Sleep medicine has always had a personalisation problem. Two patients can present with identical AHI scores from their polysomnography studies, yet respond completely differently to the same treatment. One thrives on CPAP. The other can’t tolerate the mask and ends up abandoning therapy within three months. The clinical data looked the same on paper. The outcomes couldn’t have been more different.

That disconnect has frustrated sleep physicians for decades. Standard protocols work well for average patients, but nobody is average. The variables that determine treatment success — anatomical factors, comorbidities, sleep architecture patterns, lifestyle constraints, psychological profile, medication interactions — are complex enough that no human clinician can weigh them all simultaneously.

This is exactly the kind of problem that AI excels at.

What Personalised Treatment Looks Like

The traditional approach to sleep apnoea management follows a fairly linear path. Diagnosis via sleep study, CPAP as first-line therapy, oral appliance as second-line, surgery as last resort. Treatment decisions are based on severity scores and broad clinical guidelines.

AI-driven personalisation takes a different approach. Rather than following a decision tree based on a handful of variables, machine learning models can incorporate dozens or even hundreds of patient-specific data points to predict which treatment modality is most likely to succeed for that individual.

Research published in the Journal of Clinical Sleep Medicine in late 2025 demonstrated a model that predicted CPAP adherence with 82% accuracy based on pre-treatment patient characteristics. The model identified factors that clinicians hadn’t traditionally weighted heavily — including specific patterns in REM sleep architecture and arousal index distributions — as strong predictors of whether a patient would still be using their CPAP at six months.

That kind of prediction has real clinical value. If you can identify likely non-adherers before prescribing CPAP, you can steer them toward alternative therapies from the start rather than wasting months on a treatment they’ll ultimately abandon.

Beyond CPAP Selection

Treatment personalisation goes deeper than choosing between CPAP and an oral appliance.

For CPAP users, AI algorithms are now adjusting pressure settings dynamically based on individual breathing patterns, sleep position, and sleep stage. Modern auto-titrating machines have done rudimentary versions of this for years, but the latest generation of algorithms is significantly more sophisticated. They learn from individual patient data over weeks and months, refining pressure delivery to optimise both efficacy and comfort.

For insomnia patients, custom AI development is enabling more tailored delivery of cognitive behavioural therapy for insomnia (CBT-I). Digital CBT-I platforms can now adapt their therapeutic modules based on a patient’s response patterns, sleep diary data, and progress metrics. A patient who’s struggling with sleep restriction but responding well to stimulus control gets a different emphasis in their program than one with the opposite pattern.

Medication management is another frontier. Sleep medications have notoriously variable responses across individuals, influenced by genetics, metabolic rate, and drug interactions. Pharmacogenomic data combined with AI modelling is beginning to predict which medications — and at what doses — are likely to work for specific patients.

The Data Foundation

None of this works without good data, and that’s been a barrier historically. Sleep medicine generates enormous amounts of data — a single polysomnography study produces gigabytes of physiological recordings — but much of that data has been underused clinically.

The raw signal data from sleep studies contains information beyond what traditional scoring captures. Subtle patterns in EEG microstructure, heart rate variability during specific sleep stages, and respiratory effort waveforms all carry predictive value that standard reporting discards.

AI models can process these signals in their entirety, extracting features that human scorers wouldn’t look for. That doesn’t replace the sleep technologist — it augments what they can extract from the same recording.

Home sleep testing devices and wearable trackers add another data stream. They’re less accurate than in-lab polysomnography, but they provide something lab studies can’t: longitudinal data across weeks and months of real-world sleep. That temporal dimension is crucial for understanding individual sleep patterns and treatment responses.

Australian Implementation

Several Australian sleep clinics have begun integrating AI-based treatment decision support into their workflows. The Australasian Sleep Association has published position statements acknowledging the potential of AI in sleep medicine while emphasising the need for clinical validation and regulatory oversight.

The Therapeutic Goods Administration’s regulatory framework for software as a medical device (SaMD) applies to AI-based treatment recommendation systems. Clinics implementing these tools need to ensure they’re using products that meet TGA requirements — which means validated algorithms, transparent methodology, and ongoing post-market surveillance.

The practical reality is that most Australian sleep services are still in early stages of AI adoption. The technology exists, the evidence base is growing, and the clinical rationale is strong. But integrating AI tools into established clinical workflows takes time, training, and institutional willingness to change.

The Honest Assessment

AI personalisation in sleep medicine isn’t magic. It won’t eliminate treatment failures or make clinical judgment obsolete. But it’s providing clinicians with better information at the point of decision-making, and that translates directly into better outcomes for patients.

The patients who benefit most are the ones who would have fallen through the cracks of a standardised protocol — the atypical presentations, the complex comorbidity profiles, the treatment-resistant cases. For them, the shift from one-size-fits-most to genuinely personalised care is more than an incremental improvement. It’s a fundamental change in how their condition gets managed.