Data and Health-Care Design  

View course on architecturalrecord.com »

Photo © Jiri Havran

Glasblokkene Hospital relied on an all-digital design and delivery process.

There’s tremendous buzz and a fair bit of confusion concerning artificial intelligence (AI), but it’s not oversimplifying to say that what AI boils down to is data. Health-care facilities, as one of the most complex building types that architects design, are awash in information, so it’s no surprise that the sector’s design teams are in the vanguard of data-informed design and construction.

“AI is nothing but how data is changing the way we look at our design solutions,” says Sumandeep Singh, health studio-practice leader in the Singapore office of HKS Architects. Singh’s team is developing a generative-AI tool for using the firm’s decades of data to help optimize the programming of new health-care facilities. The initiative aims to bridge a gap between the data that the firm and its clients have accumulated—including meeting notes, room templates, spreadsheets containing room-type ratios, and more—and the issues that the information could help solve. Examples might include right-sizing inpatient facilities for declining length of stays, making adequate spatial provisions for the coming wave of automation and robotics, optimizing adjacencies for operational efficiency, or incorporating post-occupancy insights into, say, how design can help prevent medical errors or patient falls. The interactive platform will facilitate new inputs (including site factors such as height restrictions), translate the data into visual formats for communication, and even connect to design software such as Grasshopper to suggest options for building massing. It is one example of how health-care architects are harnessing data to expand the scope and value of their services.

Further examples of data-informed health-care projects include Phase 2 of Glasblokkene Haukeland Hospital for Children and Women, designed by KHR Architecture, PKA ­Arki­tekter, Henning Larsen Norway, and Schönherr (Arkitektgruppen BUSP) and completed in 2023 in Bergen, Norway. The project is one of the world’s first health-care facilities to pioneer an all-digital design and delivery process. UCI Health – Irvine, delivered by a design-build partnership of CO Architects and Hensel Phelps, and expected to complete this fall in Irvine, California, used AI- and augmented-reality-integrated simulations to develop and test design innovations. And the iKure Health Hub, by HDR, now preparing to begin construction in rural India, used machine learning and generative AI to guide schematic design, helping to make the most of limited resources.

Glasblokkene’s 538,000-square-foot second phase consists, like its first phase, of a series of four narrow glass blocks designed to break down the scale of the development while optimizing daylight and views to healing gardens, greenery, and distant hills. Shoe-horned into a tight urban site, with a fully operational hospital next door and houses on the other side, the project used digital-twin technology initially to help minimize the risks and unforeseen costs of building under these constraints, and then to support the facility’s operations and maintenance, extending the value of the digital investment over the project’s life cycle. (A digital twin is a virtual representation of a physical object.) “Health-care buildings, by their nature, are especially well suited to digital-twin technology,”
says Caroline Tjernås, project architect at Copenhagen-based KHR. “They involve a high degree of technical integration, vast amounts of critical data, and operational environments where reliability and downtime are crucial factors.”

Photo © Jiri Havran

Organized to optimize daylight and views of nature, the team behind Glasblokkene (above & below) used digital-twin technology to build under the constraints of its tight site.

Photo © Jiri Havran

Photo © Jiri Havran

Image courtesy KHR Architecture

During Glasblokkene’s design and construction, the model provided a centralized repository of building information. It enhanced coordination and transparency among stakeholders, helped to eliminate conflicts within the design, and made possible a virtual dry run of the construction process. Anyone on the project team could (and facility management staff still can) click on a component—whether a wall assembly, a ventilation unit, a lighting fixture, or a clothes hook—to check on its status and access its associated data sheets. But the digital twin isn’t “one product that solves everything,” Tjernås says. The project team relied on a suite of (more or less) complementary tools, including databases, collaborative software, and virtual reality (VR) platforms, all brought together to enable the team to manage data, coordinate tasks, and visualize the project collectively and in real time.

Even working on a constrained site, during Covid, using previously untried processes, the project was delivered on time and on budget. “Having all that digital information, open to all the participants, was crucial.” Tjernås says. “Otherwise, we wouldn’t have made it.”

