From hospital beds to the seabed – the power of analytics to drive positive change

Home
Faculty and Research
Research at LBS
Impact stories
From hospital beds to the seabed – the power of analytics to drive positive change

Advanced analytics tools like machine learning (ML) and optimisation are leveraging the predictive and prescriptive power of Artificial Intelligence (AI) to predict what might happen in the future, and generate potential solutions that can accelerate and enhance human decision making. Little wonder that analytics are becoming increasingly ubiquitous in industry today.

From energy to e-commerce, media to manufacturing, retail to real estate, chat bots, personalised recommendations, facial recognition systems, route planners and more are transforming processes and operations, enhancing customer experience and creating business value at scale—so much so, that Fortune and others predict the ML market alone will grow to more than$200 billion by 2029.

But besides efficiency gains that can be calculated and measured in dollars and cents, how can AI be leveraged for societal benefit and the betterment of our world?

Jean Pauphilet, Assistant Professor of Management Science and Operations at London Business School, makes a compelling case for the potential of analytics to do good.

Jean’s work straddles two pillars. The first is methodological: he designs algorithms for machine learning, large-scale optimisation and optimisation under uncertainty, to accelerate data-based decision-making. The second is impact.

Jean works extensively with NGOs and organisations to pinpoint and articulate complex problems and then yoke these problems to what he calls “opportunities for analytics:” AI-powered predictions and solutions that he is developing and tailoring to specific needs. This is a dynamic and growing body of work that harnesses abstract methodology and real-world application to deliver significant, measurable impact in areas as diverse as resource allocation in healthcare and the enormous problem of plastic in our oceans.

Abstract and applied: Starting with hospital beds

Jean’s bi-lateral research is the product of passion – initially for mathematics and mathematical modelling. His early interest in the “power of abstraction” to wrap governing equations around physical systems led to a PhD in Operations Research at MIT, and an increasing focus on the use of mathematical tools to enhance decision-making in the real world. When an opportunity arose during his doctoral studies to translate his research into practice at a US hospital based in Boston, he leapt at the collaboration—in part because of his family connection with healthcare.

Jean Pauphilet

Assistant Professor of Management Science and Operations

jpauphilet@london.edu

View profile

Read also

“Both my parents are doctors, so applying my research in a hospital setting was very appealing to me. I never wanted to work in the clinical side and end up competing with my parents. This was a chance instead to work on operations in a sector that I do nonetheless personally relate to—and still enjoy cordial family relations!”

One of the biggest issues facing modern healthcare systems is overcrowding. In large US hospitals at any given time, at least 90% of beds are occupied; a congestion issue that is made all the more acute by lack of visibility around patient length of stay. Historically, hospital managers looking to allocate beds to incoming patients have relied on feedback from nursing staff on future availability, but this is problematic. Issues like staff turnover, different degrees of experience or lack of efficient communication channels create inconsistencies that can further hamper efficiency and undermine patient outcomes.

By feeding patient data into a machine learning model, Jean has been able to create accurate predictions of individual length of stay or need for future critical care—insights that have the potential to significantly streamline patient flow management by accelerating admissions and discharges. And it works by asking the right questions, he says.

“When we began, the hospital wanted to predict how many days each patient would stay. It is an incredibly hard question. But is it a question whose answer will change what the hospital manager should do? Instead, we asked the algorithm: Who is going to leave in the next 24 hours? From there you generate granular predictions that give you far better visibility on how likely it is you will have one or two free beds today. Switching from an open-ended to a ‘yes or no’ question was both easier for the machine learning algorithm and more impactful for decision- making.”

ML predictions can then be aggregated in a dashboard that gives hospital managers a unified and consistent view of bed availability across time and across different hospital units. The result was a reduction in admission delays of between 10 and 20%, within a trial with a hospital in Boston. And that is not all. Jean and his co-authors have also gone on to develop a dashboard that communicates these patient-level predictions to doctors directly, to help them pinpoint and prioritise patients that are closest to discharge—a tool that has reduced length of stay by between 0.1 and 0.3 days in seven trial hospitals in the States.

“This is about saving time for patients that are fit for discharge but are still waiting for the official green light. Roughly, we are seeing a four- to six-hour reduction in stay which is exactly what you want – any greater a reduction and you’re likely to be doing something wrong from a clinical point of view.”

From hospital beds to the seabed itself

Jean’s work isn’t tied to a particular sector or industry but is, he says, “proudly anchored in general-purpose methodology.”

The part of his work that deals with abstraction – with concepts rather than real-world application – gives him the ability to zoom out from specific contexts and learn principles and concepts that can be generalized to other use cases, building connections between otherwise dissimilar problems.

So while Jean’s work has been focused on transforming patient outcomes across different regions within the US, in the last few years a wholly new and quite different opportunity for application has also surfaced—specifically in the fight against plastic pollution in our oceans.

