As the volume of data in healthcare grows, device manufacturers, healthcare technology management (HTM) professionals, and clinicians look for new and improved ways to put it to work. What can these data tell us? Years into what some have described as healthcare's “digital revolution,” this same question is still being asked.
In many cases, the answer is obvious: The patient requires a kidney transplant, or the medical device needs immediate repair. Sometimes, however, the data cause confusion or their purpose and applicability aren't entirely clear.
“You can get a lot of analytics off an infusion pump,” said Loretta Dorn, director of clinical nurse liaisons with Fresenius Kabi USA in Lake Zurich, IL. “But once you have that data, how are you going to use it?”
For example, she said, some institutions are taking data gleaned from their pumps and using them as a reason to “write-up” their nurses. “They discover you bypassed the system x number of times, so they decide you're going to be dinged.”
A better approach might begin by exploring the reasons for the bypassing, said Dorn. “Is the issue that the pump doesn't have the right parameters, or is it just that your nurses don't like to use the software?”
Data from research conducted by Karen Giuliano, PhD, RN, MBA, FAAN, joint associate professor at the Institute of Applied Sciences and College of Nursing at the University of Massachusetts Amherst, and others have found that “the complexity of the device user interface, the time required to program the dose-error reduction system (DERS), and incomplete drug libraries are among the most frequently cited reasons that nurses bypass intravenous (IV) smart pump safety features. The complexity of IV medication administration and the multiple steps involved demands close attention to detail and ultimately relies heavily on human–device interaction to detect and mitigate errors. Clinicians in the busy critical care and medical-surgical clinical environments are frequently interrupted and rushed during IV smart pump programming. As a result, the overriding of alerts and programming outside of the DERS owing to time constraints and competing work demands are recognized as a part of daily clinical practice.”1
Fresenius Kabi, Dorn noted, offers its own infusion system that includes software designed to alert clinicians when drug doses or rates are attempted beyond preset limits. The dose-error reduction tool includes a drug library users can customize for specific medications and care areas. Part of Dorn's job with the company involves visiting facilities where the pump has been deployed and training nurses on using it correctly. A registered nurse herself, she previously worked as director of clinical operations with a public health department with close to 20 sites.
Dorn said she's familiar with the plight of nurses and other clinicians who feel like they're inundated with data of dubious utility or struggle with the technologies meant to put data to work. At her last job, the electronic health record (EHR) caused many employees to lose sleep at night.
“The physicians hated it, the nurses hated it, everybody hated it,” said Dorn. “If they could have gone back to paper charting, they would have been ecstatic.”
Although some organizations are hiring scribes to handle the data input demanded by technology during patient care, Dorn's previous employer did not. Further, she said, it lacked the resources needed to use the system to its full potential.
“I think we forget sometimes when we create these complicated systems and we use data analytics and connect different medical devices, if your institution doesn't have the financial means or if it doesn't have the people who know how to use them, you can have all the data in the world and it's not going to do you any good,” said Dorn.
From Data to Knowledge
Health systems certainly don't have all the data in the world, but it may seem that way sometimes. One widely cited report from the International Data Corporation predicted that the total volume of medical data generated annually would top 2,300 exabytes by 2020,2 while other research from Dell Technologies found that healthcare data of all kinds grew by 878% between 2016 and 2018.3 Those data are coming from all directions: from patient health records, to medical imaging technologies, to wearables, and to edge and Internet-of-things devices. If it's electronic and connected to a health system in any way, there's a good chance it's collecting, transmitting, or storing data.
The challenge, as Dorn's experience has shown, involves where all those data eventually wind up. If the data are in a system intended for a technician or clinician who doesn't have the time or training to decipher their meaning, they're unlikely to help with device maintenance or drive improvements in patient care. In 2019, according to a report by the World Economic Forum, 97% of the data produced by hospitals weren't used at all.4
HTM departments “have been doing the same thing since the early 2000s,” said Eliezer Kotapuri, CEng, CCE, MS, PEng, chief clinical technology officer with Mass Technologies, LLC, in Columbus, OH. “It's, ‘Let’s go get this software, let's go get that software,' because they know that software can generate data and they know having that data is important. But what they haven't been able to figure out,” for the most part, “is how to use this data to drive decision making.”
Organizations have excelled at data gathering, he said, “but they're still behind when it comes to transforming it into information and knowledge.”