The fully digital undertaking wasn’t without its challenges, however. One of the most significant pertained to the sheer amount of information contained in the digital twin and the shift away from drawing sets tailored to each component or system. “What happens when we send everything, every little piece of information we have”—which is what the digital twin represents—“and tell a contractor, ‘you’ve got to find what you need in this’?” Tjernås asks. Contractors taking their own measurements without a shared reference point, for example, could easily allow construction tolerances to accumulate into significant errors—and that’s after they’ve managed to disregard a lot of unrelated info­rm­ation in order to identify their scope of work. Another challenge came from the technology’s effect on project priorities. “You are very focused on getting everything in the model correct, sharing the model, putting all the information into all your different objects, saying, ‘This is ready to develop. Check, check, check.’ ” But what is missing is a check box for good design. “That was something we had to remind ourselves of.” Challenges notwithstanding, the result, Tjernås says, is “a really good hospital, where staff is happy, the client is happy, and they’ve had an easy handover to their maintenance team.”

The handover to operations is a key piece of the digital twin’s utility. All too often on complex projects, building information ends up hidden on a server somewhere, forgotten. For Glasblokkene, having a comprehensive and dynamic model, with staff trained to access it, is where the long-term value of the digital investment lies. “Building a hospital is expensive,” Tjernås says, “but it’s nothing compared to running it.”

Where data-rich digital tools facilitated Glasblokkene’s design development and later phases, computational design contributed much earlier in the process for the UCI Health project. The 580,000-square-foot facility, which includes a 357,000-square-foot hospital and a 223,000-square-foot ambulatory care and cancer center (as well as a parking structure and central utility plant), is innovative in a number of aspects. It is expected to be the first all-electric health-care facility in the country. It pioneers a universal room type, permitted as intensive care, for which the client can change the licensure to meet evolving needs. And it integrates both inpatient and outpatient procedures on a 164,000-square-foot surgical mega-floor, with over 20 operating and procedure rooms, all functioning as a single system.

Image courtesy CO Architects  

For its UCI Health project (above), CO Architects developed the OR layout (below) with the aid of AR (below) and used a gaming engine to simulate patient flows.

Photo © Tom Bonner

Image courtesy CO Architects

Image courtesy CO Architects 

“When you’re the first to do something, how do you know that you’re getting it right?” asks Gina Chang, a principal at Los Angeles–based CO Architects. Central to the team’s confidence in these innovations, especially for the surgical floor, was the extensive use of augmented reality (AR)- and AI-integrated mock-ups to test them. In one type of simulation, the architect used AR to superimpose medical equipment—lights, booms, gurney, and the rest—onto a mock-up of a bare operating room. Representatives for each type of surgical procedure could then don goggles and try out the setup, interacting with the virtual equipment and, in multiuser scenarios, with one another, in the physical space.

Another type of simulation, built by CO in-house using a gaming engine, modeled the surgical floor’s complex workflows. Incorpo­rating client data, such as case volumes and staffing schedules, and drawing on the architect’s accumulated health-care-design expertise, CO coded the AI agents in the engine—pretrained with basic behaviors, such as not walking through walls—to simulate hospital workers moving through a full day of operational scenarios. “In a health-care environment, there’s a whole range of asynchronous behaviors,” says Chikara Inamura, director of digital technology at CO, referring to such activities as patient transport, surgical preparation, team communication, and maintaining a sterile environment. “Through simulations, we have an ability to understand emergent collective behavior, to facilitate collaborations, and to understand the stress points of each design decision.”

Once the simulation was built, “we literally tried to break the system,” Chang says. The designers made all the surgeries 30 minutes or less. “That’s never going to happen,” she says, “but at that point, what doesn’t work anymore?” The simulations also highlighted opportunities. For example, small changes in equipment layout, work configuration, or spatial programming offered the potential to reduce patient-flow distances, shorten staff’s walking distances (a hugely important metric in a time-crunched system), cut the time that surgery-prepped patients had to wait, clear bottlenecks, and even provide options for increasing the size of the hospital’s sterile core, should that become a client priority.