The World Economic Forum predicts that the volume of plastic in the world’s oceans could triple by 2040, outweighing fish by 2050. And while the brunt of this environmental emergency is currently borne by marine life, the threat to human health is also dire. Chemicals in plastic are increasingly connected to hormonal, neurological and endocrinological problems along with a slew of cancers, as well as diabetes, kidney, liver and thyroid issues.

In 2022, Jean began a collaboration with Dutch NGO, The Ocean Cleanup, whose work in eliminating marine plastic is principally focused on the Great Pacific Garbage Patch—a notorious, swirling gyre of debris that is estimated to be three times the size of France, containing plastics that are thought to date back to the 1960s.

To capture this waste, The Ocean Cleanup operates huge nets of between 1.4 and 2.1km in length that are hoisted between boats, targeting the plastic waste while allowing marine life to pass through. Together with LBS PhD student, Baizhi Song, Jean has developed an optimisation algorithm that anticipates the movement of the plastic through water and time and optimises the trajectory of Ocean Cleanup vessels, and schedules regular emptying of the accumulated debris which is then recycled.

To date, The Ocean Cleanup has collected 300 tonnes of plastic from the Great Ocean Garbage Patch. With this advanced routing algorithm, The Ocean Cleanup estimate that they could clean the Great Pacific Garbage Patch in half the time and almost half the cost than their previous estimates. Work is currently underway to improve the quality of input data using Go Pros attached to boats and drones that capture the movement of the plastic. Jean and Baizhi are also working on dynamically expanding or shrinking the size of the net to trade-off efficiency of the collection and friction through water.

“With my PhD student, we are now looking at further optimising the size of the net and the speed of the boats in real time. We are also interested in optimising data collection: where should we send drones to take pictures of actual plastic pollution so that we improve the predictions that inform our routing?”

One body of work, two pillars

The two pillars of Jean’s work, methodology and applications, are “equally important,” he says. Bringing them together stems from a desire to connect his methodological work to real-world application and practice; to find what he calls the intersection between relevant problems and innovation.

“There’s a danger in becoming too engrossed in the methodological side of research and getting into a bubble where you start finding challenges or frontiers that are somehow detached from what’s practically relevant in the world. And I want to build new methodologies that are meaningful and impactful. To me that’s the objective. That’s the gold.”

Balancing both worlds means allowing for the applied work to inform and shape the methodological, and vice versa. Getting both right means dedicating requisite thought and time to each. At times, Jean will prioritise academic work that has no immediately obvious application. At other times, he immerses himself in real-world issues. There’s a symbiosis, he says, where the applied work is sparking new methodological questions that need to be answered, and those answers fuel solutions that can be applied.

“Maybe you need to dig into the challenge without having any preconceived idea about the kind of algorithm you’ll need. And the other way around. Sometimes you need to take time to think about the method itself without rushing to implement it to solve a practical problem. Both matter.”

Looking forward

Jean is excited by what he calls the democratisation ofAI and machine learning through tools like ChatGPT: a welcome opportunity to bring the transformational power of technology and its potential to solve problems into the hands of end-users. Academics have a role in generating knowledge, but have historically struggled with closing the loop, he says – ensuring that the outcome of research, the algorithm or message developed make their way into the hands of users who need and will use them.

Generative AI has furnished us with the potential to build interfaces that empower people to interact with sophisticated mathematical models; models that they may not fully understand but are still able to use.

“It’s inspiring to see how excited people are becoming about technology. It also has the potential to change mentality about how to use AI to find solutions. So maybe there’s overhype about ChatGPT, but it is triggering much-needed discussions about what we can and should use technology to do – what can be automated, supported by data, supported by analytics tools in general. And in the long run I think this could be a game changer.”

In terms of his own work, harnessing the abstract for real-world application and leveraging the transformational power of AI to reach end users will continue to yield opportunities in the areas of environmental impact and sustainability.

He is keen to pursue more projects in healthcare, looking for instance at A&E departments to predict patient volume and optimise staffing resources and nurses’ schedules – chronic issues that hospital administrators have long grappled with and struggle to resolve.

He is also interested in work that arrests the pollution of our oceans at source: getting plastics and other emissions out of our rivers and improving waste collection, waste operations and landfill management.

What channels his interest in these disparate areas is the urgency of need—the acuteness or relevance of the problems for managers and decision-makers, and for human outcomes and wellbeing.

From healthcare to the environment and beyond, Jean perceives the clear potential for analytics and the extraordinary advances being made in machine learning and optimisation to offer solutions; solutions that can inform individual, organisational and societal outcomes.

"We cannot afford to choose which problems to work on. Given the current crises, solutions that address the most pressing, the most acute and the most existential of human needs cannot be postponed."

Find out more about the largest clean up in history.