Preventing Bias in Medical Devices: Identifying Morally Significant Differences

This article refers to:

Materialized Oppression in Medical Tools and Technologies

Liao and Carbonell discuss the role of (supposed) racial differences and racism in two medical devices: pulse oximeters and spirometers. They show that what might seem like cases of mere bias, are in fact part of larger and pervasive patterns of oppression (Liao and Carbonell 2023). While in the case of pulse oximeters, differences in skin color were not sufficiently considered, in the case of spirometers differences were considered but turn out to be biased and incorrect. Both mistakes result in harms for people with dark(er) skin color and for people racialized as black, and these harms align with existing patterns of oppression. Liao and Carbonell conclude that there is a distinctive material dimension to oppression in healthcare.

These examples also raise the question: How can we identify which differences are morally significant for the design of medical devices, in order to prevent bias and oppression from becoming materialized? Liao and Carbonell do not answer this question. Below, we explore how morally significant differences can be identified and discuss the design of heart valve implants as an example.

MEDICAL DEVICES: ONE-SIZE-FITS-ALL VERSUS STRATIFICATION

Medical devices, especially prostheses and implants, are often designed according to a one-size-fits-all approach that allows for a single off-the-shelf available product (Søraa and Fosch-Villaronga 2020). As the example of the pulse oximeter shows, one-size-fits-all approaches are prone to bias and can materialize oppression. This happens because one-size-fits-all approaches (implicitly, if not explicitly) assume a certain end-user—characterized by social identities such as age, gender, and ethnicity—in the design process (Søraa and Fosch-Villaronga 2020). Therefore, these devices might work less well for people whose social identity does not correspond to that of the imagined end-user.

The challenges of one-size-fits-all approaches have been recognized and have given rise to so-called stratified approaches, by which medical devices can be designed and tailored to fit different groups. Yet this stratified approach comes with a new challenge: potential end-users must be categorized in groups, which—as the example of the spirometer shows—is also prone to bias and might result in harm to people grouped in biased or oppressive categories.

MORALLY SIGNIFICANT DIFFERENCES

To determine when differences are morally significant, we first return to the examples of the pulse oximeter and spirometer. We propose that the case of the pulse oximeter illustrates that grouping people based on characteristics where a morally significant causal relation can be identified is desirable because it helps to prevent bias and fight oppression. The pulse oximeter works less well for people with a darker skin tone. The morally significant difference is skin tone because there is a clear causal relation with the device’s functioning (i.e., melanin blocks the device’s light beam). Therefore, grouping people by skin tone (e.g., to allow for more testing) is defensible because it makes a morally significant distinction that helps to counter existing patterns of oppression.

The case of the spirometer illustrates that grouping people based on race is undesirable because such categories can be biased and therefore harmful and can even perpetuate oppression. Spirometers measure lung capacity and correction factors are applied to the test results on (amongst others) people racialized as black. However, there is no good scientific evidence to support the supposed biological (e.g. genetic) relation between lung capacity and race (Liao and Carbonell 2023). This is not surprising, considering that race is a social identity and not a biological trait per se (Neal 2017). Many supposedly race-related health disparities can be explained instead by differences in socio-economic status (relating to inequities in employment, living conditions, education, and access to healthcare) as well as by the negative effects of the lived experience of racism (including the impact of chronic stress) on health (Braddock 2021). Using race as a proxy for lung capacity is therefore imprecise, biased, and harmful (Neal 2017). In the case of the spirometer it disadvantages black patients who are less likely to receive the medical care they require. This example shows that biases can surface by looking for the causal explanation of a specific difference, rather than using a social identity as a proxy.

Therefore, we argue that:

1. if medical devices are tailored to different groups, differences should be defined by specific and morally significant characteristics (such as skin color or hormone levels) rather than using broader categories (such as race or gender) as proxies (cf. Neal 2017), and

2. the causal explanations for the significance of such specific differences should be scrutinized for (socio-historically rooted) biases (cf. Vyas, Eisenstein, and Jones 2020).

