Six early CPAP-usage behavioural patterns determine peak CPAP adherence and permit tailored intervention, in patients with obstructive sleep apnoea

Abstract

Background

High rates of non-adherence to continuous positive airway pressure (CPAP) in obstructive sleep apnoea hamper good clinical outcomes. Current recommendations assumes two behaviours (adherence and non-adherence) and days 7–90 follow-up post-CPAP initiation mitigates against non-adherence.

Objectives

To investigate associations between early CPAP-usage behaviours and (1) CPAP adherence at month 3 of treatment and (2) sleep centres’ treatment pathways (the procedures patients undergo that may affect barriers or facilitators of CPAP adherence).

Methods

We conducted growth mixture modelling (GMM) on retrospective data from 1000 patients at 5 UK sleep centres. Night 1 to month 3 telemonitored CPAP-usage data were downloaded from 200 patients per centre who started CPAP in 2019 (100) or 2020 (100). Adherence was defined using accepted criteria (mean CPAP-usage ≥4 hours/night for ≥70% of nights).

Results

GMM identified six distinct CPAP-usage behaviour patterns over month 1. In four (54% of patients), CPAP-usage increased or decreased, in two (remaining 46%), CPAP-usage/non-usage was consistent. 62% of the cohort were non-adherent by month 3, despite pathways following current recommendations. 98% of patients who were non-adherent by month 3 were already non-adherent by month 1. Regression analysis with a separate dataset demonstrated that early CPAP-usage behaviour explained 86% of the variance in CPAP non-adherence at month 3.

Conclusions

These data, supported by previous work, indicate that recommended day 30–90 follow-up is too late to prevent CPAP non-adherence. Determining CPAP-usage behavioural pattern in week 2 identifies risk of CPAP non-adherence at month 3 and permits the possibility of tailored interventions.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Current (American Thoracic Society, ATS) recommendations for monitoring patients on continuous positive airway pressure (CPAP), referred to in the recent International Consensus Statement on obstructive sleep apnoea, have not been formally evaluated and have failed to improve levels of CPAP adherence since its introduction in 2013. They centre around a binary classification of patients being adherent or non-adherent; however, recent studies indicate that additional behaviour patterns occur, although these vary between studies and their prognostic significance and validity have not been demonstrated. Therefore, in this study, we sought to identify early patterns of behaviour with prognostic and clinical significance.

WHAT THIS STUDY ADDS

• This demonstrates that patients can be stratified into one of six groups based on their behaviour, specifically trajectory of CPAP use between nights 1 and 14, and the particular behavioural pattern was strongly associated with risk of non-adherence at month 3, with the risk of non-adherence ranging from 100% to 7% risk. Second, we identified that, although within current recommendations, first follow-up at day 30 appears too late to mitigate against non-adherence because 98% of the patients who implement treatment but are non-adherent by month 3 are non-adherent by day 30.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• The current data interpreted in the context of previous prospective studies in patients starting CPAP and the literature on behaviour change, indicate that first follow-up in patients starting CPAP should ideally be within the first 3 weeks. Identifying early behavioural pattern and using this to guide patient management (a simple tailored intervention) in addition to adherence status is also indicated from our data, which confirms prospective data from others. This should be confirmed by prospective studies validating these early behavioural patterns with CPAP adherence and identifying modifiable determinants of these early behavioural patterns.

Introduction

Obstructive sleep apnoea (OSA) is a major global health burden. It has an estimated prevalence of one billion, a twelfth of the world’s population,¹ and is a multisystem disorder increasing the risks of type 2 diabetes, hypertension, heart attacks, strokes, cancer², and the risk of road traffic accidents³ and death.²

