The Perils of Biased Data
Data is one of the most essential resources in health care. It informs and educates care providers so they can make effective, positive decisions that benefit their patients. But can healthcare data be biased?
According to “Quick Safety 23: Implicit bias in health care”, published by the Joint Commission, “There is extensive evidence and research that finds unconscious biases can lead to differential treatment of patients by race, gender, weight, age, language, income and insurance status.” The results of the bias have led to staggering inequities and disparities in healthcare among different racial and ethnic groups.
Whether intentional or unintentional, inherent bias is often built into the ways data is collected, analyzed, interpreted and distributed. In the abstract of a study published in Science magazine, researchers found that “Health systems rely on commercial prediction algorithms to identify and help patients with complex health needs. We show that a widely used algorithm, typical of this industry-wide approach and affecting millions of patients, exhibits significant racial bias.” This biased data then becomes a barrier to actions, strategies and measurable progress toward health equity.
As a Teach for America Corps member, I served as director of entrepreneurial studies at East Central High School Entre Magnet in Tulsa, Oklahoma. During that time, I saw many underprivileged families get evicted from their homes. Why? Time and again, they were bankrupted by medical bills. In addition, within zip codes just a one-and-a-half miles from my home, there was a 10-year disparity in life expectancy.
I wanted to find out why this was happening, and soon discovered that one reason was bad healthcare data based on entrenched socioeconomic factors that created biased AI in the system. As a result, many people experienced a lack of access to essential care as well as insurance coverage to pay for it.
So how does a healthcare organization operationalize data justice to embrace diversity, equity and inclusion, and ensure the healthcare data they are using (and generating) is unbiased? How do we ensure patients receive equitable care?
Understanding How Bias is Introduced
First, we need to understand the ways in which bias has intentionally and unintentionally been built into the full data lifecycle. Social Determinants of Health (SDoH)—the specifics of where people live, education levels, insurance coverage and employment status—can greatly affect quality of life, health risks and outcomes.
EHR clinical data typically includes SDoH content such as race, religion and geographic location, and also food insecurity, homelessness propensity and even credit issues. While technological advancements such as Big Data, advanced and predictive analytics and artificial intelligence (AI) have taken organizations to new heights, there are enormous risks that this information can multiply bias and discrimination.
Because biased representation of SDoH within data can hinder progress toward more equitable outcomes, data scientists must practice responsible AI by adjusting for naturally occurring bias, truly representing study populations and prioritizing the careful review of models for bias to ensure that they are functioning as planned and the results are accurate.
As expectations of stakeholders continue to rapidly evolve, the ability to advance equity as a core aspect of an organization’s mission has emerged as the signature driver of business sustainability in the 21st century. Data integrity is foundational to meeting these expectations. But when looking at an overall fix, the task at hand can be daunting. So what can be done?
Advancing More Ethical and Equitable Practices in Data Collection
Data unto itself can either inform or misinform. The most important initial strategy is to make sure you have accurate, quality data. It is the only way to make an informed decision. Securing the quality of your data foundation and data quality practices is the basis for building a strong and successful strategy.
You can also start with the end in mind. The gap between the data workbook creator (who handles the creation of the data) and the business leader (who uses the data) is wide. Each has distinctive silos of focus. Our greatest role is to bridge that divide and become a translator between the person capturing the data and the person using it. Once we know what process of data collection best serves both silos, and all the factors in between, we can plan backwards and create specific, effective capture tools and data types.
The opportunity is in front of us now to look at the individual silos of care delivery and take all that data into a centralized location to look at trend analysis. Think of treating each EHR like a snowflake. No two are the same. Being able to establish a true baseline for a patient—who may or may not be in an underserved population—is the only way to obtain accurate data about that person. Otherwise, we can’t trust the algorithms to make appropriate, informed decisions and eliminate the implicit bias.
Rich sources of data serve as a backdrop to understanding what is gleaned from existing patient records. Health history and medical encounters are required to set the baseline for prior treatment efficacy and disease progression. Patient data can originate from multiple sources, making it important to leverage technologies that bring together these types of data over time.
We also need to revamp the way data is captured. Some organizations are looking at self-reporting to obtain a more complex picture of the patient. Archaic methods often lumped patients into much more broad categories. The CDC recently fine-tuned its social vulnerability index scores, which can lead to much more defined data on behalf of the patient.
Realize that the need to tie together quality and equality is a key component of the push toward value-based care. We see this in the NCGQ HEDIS equity measures. When quality care meets equitable standards there is a financial incentive, creating a win-win situation for all involved.
Within the various business lines in your organization, look first at what the consumer pain points are, and build a financial case for creating and curating higher-quality data. One focus could be on race or ethnicity, but this doesn’t need to be the only effort undertaken. It is important to think of this as a balance between your motives (in this case data justice) alongside the objectives of the broader organization (profit margins). Aligning the two could yield better internal receptivity and results.
Most importantly, we need to work together. Collaboration between providers, payers and other stakeholders is essential to the goal of ensuring everyone can get quality, equitable health care.
Original post: https://www.medtechintelligence.com/feature_article/the-perils-of-biased-data/
Post publish date: