"This Whole Thing Smacks of Gender": Algorithmic Exclusion in Bioimpedance-based Body Composition Analysis
““This Whole Thing Smacks of Gender”: Algorithmic Exclusion inBioimpedance-based Body Composition Analysis
Kendra Albert ∗ [email protected] Law SchoolCambridge, MA Maggie Delano ∗ [email protected] CollegeSwarthmore, PA ABSTRACT
Smart weight scales offer bioimpedance-based body compositionanalysis as a supplement to pure body weight measurement. Com-panies such as Withings and Fitbit tout composition analysis asproviding self-knowledge and the ability to make more informed de-cisions. However, these aspirational statements elide the reality thatthese numbers are a product of proprietary regression equationsthat require a binary sex/gender as their input. Our paper combinestransgender studies-influenced personal narrative with an analysisof the scientific basis of bioimpedance technology used as part ofthe Withings smart scale. Attempting to include nonbinary peoplereveals that bioelectrical impedance analysis has always rested onphysiologically shaky ground. White nonbinary people are merelythe tip of the iceberg of those who may find that their smart scaleis not so intelligent when it comes to their bodies. Using body com-position analysis as an example, we explore how the problem oftrans and nonbinary inclusion in personal health tech goes beyondthe issues of adding a third “gender” box or slapping a rainbowflag on the packaging. We also provide recommendations as to howto approach creating more inclusive technologies even while stillrelying on exclusionary data.
CCS CONCEPTS • Human-centered computing → Ubiquitous and mobile com-puting ; •
Social and professional topics → Gender ; Race andethnicity ; People with disabilities; Geographic characteristics; Cul-tural characteristics;
Remote medicine ; •
Applied computing → Consumer health . KEYWORDS data collection and curation, sex/gender, bioelectrical impedanceanalysis, body composition, critical data/algorithm studies, scienceand technology studies, critical HCI and the design of algorithmicsystems
ACM Reference Format:
Kendra Albert and Maggie Delano. 2021. “This Whole Thing Smacks ofGender”: Algorithmic Exclusion in Bioimpedance-based Body CompositionAnalysis. In
Conference on Fairness, Accountability, and Transparency (FAccT ∗ Both authors contributed equally to this research.Permission to make digital or hard copies of part or all of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for third-party components of this work must be honored.For all other uses, contact the owner/author(s).
FAccT ’21, March 3–10, 2021, Virtual Event, Canada © 2021 Copyright held by the owner/author(s).ACM ISBN 978-1-4503-8309-7/21/03.https://doi.org/10.1145/3442188.3445898 ’21), March 3–10, 2021, Virtual Event, Canada.
ACM, New York, NY, USA,11 pages. https://doi.org/10.1145/3442188.3445898
Kendra:
As a nonbinary person who was assigned female at birth,I’ve always had an uncomfortable relationship with my body weight.To be honest, before I stepped on the Withings scale, I hadn’t weighedmyself in some time. But after undergoing some medical transitionsteps, I found myself much more curious about what my body fatpercentage was. Especially since I have put on a fair amount of muscle,I was interested to see how the numbers matched or didn’t match myself perception.Because of this discomfort with weight numbers, I asked Maggieto make a profile for me when I was getting started with the scale.She started going through the registration flow for a new user, andquickly encountered a roadblock - the system required a gender.Wait, that’s not right. It didn’t require a gender, it required you topick one of two images labeled gender - a person wearing pants or aperson wearing a skirt (see Figure 1). When Maggie asked me whichone I preferred, I told her to pick one and not tell me.When I did finally step on the scale (with my shiny new profile), Ilooked at the numbers and felt...well, some sort of way. To be honest,I’m happier with my body now than I remember ever being before,but the numbers seemed very off.The next morning, we were talking about the scale at breakfast,and that was when Maggie first told me that the fat percentage wasan estimation based on a technique called bioelectrical impedanceanalysis (BIA). There was an equation behind that number - it didn’tactually measure my body fat percentage directly. I was shocked, andasked how that related to my gender. We decided to do some testing.When Maggie changed my “gender” from skirt-wearing to pants-wearing, my body fat percentage dropped 10% points. 10! The hugedifference seemed to confirm everything I’d thought about the numbersfeeling wonky. But which one was right? Or failing that, which onecloser? Could we look at how the algorithm calculated the percentagein order to see which one was likely to be more accurate for me?