Even once on treatment, an obstacle to good clinical outcomes exists. The most efficacious treatment, aside from cures such as weight loss where appropriate, is continuous positive airway pressure (CPAP).⁴ However, many patients do not use CPAP enough to benefit and CPAP adherence rates have remained unchanged over the last thirty years.^{5 6} In trials, 17%–71% of patients are not adherent to CPAP using the widely accepted criteria of a minimum of 4 hours nightly use for at least 70% of nights.^{6 7} Patients who do not meet this threshold have an increased risk of cardiovascular events and hospital admissions,8,10 while 4 hours use is also required to reduce the symptoms¹¹ and mortality associated with OSA.¹⁰

Current recommendations for practice advocate early follow-up between days 7 and 90, ideally after 1 week, 4–6 weeks and 12 weeks.¹² These align with early, small studies describing adherent and non-adherent behaviours arising between nights 1 and 4 of treatment and continuing until month 3, the first of which recommended week 1 follow-up to mitigate against non-adherence.^{13 14} However, these studies predetermined their results by dividing their samples into two based on frequency of CPAP use.^{13 14} Furthermore, the efficacy of current recommendations with respect to CPAP adherence compared with previous recommendations of day 31–90 follow-up has not been evaluated. In addition, more recent studies have described more than two patterns of CPAP-usage behaviour over the first 3–12 months of CPAP treatment,15,19 indicating that the premise that recommendations are based on is incorrect. However, the behaviours described have varied between studies and their prognostic, hence clinical and significance have not been determined.

Our primary objective was, therefore, to investigate early patterns of behaviour in relation to CPAP use and determine associations with CPAP adherence defined using accepted criteria, at month 3 of treatment. The reason for doing this was CPAP adherence defined this way impacts long-term prognosis (month 3 adherence is a predictor of long-term adherence 2–4 years after treatment initiation²⁰ and adherence to CPAP predicts hospitalisations, development of OSA-related conditions, etc8,10). Therefore, if these early CPAP-usage behaviours associate with month 3 adherence, they have potential to be associated with long-term prognosis; if they can be modified to improve CPAP use by behavioural interventions, they can potentially improve long-term prognosis. Our secondary objective was to evaluate associations between early CPAP-usage behaviours and the sleep centre’s treatment pathway, that is, the procedures that patients undergo that may affect barriers or facilitators of CPAP adherence.²¹ This would determine ‘best pathways for practice’ for dissemination.

Methods

Retrospective study design

Recruitment and ethics

22 UK sleep centres were approached: the only eligibility criterion was an existing telemonitoring programme for all patients in both 2019 and 2020. Five centres were eligible, all enrolled and are referred to by code name. See online supplemental file 1 for details of ethical approval.

Figure 1 shows each centre’s treatment pathway per year; there were nine pathways as all centres except one changed pathway when the pandemic started. All pathways delivered first follow-up between nights 1 and 90, 80% delivered this between nights 1 and 30. Median follow-ups (IQR) between nights 1 and 90 was 3 (3).

Figure 1. Schematic of sleep centres’ treatment pathways in 2019 and 2020. Consults refer to consultation with clinician or sleep technician and were in-person unless otherwise specified, for example, as remote which refers to a video or telephone consultation. Training refers to training in setting up the CPAP machine, fitting the CPAP mask and troubleshooting potential issues with the CPAP machine. Technician and physician refer to specialist sleep technicians, who have technical expertise with managing CPAP and sleep study equipment, and sleep physicians, respectively. Each sleep centre had a defined pathway of procedures that each patient’s treatment followed, the centre’s ‘treatment pathway’. There were key mandatory components common to all centres, although the timing, frequency and mode that these components were delivered varied by centres and by year for all centres, apart from DH. Pathway components shaded in white were those that remained the same within a centre between 2019 and 2020, while those in pink were different between the 2 years, because of a change at the start of the COVID-19 pandemic. In addition to the pathways shown, patients could also call the sleep department as required at all centres to access advice and equipment from technicians. The information in this figure can be used to interpret figures2 4; the effect of different pathways and individual pathway components on CPAP adherence and early patterns of CPAP-usage behaviours, respectively. CPAP, continuous positive airway pressure; H, home; IP, inpatient; Tech, technician.

