Bias in medicine research exists in datasets and in outcomes, according to a newly published report, “Fairness in Precision Medicine.”
The report, by Kadija Ferryman, PhD, and Mikaela Pitcan, supported by the Robert Wood Johnson Foundation, seeks to better understand how bias could impact biomedical research into precision medicine.
In 2017, Ferryman and Pitcan conducted 21 qualitative interviews, via phone or in person, with biomedical researchers, bioethicists, technologists, and patient advocates. In the semi-structured interviews, the researchers asked respondents to discuss the potential areas where bias could present itself. The respondents told Ferryman and Pitcan that datasets themselves can become unintentionally biased in a variety of contexts:
- Genetic data. The choices that researchers make about which kinds of data to include in a study hinge on the types of data to which they have access. “Bias through invisibility—such as lack of data on certain factors—can trigger discriminatory outcomes just as easily as explicitly problematic data,” write the authors.
- Electronic health records (EHRs). Not only do data collected in EHRs vary widely, there is also a disconnect between the purpose of the EHR—the collection of records for billing purposes—and the goals of precision medicine research. An EHR is not the same, say the authors, as a complete record of a patient’s clinical history, and the fragmentary nature of EHR data creates impediments to using these data for research.
- Diversity in participants and data types. Biomedical research has long failed to include representative samples of women and minority populations, so findings from studies may not be generalizable.
- Historical bias. Medical data themselves may have hidden historical biases; the researchers point to lung cancer screening guidelines that were developed from a study in which African American individuals made up only 4% of the study population. “Thus, data on who receives lung cancer screening is impacted by sample bias: African Americans may not be adequately represented in this data, not because they chose not to get screened, but because some may not have qualified for a screening,” write the authors.
- Analytical bias. Precision medicine research may become biased by researcher preferences or by which areas of research are seen as “important” and “fundable.”
Moreover, say the authors, bias may influence the outcomes of precision medicine for patients and the healthcare system:
- Discrimination and disadvantage. Medical interventions derived from precision medicine research could be used to discriminate against groups of people; if, for example, a health risk is present in a certain immigrant population, that group could be targeted for immigration restrictions based upon that risk. Furthermore, patient populations who face limitations in health literacy could have difficulty in accessing interventions to improve their outcomes. That fact could lead to precision medicine benefitting primarily the healthcare savvy.
- Individual responsibility versus structural influences. Precision intervention may focus too much on the individual and not enough on structural forces that shape health outcomes. “Despite the large body of evidence on how social factors influence health, interventions targeted at individuals can be preferable because tackling social problems is so complex,” say the authors.
According to Ferryman and Pitcan, although precision medicine has the potential to transform healthcare for the better, there remains work ahead to ensure that structural determinants of health are acknowledged and addressed and to institute policies that will keep data from being used to marginalize vulnerable people.
“Above all,” say the authors, “patients and the public need to be engaged in this endeavor both to guarantee its success and to hold other actors accountable for potential misuses or misunderstandings about when and how their data should be used.”