the scenario with cardiovascular effects.

The cost-effectiveness analysis entailed three components to ensure appropriate consideration of all the relevant evidence: a systematic review of previous economic evaluations, a within-trial analysis using individual patient data collected in PREDICT and a model-based analysis incorporating the data collected in PREDICT with relevant external evidence. A systematic review on the clinical effectiveness of CPAP in older people confirmed that PREDICT was the sole source of evidence in this patient population. As it is considered that the CPAP treatment effects recorded in younger patients were not generalisable to older patients, this meant that PREDICT was the sole source of evidence of the treatment effects of CPAP included in the model-based analysis. Uncertainty around the cost-effectiveness results was quantified with a range of scenarios and sensitivity analyses in the within-trial and model-based analyses.

Areas of uncertainty included whether or not the savings in health-care costs are sustained over time and the differences between EQ-5D and SF-6D QALYs. The savings in health-care costs are small and uncertain; however, on average, these were enough to offset the cost of CPAP treatment. The savings may reflect the effect of CPAP on health-care resource use. Patients on CPAP may experience fewer adverse effects caused by their OSAS than patients on BSC alone and use the NHS less often as a result. Alternatively, the average saving per patient may be as result of chance and an artefact of the trial being underpowered to detect differences in costs.

The differences in the improvements in QALYs observed with EQ-5D and SF-6D were another area of uncertainty and a key driver of cost-effectiveness. As discussed, these differences may be related to the failure of the questionnaires in capturing the impact of sleepiness on quality of life or to the differences in the frequency and setting of the administration. Future research should explore the differences between EQ-5D and SF-6D as well as how different methods to collect HRQoL data impact on the results.

Main findings

Positive Airway Pressure in Older People: a randomised controlled trial was designed to assess the clinical efficacy of CPAP in older people with OSAS at 3 months and its cost-effectiveness over 12 months. CPAP improved sleepiness after 3 months by 2.1 points on the ESS compared with BSC. The beneficial effects were maintained at 12 months, and the magnitude of the improvements was similar to those seen in middle-aged patients with equivalent disease severity. 20 This subjective improvement in sleepiness was corroborated by the improvement in objective sleepiness at 3 months.

Continuous positive airway pressure also improved quality of life, both generic and disease-specific. CPAP-related improvement was statistically significant for the QALYs calculated with the SF-6D but not with the EQ-5D, equating to 1 week and 2 days, respectively. The CPAP group also accrued marginally lower health-care costs than BSC alone over 12 months. Overall, the economic benefit of CPAP was linked to the reduced health-care usage, offsetting the cost of the equipment, making it a cost-effective alternative to BSC for the treatment of OSAS in older people. The discrepancy between the two QALY measures could be a result of the EQ-5D being a less sensitive measure of the changes in health status attributed to sleepiness than the SF-36 (from which the SF-6D is derived). 12

Additional findings

Secondary outcomes related to cognitive function did not show any difference between the two groups despite reductions in sleepiness in the CPAP group. However, the baseline cognitive scores were often within the age-adjusted normative range, which may have resulted in a ceiling effect. Cognitive dysfunction is well recognised in middle-aged OSAS patients14,157 and is potentially linked to changes in brain morphology. 158 However, the impact of OSAS on cognitive function, separate from its effects on sleepiness and vigilance, is debated. 159,160 In older people with OSAS, the benefits of CPAP could be reduced because the capacity for neuronal recovery is less, owing to a combination of neurodegeneration associated with ageing and the life-long effects of OSAS. 57

The cardiovascular risk factors showed a small reduction in total cholesterol at 3 months, which was driven by a reduction in the LDL component. These findings are similar to those in a more severe and sleepier OSAS population, following 1 month of treatment with CPAP. 161 CPAP resulted in no improvement in BP. In the BSC group there was a small improvement in the systolic BP at 12 months. This finding echoes the results of a recent RCT of cardiovascular risk in mild asymptomatic patients162 in which CPAP usage, more specifically low usage, seemed to slightly raise BP. We speculate this could be a result of the BSC group following the BSC advice more closely.