Sample size calculation

We initially set out to compare CPAP adherence at each centre between 2019 and 2020. We used UK-wide telemonitoring data demonstrating CPAP adherence, defined according to the accepted criteria,⁷ had fallen by 18% during the pandemic.²² To detect an 18% difference, with 80% power, α<0.05, n=92 was required, per centre, per year but a recruitment target of n=100 was set. This generated an ample n=1000 sample for the modelling which later became the primary analysis.

Data collection and analysis

Per centre, CPAP-usage datafiles were downloaded from 100 consecutive OSA patients who started fixed or variable pressure CPAP (inclusion: OSA, all severities, defined according to the American Association of Sleep Medicine guidelines, exclusion: prior CPAP use) from 1 April 2019 (prepandemic), 100 from 1 September 2020 (peripandemic, after UK’s first wave). CPAP use on nights 1 (night of CPAP collection and start of telemonitoring), 2 and 3, mean nightly use and total nights of use by nights 7 and 14, month 1 (30 days) and month 3 (90 days) were collected. CPAP adherence definition: mean use ≥4 hours/night for ≥70% of total nights between night 1 and time point, except between nights 1 and 3, median was used (data skewed). Clinical data including age, gender, body mass index and sleep study results were collected from electronic and hard copy hospital records. All data were collected by a single investigator (JD) to ensure standardisation across data, with data verification conducted by two other investigators (AS and LJL-C) (see online supplemental material for further details).

Statistical analysis

Analysis was performed using R (V.4.1.3). Details are in online supplemental material.

Growth mixture modelling

Once the functional form of the growth mixture modelling (GMM) was determined as a quadratic term and random intercepts and slopes, models with 1–7 latent classes were fitted. The final model was chosen using the Bayesian information criterion, entropy, minimum class size, average posterior probabilities of each class and clinical plausibility. Independent associations between patient characteristics and treatment pathway (centre interacting with year), and GMM classes, were investigated using multinomial logistic regression (LR). Characteristics and determinants of each class were estimated using the three-step approach with the pseudo-class method.²³ Latent class growth modelling was performed for a sensitivity analysis, as recommended by the Guidelines for Reporting on Latent Trajectory Studies (see online supplemental material).

Logistic regression

Associations between patient characteristics, including interactions based on previous studies and between centre and year, with CPAP adherence at 3 months were explored.

Multiple imputation

Multiple imputation (MI) was used to account for missing values. There were 115 (12%) body mass index and 41 (4%) Epworth Sleepiness Scale scores missing; there were no missing CPAP-usage data points. Results of the complete case analysis using MI or with the missing cases excluded did not differ (see online supplemental tables S1–S5).

Results

Sleep centres were spread across the UK: GSTT and WYH in large cities, MSH, CH and DH in smaller cities. Patients’ characteristics are shown in online supplemental table S6.

CPAP adherence at month 3 of treatment is poor

Three months after starting treatment, CPAP adherence across the centres averaged 34% in 2019 with significant variability between centres (range 29%–50%, p=0.025, online supplemental tables S7 and S8 and figure S1). In 2020, this averaged 42% (range 27%–51%, p=0.004), which was not significantly different to 2019 (figure 2F, p=0.24).

Figure 2. CPAP adherence rates at UK sleep centres. Plots (A–E) show the CPAP adherence rate (% of patients adherent to CPAP according to the accepted definition) over time at each of the five UK centres and (F) shows this when the data from all five centres is pooled. N=100 for 2019 and for 2020 in A–E and, therefore, n=500 for each year in (F). Differences between percentages of patients who were treatment-adherent within the centres at different time points, or between centres at the same time point, were compared using a χ² test. Differences that were statistically significant are marked with an asterisk: *p<0.05, **p<0.01, ***p<0.0001 (exact value). CPAP, continuous positive airway pressure.

