Hospitals and health systems are rolling out more tools to analyze and process data in an attempt to improve patient care-raising questions about when and how to integrate racial and ethnic data.

As the United States becomes more and more diverse, more and more Americans identify with more than one race or ethnicity, and racial data becomes more and more complex.

According to the U.S. Census, which was conducted every 10 years last year, the number of Americans who identify with at least two races has doubled in the past decade. According to the Census Bureau, the Census Bureau began letting people identify more than one race in 2000. New York Times. It is now the fastest growing race and ethnic category.

As the healthcare industry moves toward a more data-driven direction, this is a demographic change, and executives should put it first. For example, if analysis or artificial intelligence tools include whether the patient is black, white, or other races in its prediction, this may cause black and white patients to be confused.

Tina Hernandez-Boussard, Associate Professor of Biomedical Informatics, Biomedical Data Science and Surgery at Stanford University, said that multi-ethnic patients represent a growing population. Consider other data-driven tools.

She said that if health system and software developers use race-dependent algorithms or protocols without taking into account the multi-ethnic patient situation, such models may be unreliable for that patient population. This may weaken the trust of patients in the health system.

“It’s very complicated,” Hernandez-Boussard said. “By developing algorithms that were not specifically tailored to this growing population, we lost the trust of that community.”

Predict risk

In recent years, healthcare organizations have been investing in tools for evaluating data to mark patients who need additional care, those who are at risk of adverse outcomes, and those who may have other needs.According to a report, more than three-quarters of emergency and outpatient care organizations are using advanced analytics to ensure population health Polls From the School of Healthcare Information Management.

Some of these tools—from basic risk equations to advanced artificial intelligence—include race, but not always in the manner of the growing multi-ethnic population of the United States.

“How should we care most about individuals identified as multiple races?” Dr. Michael Simonov, director of clinical informatics at the hospital-supported data company Truveta, talked about risk calculators and predictive models that include race and ethnicity data. “This is an open question and a very active area of ??research.”

Some Risk prediction algorithmIt has been used in medicine for many years, requiring clinicians to report whether the patient is black or white as part of their calculations.

A tool for estimating a patient’s 10-year risk of atherosclerotic cardiovascular disease requires the user to select the patient’s race as “white”, “African American” or “other”, which may cause uncertainty for black and white patients Sex—especially if the patient chooses only one race on their intake scale, or if the doctor assumes race based on the patient’s appearance.

National Kidney Foundation and American Society of Nephrology Released An equation for estimating kidney function that does not include race-replaces the current version that asks whether the patient is black.One Calculator It is used to predict the risk of vaginal delivery if the patient had a C-section during the last pregnancy. The competition will also be cancelled this year.

“If doctors are trained to treat race as a risk factor, and the patients they encounter do not fall into a clean race category, then it will be difficult for them to make an assessment that they have been trained,” Stanford University Health Care Office Director Dr. Megan Mahoney, a clinical professor in the Department of Medicine at Stanford University, said.

“I don’t fit any clean category of their calculator,” added the black and white Mahoney.

Mahoney said she said she would like to see more data tools and calculators follow in the footsteps of the equation to estimate kidney function, regardless of race.

Next generation medicine

For many years, artificial intelligence has been touted as the future of healthcare, and if developers have the right data to use, it may provide an opportunity to integrate multi-ethnic and multi-ethnic data.

Unlike other analysis or modeling methods that tend to strictly collect specific types of data to calculate results, advanced artificial intelligence is more flexible—it can ingest more variables and complex and multi-layered data that has not been clearly programmed, Russ, Chief Health Information Officer, UCSF Health Dr. Cucina said.

but Good algorithms start with good data.

In order for an AI tool to generate generalizable insights, it needs to analyze a large amount of data that reflects the people that the tool will use.

In order to create an AI system, developers need to provide AI with a large amount of training data, from which they can learn to recognize features and draw patterns. However, if the data set is not diverse and lacks information about certain subgroups, the system’s predictions and recommendations may not be so accurate for these patient groups.

Health care institution and Advocacy group Whether to incorporate race data into algorithms is increasingly being challenged, arguing that race is inappropriately used as a proxy for other variables related to disease risk, such as ancestry, genetics, socioeconomic status, or the environment in which the patient lives.

They said it would be more appropriate to use these data instead of race.

However, even if race is not a variable in the algorithm, it is important to have a diverse data set that can be used to validate AI tools-so that organizations can test products against specific subgroups and ensure that they perform well in different demographics .

Dr. Peter Embi, President and CEO of Regenstrief Institute, said: “When we did not have a good representative data sample when developing these algorithms, we saw many examples of problems.” Embi joined Vanderbilt University of Medicine in January Center, serving as the head of the Department of Biomedical Informatics.

Take dermatology as an example, The researcher said The skin cancer detection artificial intelligence tool is mainly trained on images of light-skinned patients, and may not be so accurate for dark-skinned patients.

Suchisaria, a professor and director of the Johns Hopkins University’s Machine Learning and Healthcare Laboratory and the CEO of Bayes Health, said more research is needed to figure out under which conditions patients are noticed as multi-ethnic Or Nation will improve the accuracy of predictive tools, a company that develops artificial intelligence for clinical decision support.

Get the correct data
However, even accumulating enough data on multi-ethnic patients to train or validate AI systems is challenging.

Only about 10% of Americans are multiracial. This in itself is a diverse label, including whites and blacks, blacks and Asians, Asians and Native Americans, just to name a few—not to mention choosing patients of more than two races.

Patient data in medical records is often not captured sufficiently finely to identify patients of multiple ethnicities.

Based on Bayesian Health’s experience in handling hospital customer EHR data, Saria said that she suspects that multi-ethnic patients are underestimated in medical records.

She said that in the data used by the company, only about 1% of patients were recorded as having multiple ethnicities.

This may be because multi-ethnic patients are usually classified as “other”, or they may only choose one of the races they identify.

Collecting enough data for the research, development, and validation of analytics, artificial intelligence, and other data-driven tools will be the key to ensuring that they work effectively for patients with different backgrounds.

“If we do have data, then yes, the algorithm will be able to handle these issues appropriately,” Hernandez-Boussard said. “But the problem is that we don’t have data to train [algorithms] appropriate. “