Other secondary outcomes which showed no statistically significant difference between the two groups at 3 and 12 months were mood, frequency of nocturia and accidents. Interestingly, the patients in this trial had a relatively low prevalence of depression compared with a recent study. 8 We speculate that the expected lack of improvement in nocturia with CPAP may have been because of the multifactorial nature of this symptom in older people. 163

Comparison with other trials

A review of the clinical effectiveness of CPAP therapy in older people revealed three RCTs (from a possible 3560 titles) assessing the efficacy of CPAP treatment in OSAS patients with an average age of 60 years or older and the capacity to give informed consent. These studies included patients with cardiovascular conditions and compared CPAP therapy with sham CPAP88 or no CPAP. 89,90 None of the studies assessed daytime sleepiness or collected generic measures of health-care usage and they were not conducted in a secondary-care setting. The primary outcomes were left ventricular ejection fraction,88 baroreflex sensitivity,89 a number of neurological, quality-of-life and sleep-related effects and mortality. 90 Two studies reported BP at baseline and at follow-up;88,89 however, both of these studies focused on patients with chronic heart failure and their follow-up was short, at 3 months and 1 month, respectively. In the Egea et al. 88 study, no statistically significant differences were found in BP. In the Ruttanaumpawan et al. 89 study, the reduction in average systolic BP at 1 month was statistically significant but not the reduction in average diastolic BP. Overall, the results of these three studies are difficult to generalise to PREDICT, given their focus in patients with concomitant cardiovascular disease. Egea et al. 88 and Ruttanaumpawan et al. 89 included only patients with chronic heart failure and Parra et al. 90 included only patients who had had an ischaemic stroke.

Treatment adherence

The CPAP adherence was low at 3 and 12 months, which is likely to have diluted any treatment effect between the groups. 164 Indeed, exploratory analyses revealed that the treatment effect was larger in patients with more frequent CPAP use. The mean CPAP usage and the percentage of patients using CPAP at 12 months were similar to another UK RCT, albeit one of a shorter duration in patients with minimally symptomatic OSA. 162

The CPAP machines used in PREDICT were autoadjusting and, based on previous studies, it is unlikely that the autoadjusting machines were the cause of less frequent CPAP use. 165–167 On the other hand, we adopted a clinical approach to initiating and managing CPAP treatment, which may have resulted in a less frequent CPAP use, compared with a more intensive trial protocol. 15 However, with the approach that was adopted in this study, we have ensured that the PREDICT outcomes reflect clinical practice in the UK, which in turn has strengthened the validity and applicability of the health economic assessment. An additional factor that may have contributed to the low frequency of CPAP use in older people is reduced social support. We do not know how many patients were married, a factor that has been reported to be associated with increased CPAP compliance;168 however, just over half the patients slept alone.

Strengths and weaknesses

Positive Airway Pressure in Older People: a randomised controlled trial was designed as a pragmatic trial, recruiting older OSAS patients with comorbidity from geographically diverse areas throughout the UK. The findings are, therefore, relevant to what would be seen in clinical practice. Additionally, one of the unique elements of the trial design was the simultaneous cost-effectiveness evaluation, as well as the analysis of clinical effectiveness, measured over a relatively long time period. Finally, the high follow-up rates of patients attending at 3 and 12 months (over 80%) was impressive considering the duration of the trial.

A possible limitation of the trial was that sham CPAP was not used; therefore, as it was a physical device trial, the treatment allocation for the individual patients could not be concealed. The treatment allocation was concealed as far as possible from the member of the research team completing follow-up assessments. However, we reasoned that any placebo effect there might have been in the CPAP group would be expected to have disappeared by 12 months and that patients using CPAP might have expected an improvement but they would not have known by how much. Moreover, the results of the OSLER test and the observation of a therapeutic dose–response relationship between the treatment effect and CPAP use support a real effect.

Generalisability

With respect to generalisability, PREDICT did not focus on asymptomatic older people with OSA, and, although it could be argued that the patients studied had a relatively low mean ESS score at baseline, they were sufficiently symptomatic to seek treatment. At the other end of the disease spectrum, the exclusion of highly symptomatic OSAS patients in whom CPAP was considered mandatory is likely to have diminished the effect size. Patients with a higher baseline ESS score had a greater treatment effect in the exploratory analysis. Equally, the marginal improvement in cost-effectiveness was greater in the more symptomatic patients.

Continuous positive airway pressure prescribed for the symptom of excessive sleepiness due to OSAS in older people is more effective than BSC alone and no more expensive than BSC. The beneficial treatment effect is greater in patients with a higher ESS score prior to treatment and additionally in those the patients who used the CPAP treatment more.