Data, Kotapuri explained, “is unorganized facts and statistics.” Information, on the other hand, “is structured and ready for analysis—and that's where we can generate knowledge.” Too often, he said, HTM professionals and others in healthcare make whatever decision they're required to make and only later seek data to support that decision. “We need to get off that train and instead be thinking, ‘This is the data; let's see what my decision should be.'”
Mass Technologies provides consulting and services to clinical engineering and information technology (IT) professionals in hospitals around the country, Kotapuri said. “Our focus is on integrating people, processes, and technologies to help them operate more efficiently and improve clinical outcomes.”
In his experience, although a “fair percentage” of HTM departments turn to service history data to guide device management, too few are using data-processing technologies, such as artificial intelligence (AI), to identify potential issues before they occur. “If the question is about HTM using AI to proactively make decisions about the health of clinical technologies, there is a lot of room for progress,” said Kotapuri.
In the clinical realm, Kotapuri noted, adapting AI for mammography screening is allowing physicians to predict, with a high degree of accuracy, the presence of malignant and benign tumors without putting patients through painful biopsy procedures. Likewise, early-warning predictive modeling systems can alert clinicians to a possible decline in a patient's condition, thereby allowing them to intervene proactively.
Kotapuri said he'd like to see more HTM departments take a page out of their clinical colleagues' playbook and use predictive-modeling tools to better inform their work. “There's no reason they can't be doing the same kind of thing. The technology is available; they just need get the required training and start using it,” he said.
He knows this, Kotapuri said, because his own company uses a software program called DETECT (Dynamic Expectancy Term of Existing Clinical Technologies) that is intended exactly for this purpose. “While we may both be driving the same kind of car, our usage, driving habits, and servicing methodology will probably be much different between one another.”
Eliezer Kotapuri of Mass Technologies describes data as “unorganized facts and statistics.” Information, on the other hand, “is structured and ready for analysis—and that's where we can generate knowledge.”
Eliezer Kotapuri of Mass Technologies describes data as “unorganized facts and statistics.” Information, on the other hand, “is structured and ready for analysis—and that's where we can generate knowledge.”
With that said, he asked, “Why would the maintenance of our cars be the same? And who would ever say that our cars need to be replaced at the same time?”
Kotapuri said that Mass Technologies' AI-backed system looks at “multiple dimensions within a few domains. ... It allows us to come up with a very objective decision: maybe the device needs to be removed from the hospital this year, or maybe it's still good for five more years.”
Binseng Wang, ScD, CCE, FAIMBE, FACCE, vice president of program management with the clinical technology management division of Sodexo, believes that many HTM departments should be leveraging data better than they are currently. “Physicians are using ‘evidence-based medicine' when they treat their patients,” he noted. “We should treat medical equipment the same way.”
An HTM professional, Wang explained, is responsible for the care of medical equipment just as a clinician is responsible for his/her patient. “We want this device to be safe and reliable, and enjoy good health, because then it can produce good work.” With “evidence-based maintenance,” he said, clinical engineers and others in HTM “use data to analyze how we should provide services to our equipment.”
Just as a responsible surgeon would never perform a procedure on a patient if their diagnoses didn't suggest that procedure was required, HTM professionals should not work on the equipment under their purview without analyzing the data previously collected on that equipment. “Any other approach doesn't make sense,” said Wang.
Sodexo, an independent service organization, is “aggressively trying to leverage data” to benefit its clients and their patients, Wang said. As an example, he pointed to the company's ongoing work to develop alternative equipment maintenance (AEM) programs. Such programs allow HTM professionals to service devices in line with their needs, rather than solely in accordance with manufacturers' recommendations.
“It's not to save money,” noted Wang. “It's about using your and your hospital's limited resources more effectively for the ultimate benefit of the patients.”
Time and time again, Wang explained, hospital HTM teams and independent service organizations, such as Sodexo, have looked at the data pertaining to equipment maintenance and found that deviating from the manufacturer's directions does not sacrifice patient safety or equipment reliability. In fact, he said, in many cases, they've learned that the manufacturer's recommendations were insufficient.
“That is, we needed to do more maintenance than they suggested because of the high usage or the unique environment in which it was used,” said Wang.
The Joint Commission and the Centers for Medicare & Medicaid Services allow AEM strategies that are supported by data. However, in Wang's opinion, more communication should occur between HTM departments and manufacturers about the data being collected and what it shows.
“The manufacturers don't know what's going on out there because the majority of equipment they produce for hospitals isn't maintained by them,” he said.
The result is that their maintenance recommendations often don't reflect how their devices are used in the real world. “In my over 20 years working in this field, every time we've offered our data to the manufacturers, they've refused to come take a look at it. They say they don't have the resources to analyze the data as required by the Food and Drug Administration (FDA) and that what they've published is simply the way it is.”
According to a report by Allied Market Research, the global market for predictive analytics in healthcare, valued at $1.8 billion in 2017, is expected to top $8.4 billion by 2025.5 Driving the growth, the firm noted, are factors such as the need for greater efficiency and cost control, as well as a push for more personalized and evidence-based medicine.
AI in healthcare, meanwhile, is projected to expand at a compound annual growth rate of more than 44% over the next five years, from $4.9 billion in 2020 to $45.2 billion in 2026, according to MarketsandMarkets Research.6 The forces at play include the “increasing volume of healthcare data and growing complexities of datasets,” combined with better computing power and declining hardware costs.
Although much of the money spent on AI and predictive analytics will be targeted at technologies that directly affect clinical decision-making, experts such as Mass Technologies' Eliezer Kotapuri believe that HTM departments will get a slice of the pie as well.
“Clinical engineers need to prepare for this,” he said. “As these technologies improve and become more refined, they're going to replace a lot of the clinical engineer's work.”
Kotapuri's recommendation: get specialized training in AI programs and systems and build skills in areas such as data manipulation and analysis. He predicted that the typical clinical engineer will soon hold a title such as “clinical predictive systems engineer,” “clinical outcomes data analyst,” or “artificial intelligence clinical engineer.”
There will still be jobs in the field, he explained; they're just going to be different than what they are today. “Clinical engineers will not become extinct, but they will need to transform themselves if they're going to survive.”
Binseng Wang, ScD, CCE, FAIMBE, FACCE, vice president of program management with the clinical technology management division of Sodexo, also believes that HTM will see major changes in the coming years because of progress made in AI.
“Original equipment manufacturers of large, sophisticated equipment have already begun introducing self-diagnostics into their products to allow not only remote diagnoses but also prediction of failures,” he noted. As a result, these manufacturers are cutting back on troubleshooting training for their field service staff and advising them to follow “a strict playbook” for replacing parts in accordance with the built-in diagnostics, he said.
“They're gradually introducing AI in this process and soon will be able to pretty much do away with any technical training except for parts and assembly replacements, with a centralized, remote support center,” said Wang.
Promise and New Opportunities
The challenges and opportunities associated with leveraging data in healthcare also have been top of mind for another industry veteran. Bakul Patel, director of digital health in the FDA's Center for Devices and Radiological Health (CDRH), noted that healthcare has always had data and information.
“So what is new about digital healthcare other than it's healthcare driven by digital technologies? It's healthcare where, because of these technologies, data and information can now be made readily available to providers wherever they are,” he said.
According to Patel, the most pressing issue when it comes to digital health is actually a two-step problem. The first step is accessing data that are usable and representative, and the second is understanding the biases held by particular data held.
In AI, for example, “we have to know the limitations of the datasets that are being used,” Patel said. “You want to make sure that whatever you're training that machine with, fits the purpose and the intentions of that application.”
Data standardization has been defined by Observational Health Data Sciences and Informatics as “the critical process of bringing data into a common format that allows for collaborative research, large-scale analytics, and sharing of sophisticated tools and methodologies.”7 Standardization of data in the world of healthcare is important because it's the key to interoperability among different technologies. If the data in disparate systems aren't in a common format, exchanging information among those systems can be difficult or impossible.
Considerable progress has been made in this area thanks to organizations such as Health Level Seven International, which developed the standard HL7 FHIR (Fast Healthcare Interoperability Resources), for example, to facilitate data exchange between EHR systems and third-party applications. Still, say experts like Karen Giuliano, PhD, RN, MBA, FAAN, standardization is just one of several challenges when it comes to data sharing.
“HL7 has definitely made things a lot easier, but there is also the economic reality of a variety of vendors in the same space who are often competing with each other,” said Giuliano, who is joint associate professor at the Institute of Applied Sciences and College of Nursing at the University of Massachusetts Amherst. One company might make a patient monitor, for example, but it is likely that many of the devices connected to the monitor are made by other manufacturers.
“There are many original equipment manufacturer partnerships that are committed to interoperability and do a good job with bidirectional data exchange, but it would be nice to see a higher level of interoperability more widely available across all devices,” she said.
