Without computers, it would take Jake Luo a lifetime to sort through the sheer amount of data that is integral to his work. An associate professor in UWM’s Zilber College of Public Health and College of Engineering & Applied Science, Luo focuses on identifying patterns in massive, sprawling electronic health records to highlight disparities in care.
One dataset in particular, from the National Inpatient Sample, contains the data of 7 million patients across multiple years, Luo said. Even regular computers struggle to process this amount of information. “Sometimes, if the data set is too large, you can’t get a result because of the memory limitation or the CPU limitations,” he explained.
That’s why artificial intelligence is a powerful companion for Luo’s work. AI was designed to handle huge amounts of data and identify patterns. With resources from UWM’s High Performance Computing Center, Luo employs advanced AI computing techniques to efficiently process massive datasets.
These AI models can identify subtle patterns in how different patient populations access and experience health care services and their health outcomes, helping us understand where disparities exist and how to address them. The end result is organized information that researchers can leverage to draw conclusions about the state of health care — and build a roadmap for improvement.
Finding the gaps
Access to health care varies widely in the United States. Income level, insurance status, location, race, sex and level of education can affect each person’s experience with preventive and emergency care.
Making health care more equitable starts with addressing the disparities. But to prompt real change, professionals need to identify the gaps and who is most affected.
Luo, who directs the UWM Center for Health Systems Solutions, approaches this widespread problem by digging into the data. As an expert in bioinformatics, he sifts through electronic medical records in order to spot patterns. These databases are huge and have many data points on each patient.
“All the details about the patient — what kind of treatment they had, what kind of drug they’ve been taking, what kind of diagnosis and the (clinician) notes are in the electronic health record,” Luo said. “We leverage this particular dataset to do a lot of different kinds of research.”
Highlighting disparities
In many projects, Luo begins by collaborating with clinical investigators — physicians who directly work with patients — to suss out patterns and develop hypotheses.
“For example, they might observe that certain patient groups have lower response to certain treatments, and some patient groups are not adhering to the treatment … protocol as well as other patient groups,” Luo said.
Then, using patient data, the researchers determine if the hypothesis is true or not. “Clinical investigators give us some hint about potential gaps and challenges,” Luo said. “And then we basically drill into those areas and look into the pattern to see if that’s true or not.”
Other times, Luo’s group works backward; they get access to a large dataset but have to use machine learning to detect patterns within it. Such was the case when they studied disparities in telemedicine during the COVID-19 pandemic. Using sophisticated machine learning algorithms, they analyzed several factors simultaneously, from clinical outcomes and treatment patterns to socioeconomic indicators, to identify which patient populations may be underserved. For example, when studying telemedicine adoption during COVID-19, their AI systems processed millions of patient interactions to detect usage patterns across different demographic groups, revealing previously unknown disparities in virtual care access.
“We pooled all the patients who used telemedicine and then generated a control group who did not use telemedicine and looked into the pattern of those patients to see, for example, whether a specific group actually adopted telemedicine better than the other groups,” Luo said.
Some of the data confirmed their hypothesis – that more educated patients were more likely to use telemedicine. Other patterns were less obvious and more surprising, Luo says. For example, female patients were more likely to meet with their doctor virtually than male patients, as the team revealed in a 2021 paper in the journal Applied Clinical Informatics.
In another project, Luo is working on an initiative with the Medical College of Wisconsin called OTO Clinomics. It aims to help researchers better understand individual risk factors for otolaryngologic diseases and treatment to provide better care. (Otolaryngology includes conditions like head and neck cancer, tonsillitis, reflux and hearing loss.)
In 2021, Luo contributed to a report in OTO Open about the socioeconomic factors that correlate with a chronic rhinosinusitis diagnosis at a specialized clinic (as opposed to the emergency room). His team found that patients at the clinic tended to be proportionally older, educated, white and female. Conversely, clinics saw fewer patients who were Black, male, and had lower income and education levels.
These findings correlated with national trends related to race and socioeconomic status in health care access. In this case, Luo’s team won’t be the one to address the gaps with potential solutions, but drawing attention to these disparities can set the stage for other researchers to explore ways to help different patient populations.
Improving the patient experience
In other projects, Luo is working on more direct improvements for the patient experience. Under a grant from the National Institutes of Health, he helped design an AI-enabled voice system to help patients report data.
When patients start a new medication regimen or join a clinical trial, they’re not always diligent about reporting their health data. For example, a person undergoing diabetes treatments might need to log their glucose levels every day in an online portal. Clinicians rely on this data to determine if a treatment is working, yet patients aren’t necessarily consistent when it comes to recording their biomarkers.
So Luo’s team is working with a small cohort of participants who agreed to bring home an Amazon Alexa device that can talk to them when it’s time for a check-in. “It provides a very natural interface for the patient to do this task,” Luo says.
Instead of requiring patients to sit down at a computer and type in information each day, it’s a lot easier to just chat with the device on the go. The AI-enabled software can have a conversation with the patient, prompting them to share incremental health details as needed. Unlike simple reminder systems, this AI can engage in more sophisticated interactions. For example, asking clarifying questions if a patient reports concerning symptoms or offering encouragement when they’re consistently tracking their health data. The goal is to create an easier reporting system that streamlines data collection for patients and clinicians alike.