Implications for health care

Clinical guidelines play an important role in improving health care for people with long-term conditions; however, it is well recognised they often fail to address the effects of comorbidity and polypharmacy. 169 There is also an inequality of research in older people with OSAS170 and PREDICT addresses this. The high-quality data from this trial will add to the knowledge of age-related changes, improve the generalisability of research findings and help inform best practice in the clinical management of a population that is growing older. The results of this study clearly support the use of CPAP for the treatment of OSAS in older people.

Implication for future research

Adherence to treatment is a recognised concern, particularly in multimorbid patients; despite this, few studies have investigated how adherence could be promoted. Suggested research priorities for future research are:

• To focus on the optimisation of CPAP delivery, especially in older patients. Can changes in health-care delivery improve adherence? Stratifying older patients with OSAS according to comorbidities and social factors to assess the clinical effectiveness and cost-effectiveness of CPAP treatment could further inform the delivery of care. Given uncertainty surrounding use of EQ-5D further work could be undertaken to assess quality-of-life measures this group of patients.

• To define patient-centred outcomes for treatment of OSAS in women and in ethnic groups, both of whom are currently under-represented in clinical trials.

• To identify potential biomarkers sleepiness and cognitive function that would enable early detection, which could be used in studied to inform when in the disease cycle treatment is needed to avert central nervous system sequalae.

• To explore the hypothesis that OSA in different groups may have different causes anatomically and physiologically, with different consequences.

This last point remains to be explored and may be fundamental to the understanding and targeting of treatment of the disorder.

Conclusion

• PREDICT has been the longest and most comprehensive controlled treatment trial in older OSAS patients to date, assessing both the therapeutic and economic impact of CPAP treatment.

• PREDICT has addressed the lack of research in older people with OSAS.

• The results of PREDICT clearly show that CPAP reduces symptoms of excessive daytime sleepiness in older patients with OSAS, as it does in middle-aged populations, and that these clinical benefits are associated with reduced health-care utilisation.

PREDICT investigators

Contribution of authors

Alison McMillan drafted the first and subsequent versions of this report, with supervision by Mary J Morrell and the other authors, who reviewed and approved the final submitted report.

Mary J Morrell and Renata L Riha were co-chief investigators for the trial, and designed the trial with Robert J Davies.

Andrew J Nunn and Daniel J Bratton undertook the statistical analyses. Rita Faria and Susan Griffin carried out the health economic analysis and wrote Chapter 4.

All authors participated in data interpretation.

Publications

McMillan A, Bratton DJ, Faria R, Laskawiec-Szkonter M, Griffin S, et al. Continuous positive airway pressure in older people with obstructive sleep apnoea syndrome (PREDICT): a 12-month, multicentre, randomised trial. Lancet Respiratory Medicine 2014;2:804–12.

Disclaimers

This report presents independent research funded by the National Institute for Health Research (NIHR). The views and opinions expressed by authors in this publication are those of the authors and do not necessarily reflect those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health. If there are verbatim quotations included in this publication the views and opinions expressed by the interviewees are those of the interviewees and do not necessarily reflect those of the authors, those of the NHS, the NIHR, NETSCC, the HTA programme or the Department of Health.

Trial steering committee

Professor Walter McNicholas (Chairperson), Professor Sir Neil Douglas and Dr Ian Smith (independent members), Daniel Bratton, Professor Robert Davies, Dr Mark Elliot, Mr Frank Govan, Dr Melissa Hack, Magda Laskawiec-Szkonter, Dr Alison McMillan, Professor Mary Morrell, Professor Andrew Nunn, Dr Justin Pepperell, Dr Renata Riha, Professor Mark Sculpher, Professor Anita Simonds, Professor John Stradling and Dr John Starr.

Independent data monitoring committee

Chairperson Professor Tim Peto, Professor John Gibson and Professor David Wright.

Trial management

Magda Laskawiec-Szkonter (ORTU).

Data entry

Jack Quaddy and Assunta Sabia (ORTU).

Research staff by centre

Systematic review of existing cost-effectiveness evidence

Results

Figure 21 presents a flow diagram summarising the identification and selection of studies. A total of 3560 unique records were identified from the systematic literature searches, of which seven subsequently met the inclusion criteria. 20,111,112,114–117 In addition, three additional studies were identified from the previous HTA on the subject. 113,118,119 Table 48 reports a brief summary of the studies. More detailed data extraction summary tables are presented at the end of this section.