For data to be available and easily usable for clinicians, Giuliano added, it should be provided to them at the point of care and exactly when it is needed. “For example, if you are caring for a patient and you think they may be septic, what you would ideally like to be able to do is push a button to get all the lab values that might be relevant to that diagnosis, all the medications that the patient is on or was on, and all the relevant physical findings and vital signs that are predictive of sepsis.”
Unfortunately, she said, that information is much more likely to be stored in separate silos. “Because it's not packaged together, the clinician most commonly has to leave the bedside to a find the information they are looking for in a variety of different locations and systems, and that takes time,” which ultimately detracts from direct patient care.
Since coming to the University of Massachusetts Amherst, Giuliano's goal is to address some of these issues by focusing on improving the usability of devices and systems for clinician end users through interdisciplinary collaboration and the development of two laboratories.
“First, we are in the process of developing a healthcare innovation prototyping lab in the Institute for Applied Life Sciences, where design thinking, human factors engineering principles, and rapid iterative prototyping can be used for product creation,” explained Giuliano. “Second is to build a product usability lab at the University of Massachusetts Mount Ida campus that can be used for product testing with clinician end users, who are most often nurses. These two labs will also be designed to provide the infrastructure needed for additional clinical, academic, and industry collaborations outside of the university to further enrich innovation capabilities and facilitate a more rapid transition from research and development to clinical practice.”8
A third challenge related to data sharing and clinical decision making involves the data itself, said Kevin Lev, AI and advanced visualization marketing leader at Philips. An estimated 70% to 80% of medical data are unstructured or not organized in a way that makes it easy for a machine to process or (in the case of AI) allow its algorithms to work.9,,10
“We've seen situations,” Lev noted, “where commonly available clinical algorithms that had 90% efficacy during testing with particular datasets dropped to 50% or 60% efficacy when they were used in the real world.”
The culprit in such cases is often “data consistency,” Lev explained. “Hospital systems and physicians therein often label, measure, and quantify their data a little bit differently,” and those discrepancies can wreak havoc on an algorithm's output, especially when attempting to apply results from one group to the next.
Organizations may be able to retrain or “touch up” an AI algorithm to better fit their particular approach to data management, but even then they can still run into trouble, he said. “You have to remember, an algorithm is only as good as its coding and the means in which it was trained.”
Toward that end, Patel noted, CDRH published a discussion paper in 2019 on a proposed framework for regulating modifications to AI- and machine learning(ML)-based “software as a medical device” (SaMD).11 Technologies that use AI and ML, the paper begins, “have the potential to transform healthcare by deriving new and important insights from the vast amount of data generated during the delivery of healthcare every day.” The promise of AI- and ML-based software, it continues, has to do with its capacity for learning from real-world feedback (training) to adapt and improve performance.
“The challenge is in how you regulate these technologies,” Patel said. “How do we allow for this adaptability and change while also making sure the product remains effective and safe?”
The CDRH's proposal suggests that a premarket submission to the FDA be required when a modification to an AI/ML-based SaMD “significantly affects” device safety or effectiveness, changes the intended use of the device, or results in a “major change” to the software's algorithm. This regulation wouldn't apply to software that deploys algorithms that are trained only during development and “locked” before they reach hospital floors.
“It's a novel and innovative approach to a real problem,” Patel said. “We want these technologies to evolve over time, but it's also very important they do so in a positive way in order to improve patient care.”
One manufacturer with a possible interest in how FDA regulations eventually pan out is Philips. Among the company's many healthcare solutions with AI functions and features are intelligent workflow and maintenance functions in core imaging modalities; AI-empowered measurements of clinical anatomy and anomaly detection; an “advanced data integration, visualization, and analysis” diagnostic platform Philips calls the IntelliSpace Portal; AI-optimized research; and app deployment offerings.12 The Portal, explained Lev, is an SaMD product that helps intelligently automate complex analysis of DICOM (Digital Imaging and Communications in Medicine) images from computed tomography, magnetic resonance (MR), and other imaging modalities.
“There's a huge amount of hype around the capabilities of AI, and much of it is about what it can or will do in the clinical environment,” Lev said. Still, he noted, when it comes to deploying data, it may be the “para-clinical space” that stands to benefit the most. “I think a lot of the potential is in operational analytics and predictive analytics and applications in other areas of the hospital, including IT and HTM.”