There were six different patterns of early behaviour relating to CPAP use, with >50% of patients changing their CPAP use over the first month

The best fitting GMM had six classes, as shown in figure 3. Four of the six classes (54% of patients) increased or decreased their CPAP use over the first month, particularly over the first 2 weeks. The other two (remaining 46%) had consistent good or non-use, as shown in table 1. In the very early period including week 1, patients with different behavioural patterns could not be differentiated by their adherence status. For example, ‘good’, ‘downward drifting’ and many ‘falling users’ would be classed adherent as they had used CPAP for more than 4 hours a night for more than 70% of nights by night 7, while ‘non’, ‘upward drifting’ and many ‘rising’ users would be classed non-adherent by failing to meet these criteria. However, each of these behavioural patterns was associated with a very different outcome, as the likelihood/risk of non-adherence to CPAP at 3 months ranged from 7% to 100% in the six groups, spanning 21%–100% in the three non-adherent groups and 7%–99% in the three adherent groups. Furthermore, 98% of the patients who started treatment but were non-adherent by month 3, were already non-adherent by the end of month 1, resulting from a steady rise in non-adherers from night 3 to night 30 (see table 2). This was despite 80% of pathways delivering first follow-up within month 1. Between months 1 and 3, there was little change in CPAP use in the cohort.

Figure 3. Six class model of patterns of CPAP-usage behaviour in OSA patients starting treatment from 1000-patient multicentre UK cohort. 95% CIs around each class are not shown but are described in table 1. CPAP adherence was defined by mean CPAP use of ≥4 hours/night for ≥70% of total nights in the period between the time point and night 1 of treatment, except at night 3 where this was defined by median hours use between nights 1 and 3 being ≥4 hours as data were not normally distributed. Night 1 was defined as that on the day of CPAP collection. Class sizes and percentage of each class that was treatment-adherent 3 months after starting CPAP initiation were calculated using the three-step approach with the pseudo-class method. CPAP, continuous positive airway pressure; GMM, growth mixture modelling; OSA, obstructive sleep apnoea.

Table 1. CPAP use per night for the six groups demonstrating different early patterns of behaviour.

Class name	CPAP use per night (hours)
	Night 3	Night 7	Night 14	Month 1	Month 3
Good use	7.34(7.14–7.54)	7.24(7.04–7.43)	7.08(6.88–7.28)	6.80(6.57–7.03)	6.72(6.48–6.96)
Falling use	6.30(5.61–6.99)	5.47(4.85–6.08)	4.15(3.62–4.68)	1.82(1.33–2.31)	1.51(0.99–2.02)
Downward drift	4.57(4.07–5.08)	4.33(3.84–4.81)	3.94(3.47–4.40)	3.24(2.75–3.74)	3.04(2.50–3.58)
Upward drift	2.19(1.78–2.59)	2.54(2.16–2.91)	3.09(2.74–3.44)	4.04(3.63–4.45)	3.81(3.42–4.21)
Rising use	1.29(0.93–1.66)	2.23(1.89–2.58)	3.71(3.37–4.05)	6.30(5.90–6.69)	6.12(5.72–6.53)
No use	0.49(0.35–06.2)	0.49(0.37–0.62)	0.50(0.38–0.62)	0.51(0.37–0.65)	0.41(0.28–0.55)

Night refers to night of treatment, with Nnight 1 being the night of CPAP collection and the start of telemonitoring. Results are displayed as means and 95% CIsconfidence intervals for each of the classes identified in the six-group GMM.

CPAP, continuous positive airway pressure; GMM, growth mixture modelling.

Table 2. Percentage of each group demonstrating a distinct early pattern of behaviour that was non-adherent to CPAP at each time point.

Class name	Night 3	Night 7	Night 14	Month 1	Month 3
Good use	1	4	4	4	7
Falling use	1	43	78	97	99
Downward drift	32	60	71	77	83
Upward drift	80	82	75	74	71
Rising use	94	85	56	25	21
No use	97	100	100	100	100

Proportions were estimated using the three-step approach with the pseudo-class method.

CPAP, continuous positive airway pressure.