Maggie:
Kendra’s reaction to the scale was an interesting expe-rience because I realized that the knowledge I had about how BIAworked was non-obvious to anyone not familiar with the tech. I hadalready been thinking about how sex is often coded in the underlyingBIA equations as 1 or 0 because I had given a guest lecture about itrecently, but I hadn’t fully thought through the implications for transand nonbinary people.Seeing the numbers on the scale jump by so much was jarring. Wetalked a lot about how to interpret the results and what the exactpercentages really mean. There is a little progress bar on the scalethat indicated that the numbers (regardless of gender) were above a r X i v : . [ c s . C Y ] J a n AccT ’21, March 3–10, 2021, Virtual Event, Canada Kendra Albert and Maggie Delano
Figure 1: The Withings profile view in their Health Mate mo-bile application. the “normal range.” But that normal range itself appears to be a“normal range” for an athlete because we were both well above thatline. In terms of making the scale work for Kendra, the answer I wasproposing was to pick skirt-wearing or pants-wearing and then trackthat number over time.Kendra’s interest in using the Withings smart scale, and figuringout what “gender option” was right drove us to reexamine how BIAworks as a technology, and how assumptions about gender and sexare built into the fundamental equations that drive the algorithmicmodels used for estimating body fat percentage.
This paper, the result of that analysis, combines transgenderstudies-influenced personal narratives with an analysis of the sci-entific basis behind the bioimpedance technology used as part ofthe Withings smart scale. With these different forms of knowledge,we explore how the problem of trans and nonbinary inclusion inpersonal health tech goes beyond adding a third “gender” optionor slapping a rainbow flag on some packaging. Rather, nonbinary exclusion is the tip of an iceberg of flawed assumptions and exclu-sionary clinical testing, resulting in algorithms that are advertisedfor self-knowledge but prove to allow anything but.
This paper draws on previous work related to self-tracking, trans-gender studies, human-computer interaction, and the study of sexand gender in biomedical research. In this section, we provide abrief summary of related work in these disciplines to situate ourfindings.While the Withings weight scale is not the first commerciallyavailable scale to estimate body fat percentage, the device was oneof the original “smart” (i.e. connected) weight scales [3]. The devicewas first sold in the early 2010s at the beginning of a surge of interestin self-tracking and the advent of the “quantified self” movement[47, 55, 57]. The quantified self movement included a variety ofstakeholders including individual self-trackers, organizations suchas Quantified Self, companies and toolmakers, academic researchers,and physicians (with considerable overlap between these categories)[9]. Participants are broadly interested in the capabilities of self-tracking to provide unique, actionable, and personalized insights[50, 82]. Self-trackers engage deeply with their data through aprocess sociologist Deborah Lupton refers to as “data sense-making”[49]. Many self-trackers believe that data can “speak for itself”and should be involved in medical care [27, 59]. However, the useof data collected from commercial “wellness” devices such as theWithings scale or activity trackers like Fitbits is controversial asthese devices don’t always perform well in third-party validations,and often involve proprietary algorithms and metrics (e.g. steps,sleep scores).Previous research investigating self-tracking devices and wear-able monitors has shown that these devices, like devices in othercategories, are designed primarily for an unmarked user [12, 15].That is, the user is assumed to be a White, cisgender, non-disabledman. Cifor and Garcia, for example, use duoethnography to eval-uate gendered assumptions in the design and app of the JawboneUP3 fitness tracker. They illustrate that while the device itself ap-peared “genderless,” the design of the device and the app reinforcedmasculinist values such as competition [12]. Such issues are alsopresent in the design of algorithms - for example, Fitbit devicescount steps less reliably at slower gait speeds and with softer steps,which decreases step count accuracy for older people or peoplewith mobility related disabilities [24, 37]. Early implementations ofthe hardware and algorithms used to estimate heart rate on wear-ables were less accurate for users with darker skin tones [31, 69],though recent evidence suggests these disparities may have beenaddressed by improvements to the device’s algorithms [6].The development of algorithms without a diverse set of userscreates algorithmic exclusion . Populations are excluded from theuse of algorithmic tools because they were not included as partof the original data used in development, or because informationwas not gathered in such a way as to make their needs visible.This algorithmic exclusion means that the performance of thesealgorithms for individuals not in the original dataset are unknown;the practical implication is that these algorithms likely work lesswell for those not included in the original dataset. Algorithmic