With part of the facility already operational, UCI Health has reported significantly increased revenue generation and improved patient-satisfaction scores, Chang says, compared to operations in the campus’s existing buildings. Impressed with the simulation’s effectiveness, the client is now asking CO to do additional, post-occupancy work to help plan staffing and to organize workflows by room, procedure, and doctor. “So they’re continuing to find the value in this,” Chang says, “further than we’ve ever taken it before.”

Pushing the envelope of computational design is also what enabled HDR to help a social enterprise extend its reach in the Indian province of West Bengal. The project’s client, iKure, is itself at the forefront of using AI and machine learning (ML) to deliver health care in underserved communities. (ML is a branch of AI focused on enabling computers to infer, predict, and improve based on iterative exposure to data, evaluation of results, and adjustment.) “We thought, ‘Could we develop a design approach that emulated what they’re doing in a way that would aid us in helping them?’ ” says Jason-Emery Groen, design director in HDR’s Kingston, Ontario, office.

Image courtesy HDR

Machine learning and AI helped HDR create its scheme for the iKure Health Hub in West Bengal, India (above and below).

Image courtesy HDR

Initial form-finding studies for the 8,500-square-foot facility, which will be built using local labor and locally produced brick, was guided by an HDR-developed ML model. Incorporating publicly available climate data, flow data simulating the movement of people through space, knowledge from HDR’s health-sector experience, a library of images of vernacular buildings in the region, and the facility’s program requirements (including, in addition to operational needs, such factors as security, ventilation, shading, monsoon resilience, and thermal comfort), the model ran thousands of iterations of a random form in just hours. It evaluated each, learned from its successes and shortcomings, and saved the pro bono design team weeks of work. “This isn’t taking a job away from somebody, as you sometimes hear in conversations about AI,” Groen says. “Rather, we’re doing as much as we can with as little as we have.”

Throughout those iterations, the model came up with a courtyard scheme—every time. “We realized that the machine learning was using the data to fast-forward what had taken place over centuries, if not millennia, in that environment,” Groen says. “So the vernacular had done this already.” But for a design team not steeped in that vernacular knowledge, he says, “the machine learning allowed us to have quicker, more intelligent discussions with our local partners about functional form-making.”

AI is commonly described as either predictive (using statistical analysis and ML to anticipate outcomes based on patterns in historical data) or generative (extrapolating from patterns in data to create new content). Groen describes the role of AI in designing the iKure clinic as both. In predictive mode, the model anticipated the impacts of multiple variables at once, while, in generative mode, it applied the architect’s parameters to develop forms that fulfilled them. “From there, the human touch picked up, editing and refining the balance,” Groen says.

The resulting design shows a high-tech, low-cost clinic configured as a single-story with a central courtyard, with light wells at hallway intersections. Passive cooling techniques are deployed to keep the building comfortable year-round, with supplementary air-conditioning limited to critical zones not suited to natural ventilation. Perforated brick towers will top each light well—a reference to the kiln chimneys that signal the surrounding communities’ prosperity; during the day the towers will filter incoming sunlight, and in the evening they will stand as illuminated beacons. They will also offer connection points for stairways to a second story, should the facility need to expand in the future.

A piece of feedback that has cropped up more than once, Groen says, is an appreciation for the warmth of a design that emerged from a process based on “cold” technology.

Groen thinks the explanation for that seeming paradox is that “machine learning helped us get through the pragmatics of things quicker, so that we could get to the human conversations that matter.”

Each of these projects—Glasblokkene, UCI Health, and iKure, as well as HKS’s research initiative—illustrates the potential of data-informed design in a sector it’s particularly well suited to. Other building types, however, are also—or may become—good candidates as costs and benefits align. For architecture firms wanting to contribute value as the use of AI and computational design evolve, Singh recommends structuring and examining the data they already have. Other industries—automotive and medical, for example—are further along in this process, he says, and he encourages architects to learn from them as well as through structured study, workshops and conferences, and hands-on experience. “We must equip ourselves with knowledge of how data works—and how to work with it,” he says. “Clients will be looking to us to demonstrate how our proposed solution has been informed by data.”