Associates of early CPAP-usage behaviour patterns

These are shown in figure 4 and online supplemental tables S9 and S10, which depict the multinomial regression results.

Figure 4. Multinomial regression results illustrating associates of early patterns of CPAP-usage behaviour. GSTT, WYH, CH, MSH and DH are code names for sleep centres. Adjusted ORs with 95% CIs using the ‘good use class’ as one reference category and the second reference category highlighted in brackets. ORs>1 indicate the characteristic is more likely to be associated with the highlighted GMM class than the ‘good use’ class, ORs<1 indicate it is less likely. For example, CH had fewer non-users than ‘good users’ in 2020 compared with 2019. The ‘falling use’ and ‘rising use’ classes were not included in the analysis due to their small class sizes. BMI, body mass index; CPAP, continuous positive airway pressure; ESS, Epworth Sleepiness Scale (a normal score is ≤10/24); GMM, growth mixture model; OSA, obstructive sleep apnoea (mild, moderate or severe defined as apnoea/hypopnoea index or oxygen desaturation index of 4% being 5–14, 15–30 or >30 per hour, respectively).

Severity of OSA, age interacting with gender and centre interacting with year were associates of the early behavioural patterns. Patients with severe OSA had a lower ‘non-user’ to ‘good user’ ratio than patients with mild OSA (adjusted OR (AOR) 0.52 (95% CI 0.29 to 0.91)). Females aged over 65 years had a higher ‘non-user’ to ‘good user’ and ‘upward drifter’ to ‘good user’ ratio than females less than 45 years (AOR 3.48 (95% CI 1.20 to 10.11) and 3.40 (95% CI 1.03 to 11.20), respectively), and a higher ‘non-user’ to ‘good user’ ratio than equivalent age males (AOR 2.72 (95% CI 1.09 to 6.78), online supplemental table S9). However, these associations were not strong enough for these early behavioural patterns to be differentiated by demographics or disease factors, as shown in table 3, just as month 3 CPAP adherence was not strongly associated with these factors (see later). DH had the lowest ratio of ‘non-users’ to ‘good users’ (mean of 1:2.6 between years) consistent with its pathway providing the most follow-up consultations in the first month (three by day 14 and four by day 30). The prevalence of the four main early behavioural patterns remained unchanged between years at DH (see online supplemental table S11) consistent with an unchanged treatment pathway.

Table 3. Characteristics of the six groups demonstrating different early patterns of CPAP-usage behaviour.

Variable	Category	Total sample	Good use	No use	Rising use	Upward drift	Downward drift	Falling use
All participants (%)		100	27.1	19.2	5.0	16.3	26.1	6.3
Age (years)		53 (18)	52 (18)	54 (18)	60 (16)	53 (18)	51 (17)	51 (19)
Gender (%)
	Male	67.4	69.2	69.6	65.7	66.8	65.6	62.6
	Female	32.6	30.8	30.4	34.3	33.2	34.4	37.4
BMI (kg/m2)		34 (11)	34 (10)	33(10)	33 (11)	33 (11)	35 (11)	34 (11)
	Missing	11.5	10.3	13.6	6.5	12.4	12.2	9.7
ESS (/24)		12 (8)	13 (7)	12 (8)	11 (6)	12 (8)	13 (8)	11 (7)
	Missing	4.1	2.9	4.7	1.2	5.2	5.0	3.3
OSA severity (%)
	Mild	23.3	21.9	28.9	24.3	18.7	22.6	27.1
	Moderate	36.3	33.9	34.2	36.9	42.2	34.3	48.0
	Severe	40.4	44.3	36.9	38.8	39.2	43.1	24.8

Displayed as median (IQR) or percentages. Class characteristics were estimated using the three-step approach with the pseudo-class method due to the probabilistic nature of class membership in growth mixture modelling. As a result, absolute numbers of participants are not available because each participant’s class membership is drawn from their posterior probability matrix several times. The percentage of each class with missing BMI or ESS is reported because missing information was not imputed in this analysis.