This Whole Thing Smacks of Gender”: Algorithmic Exclusion in Bioimpedance-based Body Composition Analysis FAccT ’21, March 3–10, 2021, Virtual Event, Canada exclusion can have real world impacts as individuals rely moreand more on these data, especially when these data are used byphysicians. For example, pulse oximeter devices that measure bloodoxygenation (using a more involved technique similar to that usedby wearables manufacturers for measuring heart rate) overestimateblood oxygenation in individuals with darker skin tones [7, 25].Renewed interest in these disparities during the COVID-19 epidemicled to a study that showed that Black patients had nearly threetimes the frequency of occult hypoxemia (oxygen depravation) thatwas not detected by pulse oximetry than White patients [52, 71],potentially leading to higher mortality rates for Black patients whenthe seriousness of their COVID-19 cases were underestimated.These issues have not escaped notice within communities thatbuild technological tools. There has been increasing discussion indifferent design communities about how to create technology thatis more inclusive and/or addresses some of the disparities discussedabove. In human-computer interaction (HCI) and artificial intelli-gence (AI) research, for example, there have been efforts includinganalytical reviews, software analysis of datasets, and guidelinesabout increasing “gender sensitivity” [11, 33, 39, 63, 64] and moreintersectional approaches to addressing disparities, such as the in-tersection of gender and race [10, 62]. There have been multipleguides and recommendations for including transgender people inuser interface design [11, 53] and in surveys [74]. These recom-mendations include allowing all individuals to self-identify theirgender, not assuming binary gender, using the language users use,and protecting the privacy of participants. In the case of dealingwith medical research and “embodiment,” the guidelines recom-mend measuring physiological parameters such as hormone levelsdirectly, rather than assuming them based on gender.However, embodiment is a tricky line to draw. When one consid-ers the terms “sex” and “gender,” the common assumption is thatsex is biological and gender is social. If there is any relationshipbetween the two, it is assumed that sex influences gender, and trans-gender and intersex people are seen as outliers whose needs varyfrom “normal” populations. However, Springer et al. argue that it is sex that cannot be purely decoupled from social factors (i.e. gender)[75]. A “biosocial turn” is now beginning in the study of sex andgender [68]. Many mechanisms that were previously thought tobe due to biological “sex” differences, are in fact mechanisms thatinvolve differences based on socialization that manifest in biologicaldifferences. Springer et al. recommend using “gender” to refer tosocial relations and gender roles and the term “sex/gender” to referto those biological and biosocial factors associated with individualphysiology. In this paper, we will use the terms sex/gender andgender, unless we are referring to how these terms are used in aspecific work.
Our work draws heavily from transgender studies as an approach,while having some similarity to Black feminist methods, specificallyJennifer Nash’s love politics in the form of witnessing [56]. Weinclude conversations between the two of us throughout throughoutthe paper. Personal narrative, especially dialogue, can help uncover“common pain points and overlooked opportunities” [12]. Whereduoethnography, used by previous studies, is a research method that employs personal narrative to “simultaneously generate, interpret,and articulate data” about a shared experience [58], we includepersonal narratives throughout this paper to combine, in the wordsof Susan Stryker, “the embodied experience of the speaking subject”(i.e. our experiences using the weight scale) with “the specializeddomain of scholarship”, (i.e. the specifics of the theory and practiceof BIA for at home body composition monitoring) [72, 77]. Personalnarrative allows for a starting point to a broader conversation aboutsmart weight scales and the implications the system and algorithmdesign have for technology and biomedical research more broadly.We are approaching this topic as a White nonbinary person(Kendra), and as a White cisgender woman (Maggie). Neither of usare disabled in ways that are likely to affect our readings from orinteractions with the Withings scale. Both of us have considerablebackground in technology and gender. Kendra is a lawyer teachingat a technology clinic who also teaches in transgender studies. Theyhave, at times, engaged in self-tracking, although not previouslyaround weight. Maggie is an assistant professor at a small liberal artsschool where she teaches digital/embedded systems and inclusiveengineering design. Her research involves using bioimpedance tohelp patients manage fluid overload. She is also a self-tracker andhas presented her work at several Quantified Self conferences.
