DoseCare: An Overdose Alarm and Response System enabled by Mobile App, Wearable and Self-Evolving AI Models

The University of Stirling is collaborating with Manchester Metropolitan University and Queens Belfast University on a development of a smartwatch application. This proposal is being submitted to the Scottish Health and Industry Partnership (SHIP) call for overdose technology development under phase one proposals (4 months and a max of £100k). The project for phase 1 (4 months) will develop the machine learning model to refine the algorithm to minimise false positives. This is essential preparation work for when it is used in real life. We will do this by asking a sample of people to wear the watch for a period of time to simply collect more data from a range of people. This will only collect their biometric data (movement, heart rate in particular) and is not connected to any response system.

A sample of 20 people in total will be asked to wear the device across ‘healthy’ volunteers (n=10) and people who use substances and may have a range of other health conditions that will affect their biometric data (n=10). Stirling University and Queens Belfast University sites will seek data collection from people who use substances via hostels and will only require 5 participants in each of Belfast and Scotland. This approach has been previously piloted in Belfast where people were asked to wear a wristwatch device for a short period of time whilst in the hostel. We will work with a third sector hostel (possibly TSA or Simon Community). We will ask them to recruit 5 people to wear the device overnight whilst in the hostel.

The data collected will allow the computer scientists at MMU to work on the algorithm. As the lead organisation, MMU will seek ethical approval. However, we will also seek ethical approval for the local work led by Stirling.
The next phase of development would be to test the refined algorithm in a smartwatch in a larger sample. This is phase 2 as described below. [note the SHIP fund has a 2-phase approach and phase 2 is only funded if phase one is successful]. The plan and full funding application for the second phase of funding is worked up during phase one.

Below is the project outline from our MMU lead for the SHIP application (note this is still being refined):

Our overarching aim for the project, encompassing both phase 1 (4 months) and phase 2 (additional 9 months if phase one is successful), is to further advance and augment our existing overdose alarm system for services in community-based settings, particularly where individuals lead unconfined lives. We are convinced that by utilising consumer-level wearable devices (such as standard smartwatches or wristbands), smartwatch apps, a tailored service platform capitalising on AI/Machine Learning models, and the appropriate intervention protocol (care pathway), we can not only detect and respond to overdose incidents with precision but also provide enhanced care for individuals who use drugs.

While the idea of using wearable technology and mobile apps for personalised self-management (known as quantified-self) is not groundbreaking, it frequently falls short of its intended purpose in overdose detection and response due to the constraints in present devices, service models, and platforms. Individuals who have experienced an overdose are vulnerable and often incapable of managing their own care, depending on caregivers for assistance. As a result, ensuring suitable help relies heavily on accurately detecting overdose events and effectively communicating them to caregivers.

To tackle this challenge, we commence by classifying users into two categories based on their awareness of dosing risks. The first group (high risk-awareness) comprises users who acknowledge the risk and accept responsibility for their actions, while the second group consists of users who tend to undervalue the risks. The alarm mechanisms devised for these two groups will differ. For the first group, we can rely on their robust self-preservation instincts to actively report every dose event using lightweight communication tools or apps (via a two-step report: dosing start and dosing completed) to their caregivers. If reporting is disrupted, an overdose alarm will be activated, and medical assistance will be promptly arranged at the user's location. On the other hand, for the second group (low risk-awareness), who are prone to disregarding dosing risks, we require passive solutions capable of detecting overdose events. Our previous studies have demonstrated that wearable devices are well-received by the user group, and with the aid of AI models, overdose events can be detected and alarmed.

Drawing on the above discourse, the tangible objectives (for both phases 1 & 2) of the proposed project are as follows:

O1: Improve(phase 1) and test (phase 2) a smartwatch app that empowers users to proactively safeguard themselves by connecting their dosing activities with care support services.

O2: Develop (phase 1) and test (phase 2) a self-evolving AI model capable of identifying unusual biomarker patterns based on each individual's unique characteristics.

O3: Design, implement (phase 1), and test (phase 2) an appropriate intervention protocol that can seamlessly integrate with the proposed technical solutions.

By focusing on these objectives, we strive to create a novel, efficient, and effective solution for overdose detection and response that has the potential to significantly improve care for individuals who use drugs while also reducing the burden on caregivers and healthcare services. The innovative combination of wearable technology, AI models, and tailored intervention protocols sets this project apart from existing approaches, opening up new possibilities for personalised care and overdose management.