BMI, body mass index; CPAP, continuous positive airway pressure; ESS, Epworth Sleepiness Scale; OSA, obstructive sleep apnoea.

Early patterns of CPAP-usage behaviour explained 86% of the variance in non-adherence to CPAP at month 3 between individuals

LR was performed with demographics, OSA severity, centre and year with the 1000-patient dataset to identify potential determinants of month 3 CPAP adherence status. The model was poorly fitting with a McFadden’s R² of 0.038 (0.2–0.4 indicating excellent fitting models,²⁴ see online supplemental tables S12–S14).

We then re-ran the LR, using GMM class (derived from nights 1 to 14 data) as an independent variable, using an additional independent dataset of 200 patients from one centre in 2021 (see online supplemental tables S15 and S16). The purpose was to confirm associations between GMM class and CPAP adherence and identify whether GMM class improved the LR model, which it did. The full model had an excellent fit as shown in table 4. GMM class contributed 86% (McFadden’s R² of 0.408) to the model, alone explaining most of the variance in month 3 CPAP adherence between individuals.

Table 4. Logistic regression model of predictors of adherence to CPAP at month 3 using additional patient dataset.

Regression model	Null model PLL	Full model PLL	McFadden’s R2	Relative % change in R2	Penalised likelihood ratio test
Full model	−116	−61	0.475
Full model without gender×age	−120	−67	0.445	6	0.004
Full model without BMI	−119	−64	0.459	3	0.003
Full model without ESS	−120	−64	0.464	2	0.348
Full model without OSA severity	−118	−63	0.463	2	<0.001
Full model without treatment pathway	−117	−62	0.472	1	1.000
Full model without GMM class	−120	−112	0.067	86	<0.001

Table shows McFadden’s R2 for the full model minus each covariate. McFadden’s R2=1–(LL_full/LL_null). Note that unlike in maximum likelihood analysis, where the null model PLL does not change for the same data, the null penalised likelihood depends on the penalty (Jeffreys prior) which itself depends on the scope of variables in the model. The GMM class was derived from a separate cohort of 200 patients that was not used to derive the GMM classification, and the GMM class was calculated using the first 2 weeks of CPAP-usage data not the first month; 2 weeks data were sufficient to distinguish the six groups and to predict long-term CPAP adherence. Values in bold indicate statistical significance.

CPAP, continuous positive airway pressure; ESS, Epworth Sleepiness Scale; GMM, growth mixture modelling; OSA, obstructive sleep apnoea; PLL, penalised partial likelihood.

Discussion

62% of the patient cohort across five UK centres were non-adherent to CPAP, at the 3-month time point when non-adherence should be at its lowest. Patients exhibited six distinct patterns of early CPAP use, which independently and reliably associated with their risk of non-adherence to CPAP at 3 months, with the risk ranging from 100% to 7%. More than half of patients changed their early CPAP use over the first month, particularly over the first 2 weeks. By the end of month 1, 98% of the patients who started treatment but were non-adherers by month 3 were already non-adherent, with a steady rise in non-adherent patients between weeks 1 to 4. Patients’ pattern of early behaviour in relation to their CPAP use provides insight that adherence status cannot, including information to guide the health professional to tailor the conversation at the follow-up consultation. Current recommendations for practice need updating in light of this data.

Our CPAP adherence rate in 2020 matches another group’s report (41% at 1 month following CPAP initiation),²⁵ validating our results. However, our results are lower than previous clinical studies. This can be explained by methodological differences. Missing data from non-adherent patients who leave their CPAP switched off or have poor WIFI connection is common; an identical signature registers on the CPAP download (see online supplemental figure S2). Without linked clinical information, as available to us to include these patients, these patients must be excluded. This would result in potentially completely non-adherent patients being excluded in anonymised data studies. Other studies such as Buyse et al²⁶ had substantial patient loss to follow-up; these patients are more than twice as likely to be non-adherent than adherent patients.²⁷ Big data studies such as Cistulli et al²⁸ also used the threshold for financial reimbursement for CPAP in the USA where the 70% of nights only need apply to a single 30-day period within the 3 months not the full 3 months.