Critical analysis of the sort that we deploy in this paper requiresthe knowledge of how the measurement technology inside smartweight scales works. In this section, we present a background onBioelectrical Impedance Analysis (BIA). We should note, however,that because the specific testing and equations used by the With-ings scale are not publicly available, this background will leverageknowledge from public and peer-reviewed sources and may ormay not reflect the specific approaches that the Withings or otherconsumer-facing scales employ.At the most basic level, the body can be divided into two main“compartments:” fat mass (FM) and fat free mass (FFM) [40, 43, 46].FM includes all the fat in the human body, including what we thinkof as body fat and also visceral fat around vital organs. FFM is therest of the tissue; it is about 73% water, about 7% bone, and the restis proteins.BIA leverages the fact that the water in FFM is conductive; bydriving a small, painless current through the body via a pair ofelectrodes (in weight scales these are two of the electrodes on thescale), the resulting voltage can be measured by another pair ofelectrodes (also on the scale) and related to the electrical propertiesof the underlying tissue.If one assumes the body is a homogeneous cylinder with cross-sectional area A, the measured resistance 𝑅 (defined as the real partof the measured voltage divided by the current) is equal to: 𝑅 = 𝜌𝐿𝐴 (1)where 𝜌 is the conductivity of the cylinder, 𝐿 is the length ofthe cylinder, and 𝐴 is the cross-sectional area of the cylinder. MostBIA equations assume that 𝐿 is proportional to the body height 𝐻 .Multiplying both sides of the equation by 𝐿 / 𝐿 , the resistance can AccT ’21, March 3–10, 2021, Virtual Event, Canada Kendra Albert and Maggie Delano be related to the inverse of the volume, assuming that 𝑉 = 𝐿 × 𝐴 : 𝑅 = 𝜌𝐿𝐴 · 𝐿𝐿 = 𝜌𝐿 𝑉 (2)If one moves the volume to the other side, there is then: 𝑉 = 𝜌𝐿 𝑅 (3)This volume 𝑉 corresponds to what is called the “total bodywater” or the volume of all water in the body, which is assumedto be about 73% of the volume of the FFM. If one multiplies thisvolume by the presumed density of the FFM to obtain the FFM, theFM and body fat percentage (BF%) can then be calculated as: 𝐹𝑀 = 𝑊 𝑒𝑖𝑔ℎ𝑡 − 𝐹 𝐹𝑀 (4) 𝐵𝐹 % = 𝐹𝑀𝐹𝑀 + 𝐹 𝐹𝑀 ·
100 (5)
The methods described above require a number of assumptionsrelated to the body. In order for these assumptions to be valid, theresistivity 𝜌 of the measured volume must be homogeneous, andthe cross-sectional area must be constant throughout the body suchthat 𝑉 = 𝐿 × 𝐴 . The assumption that the FFM is 73% water mustalso hold, along with the assumed density of the FFM. Finally, itmust be assumed that the current penetrates through the wholebody in a uniform manner such that the estimated volume is trulyreflective of the total body water, and not just a fraction of it.Of course, these assumptions are not realistic; the body is nota single homogeneous cylinder with precisely known body com-position. Instead, BIA leverages different variables that correlatewith “gold standard” estimations of the FFM to estimate the FFMbased on the BIA itself. An example BIA equation for estimatingFFM might look like the following: [44]: 𝐹 𝐹𝑀 = − . +( . × 𝐻 / 𝑅 )+( . × 𝑤𝑒𝑖𝑔ℎ𝑡 )+( . × 𝑋 )+ ( . × 𝑠𝑒𝑥 : 𝑚𝑒𝑛 = , 𝑤𝑜𝑚𝑒𝑛 = ) (6)This equation involves a number of key terms: the 𝐻 / 𝑅 term, theweight term, the 𝑋 term (reactance or imaginary part of measuredbioimpedance), and sex. Each of these terms is associated with acoefficient in front (along with a constant at the beginning of theequation) that are calculated based on the best fit of the regressionequation that minimizes the error between the estimation via BIAand the estimation via the gold standard for the population undertest (in this case, the gold standard used was a technology calleddual x-ray absorptiometry or DXA).Precisely which parameters are included in the regression equa-tions and their corresponding coefficients depends on the popu-lation used to calculate the equations and researcher preference.Other researchers also include factors such as the age of the partic-ipants and whether or not participants are athletes [40]. In somecases these parameters are all incorporated into a single equation(such as the one above that has “options” for participant sex), or mul-tiple equations are generated, such as one for “lean” participants,one for “average” participants, and one for “obese” participants[66]. Parameters included in FFM estimation equations often do a lotof “work” and their role is not always clearly understood. Theseparameters and their coefficients “stand in” for things such as bodydensity, which can vary depending on both factors included in theequations and those typically excluded from the equations such asethnicity. For example, age is sometimes included as a parameterbecause there tends to be a decrease in FFM and an increase in FMwith age, and sex is included because males on average have lowerFM than females. We unpack these assumptions and coefficientparameters in more depth in Section 6. Kendra:
