CPAP adherence has been the same problem in sleep medicine for thirty years. The technology works when patients use it, and a non-trivial fraction of patients don’t use it enough. The published adherence numbers across most service settings have hovered in the 50-70 percent range depending on how generously you define “adherent” - which has remained roughly stable despite a generation of improvements in machines, masks, and clinical workflows.

The 2024 and 2025 literature has produced some useful refinements to this picture, and a few areas where the data suggests we should be doing things differently in 2026.

The mask interface choice still matters more than we sometimes admit

A consistent finding across recent adherence studies is the outsized role of the patient’s interface experience in the first two to four weeks of therapy. Patients who are switched from a poorly tolerated initial mask to a better-fitting alternative within the first month show meaningfully higher long-term adherence than those who persist on the initial mask for longer.

The clinical implication is that the conventional “give it eight to twelve weeks before changing the mask” approach is probably wrong. Earlier intervention on mask discomfort or leak issues looks like better practice. Several services have been reporting adherence improvements of five to ten percentage points after restructuring their early-therapy review schedules to enable mask changes at the four-week mark rather than the eight-to-twelve-week mark.

The trade-off is that earlier mask switching costs more in clinical contact time and equipment. The economic case is generally favourable in services where the alternative is a cohort of patients who quietly stop using their machines, but it requires upfront resourcing that not every service can provide.

Telemonitoring has matured but isn’t a silver bullet

The cloud-connected machine data has been transformative for adherence visibility. Clinicians can now see usage patterns, leak data, and residual events for the entire patient panel in close to real time. The question is what to do with this information.

The recent literature suggests that telemonitoring data alone, without an accompanying intervention model, produces modest adherence benefits. The bigger gains come from telemonitoring combined with structured outreach when the data shows specific patterns - declining usage, increasing leak, or pattern interruption. Several large services have reported that automated alerting tied to clinician-led intervention protocols has produced meaningful and sustained adherence gains.

What hasn’t worked as well as hoped is fully automated patient feedback. Studies of app-based usage feedback to patients have shown small effects that fade over time. Patients seem to need a human interaction to translate the data into behaviour change. The pure-tech solutions that promised to remove the clinician from the adherence loop have generally underperformed.

The CBT-i adjacency

A development worth flagging is the growing evidence that addressing co-existing insomnia symptoms in CPAP-naive OSA patients improves CPAP adherence. The mechanism is intuitive - patients with significant sleep onset or maintenance insomnia have a worse experience trying to initiate CPAP therapy, which colours their early adherence.

Several recent trials have looked at delivering brief CBT-i interventions in parallel with CPAP initiation for patients screening positive for clinically significant insomnia. The adherence benefits have been meaningful and durable. The challenge is operational - most sleep services don’t have integrated CBT-i delivery capacity, and the referral pathways to external psychology services are slow and variable in quality.

This is one of the areas where digital CBT-i platforms with clinician oversight may have a useful role. The platforms have improved significantly over the past few years and are getting better evidence support. The economic case for sleep services to integrate digital CBT-i into their standard workflows for selected patients looks increasingly compelling.

Patient phenotyping for intervention selection

The other area where the 2024-2025 literature has produced useful refinements is in matching adherence interventions to patient phenotypes. The pattern that’s emerged is that different reasons for non-adherence respond to different interventions, and one-size-fits-all approaches systematically underperform tailored approaches.

Patients whose non-adherence is driven by mask discomfort respond best to early interface intervention. Patients whose non-adherence is driven by anxiety or claustrophobia respond best to brief psychological support. Patients whose non-adherence is driven by perceived lack of benefit respond best to education and feedback about residual events. Patients whose non-adherence is driven by partner or relationship issues respond best to inclusion of the partner in the early therapy conversations.

The operational challenge is having the assessment workflow that identifies the relevant phenotype early enough to direct the intervention. Some services have built structured assessment processes; many haven’t.

AI-supported clinical workflows

The emerging story in 2026 is the use of AI to support clinical decision-making in adherence management. Several vendors are offering tools that analyse the telemonitoring data, identify patients at risk of dropping below adherence thresholds, and suggest specific intervention pathways based on the data patterns.

The early experience with these tools has been mixed. The pattern recognition is genuinely useful and identifies at-risk patients earlier than manual review processes. The intervention recommendations are more variable - some are useful, some are generic. The clinical value depends heavily on whether the AI tool is calibrated to the specific service’s patient population and workflow.

Some Australian services have been working with local AI specialists - including consultancies like Team400 and others operating in the healthcare AI space - to build service-specific tools rather than rely on generic vendor offerings. The early indications are that the tailored approach produces more useful clinical guidance than the off-the-shelf alternatives.

The cost-effectiveness question

A practical question for service planning in 2026 is what level of resource investment in adherence support is cost-effective. The answer depends on the patient population, the alternative care costs, and the service’s funding model, but the general direction of the recent health economic literature is that intensive adherence support pays for itself in most settings - through reduced clinical sequelae, better symptom control, and lower long-term healthcare costs in adhering patients.

The American Academy of Sleep Medicine and equivalent bodies have been increasingly explicit in their guidance that adherence support should be considered a core service component rather than an optional extra. The implementation of that principle in funded service models is uneven, but the clinical case is solid.

Where to focus in 2026

For services reviewing their adherence practices in 2026, three priority areas suggest themselves. The early mask intervention window deserves operational attention - the data supports earlier switching than most services currently practice. The telemonitoring-plus-structured-outreach model is well-evidenced and worth implementing if not already in place. And the CBT-i integration for the insomnia-comorbid OSA cohort is an underutilised opportunity with growing evidence support.

Adherence will probably never be a solved problem. But the gap between what’s possible with current evidence and what’s typical practice is wide enough to be worth closing.