A second, important, validation is that our results are consistent with what is known about treatment adherence, behaviour change (which becoming adherent to treatment is a form of) and illness behaviour, for which there are established theories, frameworks and a huge body of literature. These describe a continuously evolving process starting with patients deciding whether to initiate treatment, which is strongly influenced by their beliefs about their illness, the treatment, and themselves, for example, their self-efficacy (confidence) in using the treatment. After starting treatment, if a patient’s experience and outcomes of treatment differ from their expectations, these beliefs are revised, and if so, their use of treatment will change,^{29 30} just as seen in the 54% of patients in the four ‘changing CPAP use’ groups. Consistent with this, it has, for example, been demonstrated in prospective studies that self-efficacy in using CPAP and other relevant beliefs in relation to OSA and CPAP change over the first week of treatment. Self-efficacy at 1-week post-CPAP initiation (but not pre-CPAP initiation) associated with CPAP adherence at 1 month, indicating that first follow-up after month 1 will not mitigate against non-adherence.^{31 32}

Our findings also fit well with other studies’ descriptions of multiple CPAP-usage behavioural patterns in patients starting CPAP, and with other adherence behaviours, for example, where there are multiple medication non-adherence behaviours described.³³ From a case series of 71 patients observed over 1 year, Aloia et al described seven different behaviour patterns (‘good users’, ‘non-users’, ‘slow improvers’, ‘slow decliners’, ‘variable users’, ‘occasional attempters’ and ‘early drop-outs’), where the first four are similar to our descriptions.¹⁷ Yi et al conducted a growth mixture modelling analysis on CPAP-usage data at week 1, month 1 and month 3 from 301 patients.¹⁵ They described three behaviours: ‘adherers’ and ‘non-adherers’ and ‘improvers’, the latter starting with poor use that improves over 3 months. By month 3, 54% of the ‘adherers’, 71% of the ‘improvers’ and 0% of the ‘non-adherers’ were adherent according to the accepted criteria. Their ‘improver’ group was similarly sized (19% of their cohort) to our two ‘increasing use’ groups together (21% in total). Consistent with our findings, demographic and disease factors did not differentiate their groups, but behavioural factors, coping styles and personality traits did.¹⁵ Teuling et al used non-parametric mixture modelling with a big data cohort of 10 000 patients to investigate different aspects of CPAP usage: attempts to use CPAP plus duration and continuation of CPAP use over the first 3 months of treatment, in patients using CPAP beyond 3 months.¹⁶ Nine different groups were identified and motivation, one of the three factors necessary for any behaviour change including treatment adherence,²⁸ marked continued CPAP adherence.¹⁶ Other studies using less rigorous modelling techniques and smaller samples, of less than 250 patients, have also identified multiple behaviours relating to CPAP use. For example, Wohlgemuth et al described ‘attempters’, ‘adherers’ and ‘non-adherers’, with ‘non-adherers’ differentiated from ‘attempters’ and ‘adherers’ by lower self-efficacy in using CPAP.³⁴ Bros et al described four groups: ‘regular adherents’, ‘regular non-adherents’, ‘persistent non-adherents’ and ‘non-persistent non-adherents’ from a longitudinal cluster analysis of CPAP usage over 1 year in 204 patients.¹⁸ The limitation of all these studies is that it was not determined whether outcomes were/were likely to be different between different typologies, for example, between ‘attempters’ and ‘non-adherers’ or ‘regular non-adherent’ and ‘persistent non-adherents’; therefore, whether they have any clinical relevance is unknown. However, Aloia et al made an important observation that our results concur with and that is adherence status calculated in the conventional way is poorly discriminatory compared with longitudinal analysis of CPAP-usage data. They state ‘if they had analysed the data in a more traditional way…using measures of central tendency… several of their groups would be combined. The implication is that identifying group membership early in the treatment process may help tailor treatment approaches to different groups of patients, thus enabling them to achieve the goal of consistent and regular use of treatment.’¹⁷