Of course, I didn’t know how BIA worked before using thescale. Nor would looking at the Withings website have revealed anyof the fraughtness of BIA to me - when I look at their ads now, theycall the technology “body composition.” It’s not obvious from theiradvertising that it’s estimating body fat percentage based on a set ofassumptions and an algorithm, rather than providing an individual-level ground truth. If you don’t know how the technology works, it’sactually quite easy to draw the conclusion that the scale just magicallyknows your actual body fat percentage.Even if I review the “technical specifications,” the information con-tained requires quite a bit of context to determine that what is producedisn’t an individualized number. The bullet points say “BioelectricalImpedance Analysis / Athlete and non-athlete mode / Unit: body fat %,total body water %, muscle mass kg or lb, bone mass kg or lb.” There’snothing there that tells me, as an end-user without a lot of expertise inBIA, that it’s engaged in an estimation based on plugging particularvalues into an equation.That brings me to the question, Maggie, what were you thinkingwhen you bought the Withings scale? How did the body compositionstuff play into it?
Maggie:
I’ve had the scale for a long time - since 2012. That wasalso the time when the number of people talking about self-trackingwas growing, and organizations such as Quantified Self were facilitat-ing some of the first meetups and conferences in this area. QuantifiedSelf emphasized self-knowledge through numbers, often using theframe “self-research” or “personal science” [82]. Over the next fewyears, the idea of self-tracking would become very hyped, and anentire commercial ecosystem Whitney Erin Boesel dubs “quantifiedself” (i.e. little q, little s, vs big Q, big S) was formed [9]. Lookingback at my data, my first weigh in was March 23rd, 2012. I wanted tolearn more about the tech that was out there and see what I could doto make sense of things, and was also inspired by a Quantified Self“Show & Tell” talk by Amelia Greenhall about how she weighed herselfeveryday and sustained her weight loss long term [30]. I was excitedabout the possibility of self-tracking and consistent habits improvingmy fitness and my life. I wanted to learn more and then translate thatknowledge to help others.
Kendra:
This idea of self-knowledge is really exciting. That’s whatI was hoping for when I stepped on the scale as well - some numbersto help me quantify what I was feeling about my body. But of course,that’s not what I got. As a White nonbinary person, what I learnedis that this tech isn’t build for me - in part because of the choicesthat technology companies make, and in part because of the failureto meaningfully account for transgender experience as part of the
This Whole Thing Smacks of Gender”: Algorithmic Exclusion in Bioimpedance-based Body Composition Analysis FAccT ’21, March 3–10, 2021, Virtual Event, Canada underlying clinical testing. And it’s worse for non-White nonbinaryor intersex folks, who are both not represented in the studies in termsof sex/gender or race/ethnicity. So much for smart scales.
Many limitations of BIA have been well established in the medicalliterature [40], though some researchers argue that the techniquesare still appropriate to use under the correct circumstances [80].Researchers suggest caution when using BIA, especially when work-ing with populations that have altered body composition, such aspatients with fluid overload. In these cases, some researchers havedeveloped equations specifically for a particular patient population,or have used alternative methods of body composition assessmentthat don’t rely on regressions (see e.g. [38]).A major challenge with body composition using BIA is that thetwo compartment model of “fat mass” (FM) vs “fat free mass” (FFM)inherently requires assumptions about the composition of the twodifferent compartments (in addition to other assumptions suchas homogeneous composition as discussed previously in Section4.1). Uniformity of the FM is a fairly reasonable assumption acrossindividuals [51], but assumptions about the composition of the FFMare not [16]. Additionally, when using the regression equations, theassumptions about the composition of the FFM are obscured intothe coefficients associated with the included variables such as age,weight, and sex.Because the linear regressions are optimized at a populationlevel and most studies do not examine the accuracy of the estima-tions at an individual level, there is no guarantee that a specificequation is accurate for any given individual. Additionally, onceone begins to consider any individual that is not perfectly matchedfor the population that was used to create the equations, the role ofthese variables becomes increasingly murky but also increasinglyimportant in order to design equations that work for populationshistorically not included in the populations used to generate theequations. This includes non-White people, trans and nonbinarypeople, intersex people, people with chronic illnesses, and those atthe intersections of these categories.
