Every 30 minutes a stroke patient who could have been saved dies or is permanently disabled, in many cases because they were treated in the wrong hospital. In Europe, only about a third of stroke patients have access to the organised stroke care they need to survive a stroke with minimal or no disability.
Stroke treatment is particularly time-sensitive. The so-called door-to-therapy (D2T) time describes the time interval between a patient arriving at the hospital and the initiation of their therapy. D2T time is critical for patient outcomes: on average, each single minute saved adds two days of healthy life and every 15 minutes saved adds one extra month of disability-free life.
The challenge involves getting thousands of hospitals fit for optimised stroke care
Motivated by those findings, Boehringer Ingelheim (BI) started the “Angels Initiative” with one simple mission: leverage new AI and data technologies from Silicon Valley to increase the number of patients treated in stroke-ready hospitals and optimise the quality of treatment in existing stroke centres. Together with its partner the European Stroke Organisation, Angels is establishing a community of 1,500 stroke centres and stroke-ready hospitals.
This challenging goal involves getting thousands of hospitals fit for optimised stroke care. Therefore, the Angels team under project lead Jan van der Merwe, together with BI’s in-house data science team, developed SCOPE – an AI-based Stroke Care Optimisation Engine.
SCOPE is based on treatment pathway data from hospitals, which comes in the form of simple time stamps. Whenever a stroke patient enters a participating hospital, the stopwatch is set to zero and the time of individual treatment steps is recorded. For instance, a blood sample may be taken ten minutes after arrival at the hospital, the stroke specialist might see the patient three minutes after arrival, and the results of a brain scan may be available 40 minutes later.
SCOPE then analyses the statistical dependencies and patterns in this data and identifies, evaluates, and prioritises improvement suggestions. The recommendations are then provided to the stroke specialists in the hospital in the form of natural language, together with insights and recommended actions.
The difficulty comes from the variability of stroke treatment across facilities
Part of the difficulty in reducing treatment times comes from the variability of stroke treatment across facilities. Every hospital has an individual way of treating stroke and these manifest as patterns in the timestamp data collected by SCOPE. The Angel’s team has discovered that when the time of a certain treatment step shows strong fluctuations it often times hints toward an inefficient process in the hospital.
In some hospitals, door-to-first-medical-contact time was under two minutes with fluctuations of only one minute, whereas other hospitals only a few miles away would show an average door-to-first-medical-contact time of ten minutes with fluctuations of 40 minutes. Furthermore, parallel treatment is key for saving time in stroke care and SCOPE is able to utilise causal inference models to identify dependencies between treatment steps, providing recommendations so that the hospital’s stroke specialists can use to parallelise non-dependent steps.
Beyond these simple cases, SCOPE is also able to find complex statistical patterns in treatment data that humans had previously missed. For example, according to the American Stroke Association Stroke Council, the goal is to complete initial evaluation and assessment, including imaging and laboratory studies, within 60 minutes of patient arrival. SCOPE discovered that significant treatment delays occur when seemingly efficient sub-processes in stroke treatment collide in unfortunate ways that delay each other.
Managing all of those complex processes in an hour is a challenging task and cases of sub-processes delaying one another have been found to cause massive delays of 30 minutes or more in overall treatment times - delays which can cost a stroke patient their mobility or even their life.
Traditional healthcare companies are embracing the new generation of big data and AI
Projects like SCOPE and Angels are part of a larger trend in healthcare today. Traditional healthcare companies are embracing the new generation of big data and AI technologies coming from Silicon Valley to improve patient care and outcomes.
The Data Incubator has worked with hundreds of traditional companies looking to expand their data processing and analytics capabilities. In the last six to 12 months, we’ve seen a near tripling in interest in data science hiring and training from our healthcare-related clients. While much of this demand comes from the start-up world, large, established players like Boehringer Ingelheim are following closely, leveraging many of the same techniques as their smaller competitors.
SCOPE is a perfect exemplar of the new generation of healthcare initiatives from large healthcare enterprises. Like many of the new generation of data and AI technologies, SCOPE is able to pool what it learns from all participating hospitals so that stroke centres joining the program will not only profit from insights gained from their own data, but also from the knowledge that was built up by analysing their peer-group’s data, even as no data is being shared.
Borrowing another page from the Silicon Valley playbook, SCOPE is an iterative and adaptive process, not run-once-and-forget batch processing. It keeps an eye on changes in the treatment process. If a hospital is focusing on improving a specific treatment step and, in doing so, pulls resources from a different step, this can lead to new delays which the engine will be able to identify and communicate to the right stakeholders.
And the hard work is paying off. The Angels team has worked with hundreds of hospitals worldwide to implement organised stroke care, reduce treatment time, and improve patient outcomes. The program has already impacted on 2,500 patients in partner hospitals. Angels’ goal for the next 24 months is to treat at least 50% of all stroke patients in under 60 minutes. The current European average is around 80 minutes. When these quality targets are achieved, 100,000 lives will have been saved in Europe alone.