Implications of our findings

From these data, supported by previous data demonstrating that health beliefs change within a week of starting CPAP to strongly associate with non-adherence to CPAP at month 1,^{31 32} and behavioural science, more broadly,^{29 30} first follow-up cannot occur after month 1. Therefore, current recommendations for days 7–90 follow-up require revision; we would suggest a change to days 7–21. We would also recommend an addition to the recommendations where early behaviour is determined at/around day 15 and (1) the patient’s risk of non-adherence at month 3 (without intervention) is calculated and (b) the early behaviour is used as a guide to tailor the conversation at the follow-up consultation. For example, patients whose use CPAP has fallen dramatically since the first few nights (‘falling users’) versus not implementing treatment at all (‘non-users’) versus tentatively increasing use from a low baseline (‘upward drifting users’) would likely benefit from consultations tailored to their behaviour.

Future work and impact of this work on research

This new classification offers new opportunities for research that will foster development of tailored interventions to improve adherence to CPAP, based around the six early patterns of behaviour. Tailored interventions for treatment adherence are more effective than non-tailored interventions.³⁵ This will require longitudinal studies confirming our findings here and exploring modifiable determinants of each behavioural group, particularly beliefs about illness, treatment and the individual, given their association with CPAP adherence at 3 months.³⁶ Understanding CPAP’s therapeutic potential and the exact ‘dose’ for preventing negative sequelae of OSA have been hampered by CPAP trials failing to achieve the required power because of unpredictably high rates of non-adherence. Extending run-in periods from 1 to 2 weeks to 4 weeks should exclude all early non-adherers to improve this.

Limitations of study

As a retrospective analysis, some data were unavailable, for example, on demographics such as ethnicity, socioeconomic class and treatment side effects which have been associated with CPAP adherence, although inconsistently.⁶ Fortunately, when they have been associated with CPAP adherence, the variance in CPAP adherence accounted for by these factors has been relatively small, for example, between 6% and 12% for ethnicity, socioeconomic class, employment status and education combined^{6 37} particularly when compared with 40% for behavioural factors.³¹ We were confined to the nine treatment pathways. The natural experiment was helpful because one could broadly assume that centres’ staff and culture was similar in both years. However, the potential confounding effect of the COVID-19 pandemic must be considered when interpreting differences between years. We were reassured by DH’s stable adherence rate and prevalence of early behavioural groups as this suggested that major systemic pandemic effects on the population affecting CPAP adherence were not present during our data collection period. This was consistent with a study reporting that although the mental health of a proportion of the UK population deteriorated during the first wave, for the majority this had normalised by the time we started collecting data.³⁸ We also used data from the UK only. However, algorithms predicting long-term adherers using the first 2 weeks of CPAP—usaage and data on health beliefs predicting month 3 adherence over the first month of treatment were derived from non-UK cohorts^{31 39 40} and support the generalisability of our model to other populations. The sample size was limited for between-centres comparisons in less prevalent ‘rising’ and ‘falling use’ behaviours; we, therefore, did not report this. Our study’s strengths include rigorous data collection and statistical methods, including several sensitivity analyses.

To conclude, untreated OSA is responsible for a huge global burden of chronic ill health and mortality and its prevalence is increasing. CPAP is the most efficacious treatment and is likely to remain so, given its mechanism of action, aside from curative treatments. Adherence to CPAP in a sample of UK centres is extremely low: just 38% 3 months after CPAP is started, consistent with previous international trials. We describe six early CPAP-usage behaviours that robustly associate with the later likelihood of being non-adherent to CPAP and provide insight beyond adherence status that enables tailored management. We also describe a time course of non-adherence that necessitates early patient follow-up within weeks 1–3. Both are not in line with current recommendations for practice. The new behavioural classification also holds promise for research where interventions to improve CPAP adherence can be tailored to behaviour.

Data availability statement

Data are available on reasonable request.