We begin with the assumptions and role of “sex” in the equations. Sex in BIA is either coded as 0 for female and 1 for male (effectivelychanging the offset of the equations, as in Equation 6) or thereare separate equations created for male participants and femaleparticipants, as in Sun et al. [78]. What the literature means by“male” or “female” is unclear, and these terms are often confoundedwith gender identities of “man” and “woman” as in Equation 6. Aswe discussed in Section 2, “sex” as a concept is just as fraught andcontingent as gender [18, 23, 75]. This is not a problem unique to(or caused by) the existence of trans (or intersex) people.Although the methods by which “sex” was evaluated in the BIAliterature is unclear, it is common for reported sex to be a partici-pant’s sex assigned at birth. And “sex assigned at birth” is generallyonly a proxy for someone’s external genitalia at birth, which is onlyone of the many characteristics that are often encompassed under We use the term “sex” here given that this is the term used by the researchers. However,sex and gender are entangled as described in Section 2. the word sex [18, 23]. Others include hormone balance, chromoso-mal makeup, and internal reproductive organs. We could not findan example of a study that produces BIA estimates that discusseswhat sexual characteristics they round up to a determination of sex,and it is generally not clear how the identification of sex was made(i.e. whether self-identified or identified by the researchers). Thislack of specificity is one of the first and most significant barriers tocreating a more inclusive algorithm for transgender people. Givenhow large the sample sizes were for some of the populations usedto create BIA equations (upwards of 5,000 in [45]), it is statisticallyunlikely that no transgender people were involved. But we don’tknow how they were accounted for or counted in the calculations.The use of “sex” itself is also a “stand in” for other parameters.What the researchers presumably intend is for “sex” to stand infor the roughly dimorphic variation in body composition betweenthose assigned male at birth and those assigned female at birth [18].However, it is not the fact that a person is assigned male at birth(i.e., they have genital erectile tissue of a certain size) that makesthose assigned male at birth have lower FM on average comparedwith those assigned female at birth. In fact, FM and the distributionof fat may actually be biosocial phenomena: they may depend onboth sex AND gender (sex/gender). For example, one effect of largerquantities of testosterone is changes in fat distribution and increasesin muscle production, resulting in lower body fat percentages [73].However, most research into sex differences in body compositiondo not explore other explanations, such as the differences in dietbetween men and women in some cultural contexts [36, 61]. TheBIA literature also does not disaggregate sociocultural context thatmight be caught up in the word “sex”, nor account for the myriadways in which someone assigned male at birth might not match thepresumed physiological composition associated with those assignedmale at birth on average.With only two sex or gender options, some intersex and mostnonbinary users experience what D.E. Wittkower has termed adysaffordance – they have to pretend to be something they are notin order to access the service [15, 81]. The lack of transparency inthe BIA equations makes it difficult to tell exactly what adjustmentsmight need to be made to make equations more inclusive, or to atthe least advise individuals as to which “sex” they should choosewhen using consumer technologies that incorporate BIA. On theother hand, the current setup of the Withings scale that asks foruser gender may actually produce more accurate results for transwomen and trans men who have undergone gender-affirming hor-mone therapy, as their body composition may resemble that of a ciswoman or cis man, rather than their sex assigned at birth [41, 73].
White nonbinary or intersex people, despite experiencing a dysaf-fordance, may be better off than their siblings of color becausethe composition of the study populations used in BIA equations inthe Withings scale are unclear, and there is no way to enter racialor ethnic information. The majority of measurement studies in-volving BIA include primarily Caucasian subjects and assume that We use both the terms race and ethnicity because the literature is mixed on whichare relevant factors, likely because of the entanglement of biological and social factorssimilar to sex/gender as discussed in Sections 2 and 6.