FIGURE 21.

Flow diagram showing number of studies identified and included in the review of cost-effectiveness of CPAP.

TABLE 48 - Summary of cost-effectiveness studies assessing CPAP included in the systematic review

AHI, Apnoea–Hypopnoea Index.

First author (year)	Perspective	Country	Population	Comparators	Results
Chilcott (2000)118	Third-party payer	UK	Middle-age patients referred to sleep clinic	No treatment	ICER = £3200/QALY
Tousignant (2003)119	Third-party payer	Canada	Middle-aged patients (average age = 57 years)	No treatment	ICER between CAN$3397/QALY and CAN$9792/QALY; ICER without three outliers = CAN$18,637/QALY
Mar (2003)113	Third-party payer	Spain	50-year-old male with moderate to severe OSAS, defined by AHI > 30 events/hour and ESS score of > 10	No treatment	ICER 5-years = €7861/QALY; ICER lifetime = €4938/QALY
Ayas (2006)111	Third-party payer and societal	USA	Patients with moderate to severe OSAS (AHI ≥ 15 events/hour), aged between 25 and 54 years, drivers	No treatment	Third-party payer ICER = US$3354/QALY; societal ICER = US$314/QALY
Tan (2008)115	Third-party payer and societal	USA	Patients with moderate to severe OSAS, aged between 30 and 59 years, drivers	No treatment	Third-party payer ICER = CAN$3626/QALY; societal ICER = CAN$2979/QALY
Guest (2008)112	Third-party payer	UK	55-year-old patient with severe OSAS (AHI > 30 events/hour) and ESS score of ≥ 12	No treatment	CPAP dominates no treatment. Probablilty that CPAP is cost-efficient at £20,000/QALY is 0.99
McDaid (2009)20	Third-party payer	England and Wales	50-year-old male	Dental devices and no treatment	ICER = £3899/QALY; Probability cost-efficient at £20,000/QALY = 0.80
Sadatsafavi (2009)114	Third-party payer	USA	Patients with moderate to severe OSAS (AHI ≥ 15 events/hour), aged between 25 and 64 years	Dental devices and no treatment	ICER = US$27,540/QALY
Gander (2010)116	Societal	New Zealand	Patients between 30 and 59 years	No treatment	ICER = NZ$506.79/QALY
Pietzsch (2011)117	Third-party payer	USA	50-year-old male with moderate to severe OSAS	No treatment	ICER = US$16,172/QALY over 10 years and US$15,915/QALY for lifetime

All studies except Gander et al. 116 evaluated the cost-effectiveness of CPAP from a health-care or third-party payer perspective. Gander et al. 116 estimated the economic burden of OSAS from a societal perspective and, as a secondary analysis, presented an estimate of the ICER for treating OSAS compared with no treatment. Three studies were conducted in the UK,20,112,118 3 in the USA,111,114,117 two in Canada,115,119 one in New Zealand116 and one in Spain. 113

All studies compared CPAP with no treatment, and two also compared CPAP with dental devices. 20,114 Gander et al. 116 bundled CPAP together with dental devices and surgery under the treatment intervention. Pietzsch et al. 117 evaluated both the diagnostic options and CPAP treatment compared with no treatment. The time horizon employed was 5-years in six studies. 111,113–115,118,119 A lifetime horizon was used in three studies. 20,113,117 Alternative time horizons were used by three studies: 1 year in Gander et al.,116 10 years in Pietzsch et al. 117 and 14 years in Guest et al. 112

In six studies, the patient population consisted of a cohort of middle-age men with moderate to severe OSAS. 20,112,113,117–119 Three studies presented results for a patient cohort with moderate to severe OSAS, which were calculated from averaging the results from six patient subgroups defined by sex and age. In Ayas et al. 111 the cohort was aged between 25 and 54 years; in Tan et al. 115 the cohort was aged between 30 and 59 years; and in Sadatsafavi et al. 114 the cohort was aged between 25 and 64 years. Gander et al. 116 included patients aged between 30 and 59 years with any level of OSAS severity. Only McDaid et al. 20 and Pietzsch et al. 117 conducted subgroup analysis. Subgroup populations were defined by age, sex and disease severity by McDaid et al. 20 and by age, sex and disease prevalence by Pietzsch et al. 117 Note that no study looked specifically at patients aged 60 years and over.