The Philips MR system, for example, continuously collects and analyzes data pertaining to its magnet in order to give tech teams early warning of pending failures. In addition, Philips continually invests in new technologies that not only improve patient care pathways but also help hospitals benefit from operational optimization, such as “predicting patient no-show patterns or predicting whether patients will be compliant with their regimen for therapy and treatment,” Lev explained.
“Diagnostics and equipment maintenance are obviously important, but there are so many other things we can do with data,” he said. “If someone doesn't come in for their MR scan, that's a lot of revenue that is lost.”
Addressing the Present and Looking to the Future
It was the need for better operational efficiency, in fact, that led the FDA, back in May, to issue an Emergency Use Authorization (EUA) for an AI-powered predictive analytics system that it hoped would help health systems win the fight against COVID-19.13 Built by the U.S.-Israeli manufacturer CLEW, the platform uses AI-based algorithms trained to identify respiratory failure or hemodynamic instability hours before a patient's condition begins to visibly deteriorate.
Digital Health Center of Excellence. www.fda.gov/medical-devices/digital-health-center-excellence
Medical Device Data Systems, Medical Image Storage Devices, and Medical Images Communications Devices.www.fda.gov/media/88572/download
Proposed Regulatory Framework for Modifications to Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD). www.fda.gov/files/medical%20devices/published/US-FDA-Artificial-Intelligence-and-Machine-Learning-Discussion-Paper.pdf
Its ML models, according to the company, were developed using data from nearly 100,000 patients who previously spent time in hospital intensive care units (ICUs). “Healthcare providers,” said CLEW chief executive officer Gal Salomon in announcing the FDA authorization, “need more than simple analytics.”14 Their system, he continued, integrates into the clinician's workflow to put actionable data in their hands. “The CLEWICU platform is designed to enable healthcare providers to monitor patient predicted risk levels across all units in real time, allowing for smart decision making about clinical resource allocation, ensuring prompt, proactive, and efficient patient care,” said Salomon.
In its EUA, the FDA noted that, as of yet, no devices have been approved that can provide predictive screening for complications associated with COVID. “During ICU hospitalization, continuous monitoring results in the accumulation of vast amounts of multidimensional data,” the agency said. “Without predictive analytics, floor management and discharge decisions are determined based on patients' current state of health.” CLEW's AI software would give providers a chance to get ahead of the virus. By identifying their highest-risk patients, they could prioritize treatment for those who needed it most.
It's use cases just like that—where the data in a health system has immediate applicability—that make G. Wayne Moore, MBA, FASE, excited about what the future may bring.
“2020 has certainly been interesting,” said Moore, who is president and chief executive officer of Acertara Acoustic Laboratories in Longmont, CO. “There's been a lot of great movement in this area, and I'm certain we'll see that accelerated in 2021.”
Moore's company specializes in diagnostic ultrasound, and one of its products—a testing technology—uses AI to estimate when an ultrasound probe may need to be replaced or repaired. HTM professionals use the tool to make smart decisions about their inventory, he said.
A member of the Medical Imaging and Technology Alliance (MITA), Moore sits on the group's Artificial Intelligence Committee, which is advocating for the development of policies and regulations favorable to AI adoption in the industry. He said he sees little standing in the way of progress as companies like his own turn to the data to improve their technologies and healthcare overall.
“One thing we really have to focus on is not just getting data for these systems, but getting good data,” said Moore. “So the key question now is: How do we define good data? And the companies that are developing these systems, can they agree on these definitions?”
MITA and other industry organizations are working to develop consensus standards that will settle questions such as these, Moore said. “We're pretty far along in the process, but we're not there yet.”
In HTM, Moore added, he believes the sky is the limit when it comes to how data—and AI and analytics—might help an HTM professional do their day-to-day work.
“On the clinical side of things, there are radiologists and other physicians who fear these technologies may threaten their jobs. But the biomeds, they're overwhelmed; they've got a billion things to do no matter what. I'd think that anything that can help them successfully and quickly manage these tasks is probably going to be welcome,” he said.
Moore predicted that HTM professionals will see major changes to their field in the coming years. They're going to have to learn new skills, and they'll have to become familiar with AI-based systems and understand how they interact with other devices on their hospital networks. Still, he thinks, most are going to thrive in this data-drenched world, and there's plenty of time for all to adapt.
“This is really just a continuation of things that have been in motion for some time,” said Moore. “This is a process rather than an event.”
References
About the Author
Chris Hayhurst is a freelance reporter and writer. More information can be found at www.chrishayhurst.com.