AccT ’21, March 3–10, 2021, Virtual Event, Canada Kendra Albert and Maggie Delano
Caucasian subjects are to serve as a “reference” for other ethnici-ties. This is an issue because body composition can vary amongethnic and racial groups due to the environment, diet, cultural fac-tors, and anthropomorphic measurements such as limb length andbody size [19]. Researchers or device manufacturers interested inusing BIA equations validated in a Caucasian population thereforeneed to cross-validate the equations in the population of study.As with sex, race or ethnicity is likely standing in as a proxy forsome other variables, such as environmental racism, socioeconomicstatus, or some actual relationship between FFM composition andcertain genetic factors. Some studies suggest that ethnicity specificcompensations are needed to use body composition equations indifferent ethnic groups [13, 65], whereas other studies have shownthat stratification based on adiposity (i.e. body fat percentage) ismore important than race [4]. Regardless, cross-validation studiesare necessary to ensure that the assumptions of equations produceaccurate results for all populations.Another major issue with BIA is that the regression equationsthemselves are validated against “gold standards” that in turn havetheir own assumptions. For example, BIA equations are often com-pared against air displacement plethysmography or hydrostaticweighing. But both of these techniques require assumptions aboutthe density of the FFM. The density of the FFM depends on numer-ous factors such as age, sex, and ethnicity [22]. Therefore any “goldstandards” that likewise depend on the two-compartment model (i.e.FM and FFM) are themselves are only valid in the same population,and most of those are also primarily tested on White (presumablycis and binary gender) subjects.Even techniques that do not depend on the two-compartmentmodel such as dual x-ray absorptiometry (DXA) are found to signif-icantly underestimate fat mass when compared with CT scans [42].Ultimately, the only way to validate a body composition device isusing cadaver studies and chemical analysis, which have some ofthe same issues - the results would then only be validated for thosesimilar to the cadavers [70]. Although measurement problems ofthis type are common in all areas of science, the lack of physiologi-cal basic science about how and why FFM varies with respect toboth social and biological factors in an intersectional way meansthat it is difficult to determine which assumptions will hold acrosspopulations. Because of this, the lack of population-specific testingin each individual “gold standard” testing regime complicates thepossibility of meaningful validation for folks who do not fit thedefault.
The gap between promises and reality in the smart medical devicespace is in part due to the limited scope of review by the UnitedStates’ Food and Drug Administration (FDA). The vast majorityof medical devices, including “smart” and “connected” devices, areapproved through a program at the FDA known as the 510k ap-proval process. This process requires no clinical trials and very littleoversight - device manufacturers merely need to prove that theirproduct is “substantially equivalent” to that of an already approved The use of White people as a default mirrors Kimberle Crenshaw’s critiques of USlaw, as well as Ruha Benjamin’s critique of photographic film [5, 17]. medical device. The Withings scale discussed in this paper received510(k) clearance [1]; similar scales on the market such as the FitbitAria scale received approval even after the devices were alreadyon the market [2]. In 2014, the FDA announced their intentionsto cease requiring even 510(k) approval for devices such as smartweight scales. This means that there is very little regulation ofthese devices, and certainly no required clinical validation of thealgorithms used to calculate body composition, despite the label of“clinically tested” on the Withings website.This lack of regulatory oversight results in most consumer-focused deployments of technologies like BIA being “black box”algorithms. Popularized in the context of technology studies byscholar Bruno Latour, an algorithm is considered a black box “whena statement is simply presented as raw fact without any referenceto its genesis or even its author.” [34]. When using the Withingsscale, only the final body fat percentage is made available, and thereis no explicit reference to an algorithm in the app or in most ofthe marketing materials. Additionally, Withings does not releasethe equations its scales use to calculate FFM or the populationsthat it used to calculate those equations. The power of the blackbox is that we cannot thoroughly investigate a subject about whichnothing is known [60]. We are left with the assumption that, unlessproven otherwise, Withings’ internal algorithm production mirrorsthe biases of the research at large, but again, it is impossible totell. All that we know are the inputs that Withings asks users for(binary gender, height, age and athlete status).We can draw some conclusions even with just publicly availableinformation. The binary approach to the gender question withoutany explanatory text both erases nonbinary people, and does not askthe