Remote monitoring keeps patients safe at home

The mission is to keep covid patients both isolated and properly monitored, in case they suddenly deteriorate.

---

As our national focus shifts away from covid-zero and toward “living with the virus”, we’re left with questions about how the hospital system will manage the increase in patients.

Virtual covid wards, where patients with milder symptoms are monitored at home, offer a shiny new solution to the issue.

But with covid patients prone to rapidly deteriorate without showing any outward distress – a phenomenon dubbed “happy hypoxia” – being able to quickly escalate care is also vital.

Given the highly infectious nature of the virus, in particular the Delta strain, clinicians have been particularly concerned about the potential for spread among vulnerable hospital inpatients.

These fears have been realised in the latest NSW outbreak, with fatal covid outbreaks at six Sydney hospitals since June 16.

Given these risks, it is unsurprising that at-home monitoring of patients with mild symptoms holds promise.

“If you’ve got 600 cases in a smallish town, but you’ve only got a couple of hundred hospital beds, then you are going to swamp that hospital very quickly,” former emergency physician and current University of New England research professor Rod McClure tells The Medical Republic.

“You can’t put covid-positive people in just for observation, you’ve got to bring them only when you need them – if you can observe them at home, that’s great.

“It keeps infectious people out of hospital, which means that the people with non-infectious diseases that need hospital care, can get it.”

In March last year, just as covid cases began to appear in Australia, Professor McClure worked with the Hunter New England local health district to establish one of the country’s first virtual covid wards at Armidale Hospital.

It was stood down when it became clear that Australia had largely avoided a first-wave outbreak, but Professor McClure says the system is simple enough to be rolled out again on short notice.

Each patient in his trial wore a unit on their wrist the size of a large watch, which measured oxygen saturation, blood pressure, temperature and even made a simple ECG reading. These measures were then transmitted continuously through a private cloud-based system to a central monitoring site set up “very much” like an ICU hospital ward, just without patients.

“It’s an ICU-quality monitoring system that can be worn very easily on the wrist of patients wherever they happen to be,” Professor McClure says.

While most hospital and health systems have community-based subacute care programs, widely known as hospital in the home (HITH), he says his virtual ward is different.

In some areas, HITH models have been souped up to create a semi-virtual ward, but these can be relatively low-tech.

“[Health services have] used existing HITH programs for home-managed covid patient wards, and they’ve done that simply by posting out or dropping off a pulse oximeter, and then giving people telephone calls once a day,” Professor McClure says.

“Pulse oximeters are really simple: you just slip a finger into a little bulldog clip and it reads your oxygen at a point in time.

“Some of the wards just rely on the reading from a thermometer which the patient has put under their tongue or under their arm, as well as a self-recorded pulse oximeter reading and a daily chat via a telehealth type model.”

While one health district in Queensland found this model easy to set up and roll out, it did acknowledge the need for enhanced visual communication and selective patient monitoring capabilities in the setting of a larger outbreak.

Professor McClure says the continuous nature of the monitoring in his trial is an especially important distinction from the low-tech HITH models, which generally only take readings once per day.

“If people were gardening or washing up, they basically just took the equipment off, and then plugged it back in twice or three times a day, allowing us to get a very reliable recording at the times that we actually needed it,” he says.

“Now, that is analogous to a hospital because patients in the wards tend to only get their observations done once a day or every four hours, depending on where the hospital they are – if they’re in intensive care, they might get their observations taken once an hour.”

More observations allow the nursing team at the hospital to better identify clinical deterioration points as they happen.

“This system will pick up the happy hypoxic, and it will pick it up before the person does,” Professor McClure says.

“If we ring the patient and ask how they were going, and they say they feel fine but we’re looking at their data and realising that they’re feeling ‘fine’ at 90% saturation, then something’s happening.

“We’re quite sensitive to that quick change because we can see it happening.

“We can call the ambulance before they know they need it – and if they live in, say, rural Australia, then we can actually notify the hospital get them in before we have a problem at two o’clock in the morning with poor transport.”

In Melbourne, Deakin University applied AI researcher Professor Rajesh Vasa and Alfred Health deputy director of trauma services Professor Joseph Mathew took a slightly different approach.

Instead of relying on taking more measurements, the duo focussed on creating an AI that mimicked an experienced clinician’s decision-making process to triage patients.

Like Professor McClure’s system, the hardware involved was relatively simple to operate, with patients just given a pulse oximeter and standard digital thermometer.

Three times per day, the covid-positive participants in the Melbourne pilot would receive a notification via smartphone to input readings from the devices into an app interface.

Upon first downloading the app, people were prompted to input relevant information on comorbid conditions, the details of which were then stored.

The process of inputting information was designed to take less than 45 seconds, increasing the chance of patients completing the process.

The AI then got to work calculating that patient’s risk of clinical deterioration, categorising people as green, amber or red: green meant no intervention needed, amber meant to increase the intensity of monitoring (the system sends requests for tests more frequently) and red meant an immediate intervention was needed.

“We designed the AI to mimic how medical professionals would diagnose and process a patient if they were to see them,” Professor Vasa tells TMR.

“Technically, they would call it a heuristic AI, because it’s using a set of rules, weights and biases.

“But it is based on how a human expert would assess a case – we taught it by showing doctors a lot of sample cases and asking what they would do, and their responses formed the basis of what the AI learned.”

When Professors Vasa and Mathew cross-referenced the AI assessment of real covid patients, they found very little conflict between the two judgments.

“Internally, the system makes an assessment and says, ‘I think this is green’, and then a human either confirms or rejects it,” Professor Vasa says.

“If they reject the AI decision, it learns why they may have rejected it. Over time, it mimics how the humans will diagnose and classify green, amber and red.”

In every case, the algorithm was able to identify the cases requiring immediate attention exactly in line with a clinician.

According to Professor Mathew, the strength of their system stemmed from this ability to automate triage, making it more scalable.

“A lot of patients are being managed by HITH groups, which are under stress because of the sheer number of patients [which they individually contact via telehealth],” he says.

“Our system is so simple that we can deal with a million patients if possible and still pick up clinical deterioration on a very achievable nurse to patient ratio.”

The only thing hindering adoption of this model is interoperability issues.

“Hospital IT departments don’t want to have a scattering of different systems to support; instead they tend to select a big, reliable, sophisticated system,” Professor Vasa says.

“But in a pandemic, you need something lightweight, simple, user-centred first, so we have to relax the constraint that everything has to be integrated and perfect on day one.

“It’s very difficult for people in IT to accept that particular perspective.”

Despite this challenge, the duo remains confident that their system can help save lives.

“We would prefer to just get it out there and then work our way over time to fully integrate into the existing technology systems.”

Acknowledging that this could take some time, they have recently launched a version of the program in the US.

“This innovation can’t just be stuck in Australia, we have to go global with it,” Professor Vasa says.

“At the moment, it looks like we might be getting used globally more than within Australia.”