In order to synthesise the available evidence and extrapolate over the chosen time horizon, most studies employed a decision-analytic model. A Markov model was employed in six studies. 20,111–115 Gander et al. 116 used a decision tree. Pietzsch et al. 117 used a decision tree for the diagnostics pathway and a Markov model for subsequent treatment decisions. McDaid et al. 20 used the structure first proposed by Mar et al. 113 of a Markov model with four health states [OSAS event-free, post stroke, post CHD and dead] and three events (CHD, stroke and RTAs). Guest et al. 112 used a similar model, but including a health state following the RTA event. Tan et al. 115 used the same model as Ayas et al.,111 which has an event-free OSAS state and six health states post RTA, corresponding to increasing levels of injury or death. Sadatsafavi et al. 114 used the same structure for the RTA component of the model but included post-MI and post-stroke health states. The Markov model in Pietzsch et al. 117 had five health states: event-free, hypertension, post MI, post stroke and dead. Patients were at risk of developing clinical hypertension and of experiencing an MI, stroke and RTA. The decision tree in Gander et al. 116 modelled the long-term consequences of increased costs owing to diabetes, cardiovascular disease, RTAs and work accidents in the untreated patients. Tousignant et al. 119 quality-adjusted the life expectancy of the patients included in the study, estimated based on Canadian lifetables, to calculate the cost-effectiveness of CPAP over a lifetime horizon. Chilcott et al. 118 synthesised data obtained from a review of the literature.

In line with the various model structures employed, treatment effectiveness was incorporated across the studies differently. A lower risk of RTAs was considered by eight studies. 20,111–117 Six studies included a reduction in cardiovascular events. 20,112–114,116,117 Gander et al. 116 included a reduction in the risk of diabetes and in the risk of work accidents. McDaid et al. 20 was the only study which incorporated a reduction in daytime sleepiness, as measured by ESS scores. Most studies based the effectiveness estimates in observational data on patients with moderate to severe OSAS. Seven studies assumed that CPAP reduces the risk of the various events to that of the general population and took the baseline risk of events in the untreated patients from observational studies. 111–117 McDaid et al. 20 obtained data on the effectiveness of CPAP in reducing ESS score and BP from a bivariate meta-analysis of RCTs and on the reduction in RTAs from observational evidence.

All studies included the benefits of CPAP in terms of HRQoL. Chilcott118 converted SF-36 data into health utility weights using the Brazier et al. 121 algorithm. Tousignant et al. 119 obtained health utility values before and after CPAP treatment directly from patients receiving CPAP with a standard gamble exercise, although the before treatment scores were valued retrospectively. Mar et al. 113 administered the EQ-5D instrument to patients before and after CPAP treatment; these values were also used by Guest et al. 112 Ayas et al. 111 and Tan et al. 115 sourced health utility weights from Chakravroty et al.,171 which evaluated HRQoL before and after treatment with a standard gamble exercise. McDaid et al. 20 developed a mapping function relating ESS score with EQ-5D and SF-6D, in which a reduction by 1 point in the ESS score was found to correspond to an improvement in health utility of 0.01. Sadatsafavi et al. 114 used this relationship in their model. Gander et al. 116 assumed a QALY gain of 5.4, based on data from the Sleep Alliance and Tousignant et al.,119 although it is unclear how this estimate was obtained. Pietzsch et al. 117 used data from the US Medical Expenditure Panel Survey to estimate that age- and sex-specific HRQoL in terms of EQ-5D was reduced by 16% in untreated OSAS patients and by 7% in OSAS patients treated with CPAP.

Across all studies, the results of the cost-effectiveness analysis led the authors to conclude that CPAP was a cost-effective treatment for patients with OSAS. The ICERs were lower than the typical willingness to pay for an additional QALY used in each of the countries considered, and in Guest et al. 112 CPAP was found to be associated with reduced costs and increased health benefits compered with no treatment. In general, the ICERs were robust to alternative assumptions on parameter inputs with the exception of the HRQoL gain from treatment with CPAP. HRQoL gain due to CPAP treatment had the greatest impact on the ICER for all studies that included this parameter in their sensitivity analysis. 20,111,113–115,117

Data extraction tables