Being precise about what differences matter allows one to investigate if there is a plausible and morally significant biological or social explanation, rather than a (normative) bias possibly derived from oppressive (such as racist or sexist) roots. While we do not claim that such an approach is likely to eliminate bias by itself, we think it could be one of many needed strategies to reduce bias and prevent bias from becoming materialized in medical devices.

EXAMPLE: HEART VALVE IMPLANTS

This line of reasoning can be translated to the design of heart valve implants, where the question how to account for differences when stratifying users to different implants also is relevant. Valve implants are available in different types and sizes, which have different clinical benefits and downsides. Mechanical valve implants require risky anticoagulation therapy, while biological valve implants (made of dead porcine tissue) require replacement after several years.

One of the aspects to consider is whether sex and gender are morally significant differences for these valve implants. It is known that the physiology of the heart differs between (cisgender) women and men. The average size of the heart in cisgender women, including the circumference (annulus) of the aortic valve, is smaller, while for both sexes, the size and weight of the heart increase with age (St. Pierre, Peirlinck, and Kuhl 2022). Sex hormones—the levels of which can change throughout a person’s life—also play a role in cardiology. For example, a high level of estrogen, which is most common in cisgender women, reduces inflammation responses and promotes the elasticity of cardiovascular tissue (De Kort et al. 2021). It is also speculated that higher levels of female sex hormones can reduce the risk of failure of biological valve implants (Kulik et al. 2009). Moreover, treatment outcomes are dependent on sex and gender. For example, during pregnancy mechanical valve implants lead to an increased risk of thrombosis (Cho et al. 2021).

We propose that two morally significant differences between valve implant users are heart size and hormone levels, not sex or gender. This means that accordingly, stratification should be based on these morally significant characteristics. In relation to heart size, it is already common practice to fit the size of a valve implant to a user based on their heart size (Lamata n.d.), and—perhaps unsurprisingly—not based on sex or gender as a proxy, even though these also correlate to the size of the valve implantation site, as described above. One reason why this makes sense is that not only sex and gender but also age affects the size of the heart. While it might be clear that heart size is a morally significant difference, this might be less obvious—but as important—for hormone levels.

We propose that—as far as it is not common practice yet—the type of valve implant should be fitted to a user based on their sex hormone levels, and not based on sex or gender. As described above, it is thought that sex hormones can affect the outcomes of valve implants. Stratifying people based on sex hormone levels, rather than on sex or gender itself, prevents gender stereotyping that could for example occur by overlooking interpersonal differences and intrapersonal changes in hormone levels, and prevents wrongly categorizing all women into the same group. Moreover, it fosters attention to people whose hormone levels do not match that of the archetypical cisgender female or male body, for example transgender people undergoing gender-affirming hormone therapy. Especially for the transgender population, the effect of sex hormone levels on the cardiovascular system is insufficiently understood (De Kort et al. 2021). Such knowledge deficits are generally associated with stigmatization and could negatively influence the quality of healthcare for this group. Thus, more research into the effects of (sex) hormones on valve implant outcomes is needed. This could allow stratifying people based on sex hormone levels, prevent sexism and stigmatization, and promote safety for everyone.

To be clear, we do not mean to argue that categories like gender and race (or racialized groups) are entirely obsolete in the context of medical devices, but rather that they should be used with the utmost caution and need further operationalization and specification. Overall, we conclude that stratifying users of heart valve implants—as well as medical devices more generally—based on carefully identified, morally significant differences could be one of many needed strategies needed to reduce and prevent bias and oppression from becoming materialized in medical devices.

ACKNOWLEDGEMENTS

We would like to thank Sanne Peters for commenting on a draft version of this commentary.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This research was financially supported by the Gravitation Program “Materials Driven Regeneration (MDR)”, funded by the Netherlands Organization for Scientific Research (grant no. 024.003.013); and by the Horizon 2020 Research and Innovation Program “Ink-based hybrid multi-material fabrication of next generation implants (INKplant)”, funded by the European Union (grant no. 953134).