The model could make it quicker and easier to decide whether patients need continuous positive airflow pressure machines or not.
Investigators in Australia have designed a new method to evaluate whether oral appliances are likely to be effective in treating obstructive sleep apnea in individual patients.
The strategy uses clinical data and standard patient testing in order to create individualized virtual patient models that can predict the treatment’s success or failure.
The model was developed by Danny Eckert, Ph.D., of Flinders University, and colleagues. Writing in the Journal of Clinical Sleep Medicine, Eckert and colleagues said oral appliances have a number of advantages over continuous positive airway pressure (CPAP) devices, namely, that are well-tolerated and less invasive than CPAP machines.
The catch? Oral appliances do not always work, and clinicians have little way of knowing which patients will benefit from them.
“A major clinical challenge with dental appliance or mandibular advancement therapy is the inability to accurately predict who will respond,” Eckert said, in a press release. “At present, it is a toss of the coin.”
Eckert and colleagues had previously developed a computer model designed to estimate pathophysiological endotypes of patients with obstructive sleep apnea. Having succeeded in that effort, they wondered whether the model might also lend insights into the likelihood of an oral appliance helping a particular patient.
They recruited 62 people with obstructive sleep apnea, performing an in-laboratory diagnostic exam and then oral appliance treatment efficacy polysomnography. The investigators culled seven polysomnographic variables from that study and then paired that information with age and body mass index data. They alsoused a machine learning algorithm to predict therapy response. Forty-five of the patients were used to train the model. The remaining 17 patients were used as a validation cohort, with intervention success based on standard definitions of the apnea-hypopnea index.
Using a 10-fold cross-validation system, investigators found that the model was able to predict oral appliance responders versus nonresponders 91% of the time, on average. The authors used fewer than five apnea-hypopnea index events per hour as their definition of efficacy.
Eckert said a model like this has significant potential to help ensure that patients are given the most appropriate and most effective sleep therapy possible, without the usual trial-and-error approach.
“An accurate model that takes into account the different causes of sleep apnea to match therapy to each person should produce much better results from a dental appliance, particularly when CPAP or other treatment aren’t appropriate or preferred,” he said.
When investigators ran the model using different definitions of success (ranging from <5 to <20 apnea-hypopnea index events), the model’s success rate varied, ranging from 60% to 100%.
Eckert and colleagues said those accuracy ranges are promising. They said the next step is to try to test the model on larger data sets to see how well its accuracy holds up across a wider range of patients.
If the model proves successful, it could become a low-cost, low-burden method of optimizing patient care. Eckert said it is also a demonstration of the ways in which machine learning can be leveraged to make healthcare more efficient.
According to the American Sleep Association, sleep apnea mouth guards can cost up to $2,000. CPAP devices have price tags in a similar range, though the highest-cost models can be around $3,000, the association found.
Aside from cost differences, one reason oral appliances might be preferable is patient comfort. A 2013 study found outcomes were similar in patients given CPAP machines and patients treated with mandibular advancement devices for obstructive sleep apnea; however, they said the difference may have been due to better rates of compliance among patients given the mandibular device compared to those prescribed CPAP treatments.
If left untreated, obstructive sleep apnea can increase a patient’s risk of obesity, diabetes, and heart disease, among other diseases.