How Storm ID used AI to breathe new life into COPD care

What do you think of when you hear the term artificial intelligence? Maybe your mind goes straight to science fiction; to Hollywood blockbusters that centre around a battle between human and machine? Or perhaps AI means convenience to you? Do you think of virtual assistants like Siri, that help guide us through our day; or personalised Netflix recommendations, that help us choose which series to binge-watch next?

At Storm ID it means a little bit more than that. Here, AI is the cornerstone of some of the most important work we’ve ever done. AI has allowed us to create preventative healthcare services that can pre-empt flare-ups in patients with chronic conditions – before they result in emergency hospitalisations.

An AI Case Study

Over the last 18 months we successfully trialled our Lenus COPD Management Service in NHS Greater Glasgow and Clyde. As part of the trial, 100 COPD (Chronic Obstructive Pulmonary Disease) patients were offered the opportunity to try our innovative service that uses the daily patient reported outcomes as well as input from wearable-device data – i.e., their heart rate, activity, sleep activity – to gauge their likelihood of COPD exacerbation. As a result, we saw an average drop of 1 hospital visit per patient that year and the quality of patients’ life increase exponentially.

Owing to its remarkable success, a further 2,000 COPD patients have been offered these same tools and technologies to help manage their condition too – with a view to national rollout.

How Does it Work?

Patients answer Patient Reported Outcome (PROs) questions daily, weekly, and monthly through our Lenus Health app. A subset of patients are also issued with ResMed home NIV masks for respiratory data and FitBit activity trackers that continuously monitor the heart rate, sleep, and physical activity of the patient. This data provides the basis of our work.

We then enrich our dataset with historical EHR (Electronic Health Record) data and by using Feature Engineering: this prepares it for the machine learning algorithm and helps improve its performance later. We do this by taking descriptive statistics from the raw data and combining it with more advanced feature creation. The advanced features are developed using insights discovered during the data exploration phase, together with input from experienced clinicians. The clinician-driven features are further used to gain insight into how the model makes a decision and to give it explainability.

For example: Where we noticed that activity reduced in the days prior to a flare-up in the patient’s condition, a clinician could explain this and we could feed this into the framework as a feature.

Machine learning algorithms are then used to build models capable of answering the most import clinical questions. Every step of the model training and validation process is carried out rigorously to eliminate any possible biases and to ensure no data leakage occurs. The result is a robust suite of models which provide clinicians with in-time decision support to prioritise patients that are most at risk.

Figure 1 – Schematic illustrating the machine learning process from raw data through to a model ready for production.

Lenus

The work is made possible through our Lenus Health Platform. Its purpose is to use patient data to create and inform new models of healthcare that are preventative, proactive and participatory. Thousands of patients and clinicians use the platform daily and as a consequence generate swathes of clinical data – all of which can be used in combination with machine learning to improve outcomes and patient care.

What does this mean for the future?

As a result of our work, we have refined ideas and developed techniques that are not confined solely to the COPD condition.

The transferable nature of our AI advances means that there is a profound and exciting opportunity to adapt these same techniques to a broader spectrum of illnesses.

There are many existing conditions that take into account the same or similar health metrics that we used as part of the COPD Management Service.

How that health data can be harnessed and repurposed in different ways could potentially affect the lives of millions of patients around the world who live with what today is considered a debilitating illness, but tomorrow, could be entirely manageable.

If you would like to learn more about Lenus or are interested in learning how AI could help, reach out to the Storm ID team and we would be happy to have a discussion!