Powering personalized sleep apnea patient care with AI

Linde’s new sleep monitoring system harnesses the predictive power of AI to enable better care for sleep apnea patients in Europe.

The human touch: AI enables better care for sleep apnea patients.
The human touch: AI enables better care for sleep apnea patients.

As Linde continues to leverage data and AI to optimize processes throughout the value chain from plant maintenance to logistics, its Homecare business is bringing that value directly to patients. Clinical experts from Homecare in Europe teamed up with Linde’s global AI team to bring an innovative idea to life. Together, they have developed an AI-powered solution that allows care teams to better tailor their interventions to individual sleep apnea patients. The dream? Improved treatment compliance and outcomes. Having already been launched, the product is showing very promising results. 

Unmasking the problem of sleep treatment compliance

For anyone with Obstructive Sleep Apnea (OSA), the much sought after “good night’s sleep” can be painstakingly elusive. It has recently been estimated that around one in five people could be suffering from this chronic condition characterized by irregular breathing and awakening at night due to repeated upper-airway collapses during sleep1. OSA is a large and growing health concern that can affect health and quality of life in a variety of ways, but if properly diagnosed, patients can be given treatments to manage the condition. The most common and most effective treatment is to fit patients with a CPAP (Continuous Positive Airway Pressure) device in their home. These devices deliver a continuous flow of pressurized air into the patient’s airways via a facemask, preventing collapses and resulting in a better night’s sleep … provided, that is, patients actually use them. Studies suggest that up to 50% give up within a year2

“A significant portion of our sales revenue comes from sleep treatments,” says Dr. Tiago Esteves, Senior Business Manager at Linde Homecare for Iberia, France and Benelux. “But our business is not about delivering devices; we aim to deliver the best possible care to our patients – that means the constant strive to address non-compliance through intervention,” he explains.


AI supports our aim to provide our patients with the right care at the right time via the best means.
Obstructive Sleep Apnea occurs when upper airways become blocked during sleep.
Obstructive Sleep Apnea occurs when upper airways become blocked during sleep.

A dream use-case for Artificial Intelligence to create value

Interventions – whether text messages, phone calls or home visits – are opportunities to provide device usage support and address issues like ill-fitting masks – which often is the cause of non-compliance. Typically, these interventions are planned using a blanket approach that doesn’t take into consideration individual patient variables like current compliance. “It’s not the most efficient process,” says Esteves, “and because interventions are reactive, patients in greatest need are not being targeted.” With around 1000 professionals providing care to around 400,000 patients in Europe, scaling-up the Homecare workforce is an expensive means to address the issue. According to Esteves: “We need to predict and prioritize non-compliant patients, personalize the intervention effort and be more proactive.” Enter AI.

“Despite significant development in AI technology in recent years, the technology isn’t the right fit for every business problem,” explains Dr. Dexin Luo, who leads Linde’s global AI team that houses a collective of innovation professionals and hands-on AI technology builders. Luo believes the real value in AI lies in its capability to derive insight from large and complex datasets beyond that of human capability, and then recommend actions – but only for well-described use-cases: “With Linde’s homecare patients increasingly using connected devices, AI supports our aim to provide our patients with the right care at the right time via the best means”.


From reactive to proactive patient care

As the core product development team, Dr. Florian Voiglaender and Amine Kechaou are very excited to see the result of their work having real impact. “Our modelling approach applies supervised learning and transfer learning on large datasets with the goal to forecast treatment compliance for the next 20-30 days,” Dr. Florian Voiglaender explains. Patients are scored on well-defined KPIs and scores are used to recommend an intervention for clinical teams to review. “The system aims to predict drop-offs in compliance and signal the need for an intervention to get patients back on track,” Kechaou adds. In the end, this data-driven approach allows clinical teams to better target their interventions to those most in need. 

Luo sees this as just the beginning: “This sort of insights-driven patient monitoring is a blueprint for transformative, embedded AI in our businesses.” That means no dashboard, just an AI engine that runs and recommends allowing a shift from reactive to proactive workflows. "We will continue to build AI tools that enable our business to improve patient care."


Mask discomfort can lead to less usage: something AI can help flag from device data.
Mask discomfort can lead to less usage: something AI can help flag from device data.
We are seeing a meaningful impact even on patients with low to zero compliance – a really promising sign.

Promising preliminaries from patient testing

Linde’s AI-powered sleep monitoring system is already being put to the test in a pilot study with thousands of Linde’s homecare patients divided into a control (intervention not executed) and experimental (intervention executed) groups. As Dr. Sandra Guedes, Project Manager, reports, “Our evaluation of patient adherence to treatment has been running over the course of several months and we are seeing a meaningful impact even on patients with low to zero compliance – a really promising sign.” Anecdotal reports continue to emerge that capture the powerful coupling of data and the best trained clinical eyes: “We had one case where a patient’s compliance dropped early in the treatment course. The algorithm prompted a text message intervention and on the basis of that, Linde could proactively get in touch with the patient,” Guedes explains. From that touchpoint, the clinical team could easily resolve the issue. As well as being proactive about her own treatment resulting in increased compliance, “she really felt cared for, that we were looking out for her,” adds Guedes. At Linde Homecare, that is always the end goal. And that is why Linde is now expanding this sleep monitoring solution in Europe.

The outlook: What’s next?

The World Economic Forum’s Global Health and Healthcare Strategic Outlook report, which sets out a vision for global health and healthcare by 2035 talks about an “AI-enabled healthcare future” where “Health-focused technologies have the vast growth potential to bring about positive change.” With AI-powered patient monitoring, Linde Homecare is already preparing for that future. “There is a sustainability problem for most of the healthcare systems around the world,” explains Esteves, “More and more care will likely be provided in the home – and not just for sleep treatment.” With AI bound to play a vital role in enabling the transformation from reactive to proactive and eventually predictive and preventative healthcare, it’s an area where Linde will certainly not be caught sleeping.


1. https://www.news-medical.net/news/20230511/Study-Nearly-one-in-five-people-may-be-suffering-with-obstructive-sleep-apnea.aspx
2. https://www.weforum.org/agenda/2023/04/world-health-day-how-ai-and-digital-health-healthcare/