right question about the embodiment of the user. The Withingsscale does not prompt the user to enter information related toethnicity, so it is not possible that the scale is using an equationthat adjusts variables to compensate for different FFM factors inracial or ethnic groups. Because of the marketing of the technology,non-White users may not even know that it might be relevant. TheWithings scale’s algorithm is, in the words of Ruha Benjamin, ananti-Black box [5].Furthermore, what evidence we do have points to BIA being unre-liable even on the populations that it is theoretically well positionedfor. We reviewed the list of studies that Withings shared on theirpage for researchers and found only one study that specifically eval-uated the body composition aspects of the scales [14]. The resultsof the study compared the performance of the newer Body Cardioscales with an air displacement plethysmography device called theBod Pod (a “gold standard” discussed in Section 6.2). The meanabsolute percentage error of the body fat estimation of the BodyCardio weight scale compared with the Bod Pod was greater than25%, well above a previously established 1.5% threshold deemedas an acceptable error [14]. This suggests that any results fromthe Withings scales should be interpreted with extreme caution,even on the target population who is most well represented in thestudies likely used to create equations. It is important to note, however, that while air displacement plethysmography isoften used as a comparison device/gold standard, it relies on assumptions that cancompromise its effectiveness as we discuss in Section 4.1. Withings also argues thatthese scales should be used to indicate trends rather than for absolute assessment [83].
This Whole Thing Smacks of Gender”: Algorithmic Exclusion in Bioimpedance-based Body Composition Analysis FAccT ’21, March 3–10, 2021, Virtual Event, Canada
Kendra:
The distance between Withings’ promise of self-knowledgeand the reality of regression equations is upsetting. They advertise allof these benefits to self-quantification, but it’s actually limited by thetechnology [26]. As with many algorithmic technologies, the creationof regression equations based on a limited sample cannot and will notcreate accurate self-knowledge amongst those who do not fit withinthose samples.If the scale was actually individually calculating a ground truthnumber, as opposed to using a regression based on height and skirt-wearing vs. pants-wearing, being nonbinary wouldn’t matter! To behonest, 90% of the time when I’m asked my gender or sex, it doesn’tmatter. It’s an arbitrary box checking exercise. So why would theWithings scale be different?There’s this sleight of hand involved in not revealing to people howthe technology works that creates the situation in which a nonbinaryperson could go in expecting self-knowledge and getting a predictiontotally disconnected from what they themselves could tell you abouttheir body. I could have told the scale more about me to make theanswer more accurate, but that wasn’t an option. I don’t necessarilymind sharing information about my hormone status, my sex assignedat birth, or other facts about my body if they’re actually useful. Andalthough my views on volunteering race are shaped by my member-ship in a privileged racial group, I suspect many users would prefer toshare race or ethnicity information if it meant that they would getmore accurate results.Withings asks for my gender, but it doesn’t want it, even asidefrom the app’s confusion between gender and sex. I know things aboutmyself that are relevant to its guessing, but there’s no way to translatethis knowledge within the limited frame of reference produced bythe clinical trials. There’s no way to come out with a more accuratepicture or contest the bounded understanding of the system. That feelserasing, even more than the mere existence of the binary prompt.It makes me wonder about all of the other random gender/sexrequests that I’ve encountered when using technologies around health.Does Fitbit want my gender/sex for calorie burning calculation? DoesApple? What about the Ring Fit? How deep does the rabbit hole go?
Trans and nonbinary people of all races deserve to have accessto inclusive self-tracking technologies that do not collapse theiridentities or force them to ignore relevant self-knowledge. Whatcan be done to improve these technologies? We evaluate threeoptions for how to handle sex/gender under the hood of a BIA-calculating device such as the Withings scale, and then provideoverall recommendations as to how to handle the use of regressionequations based on limited medical testing.It would be inappropriate to continue without noting that BIAas a technology deployed in smart scales may be fundamentallyinseparable from the fatphobic medical and cultural context thathas created concerns about body fat in the first place [48]. Withingsmay claim that “there is more to you than just weight” (see Figure2), but the subtext of its advertising indicates that you should wantto weigh less. That is not fixable through recommendations around the handling of sex (or gender, or hormone status). It might be rea-sonably asked - given all its flaws, should BIA be used in consumerdevices at all? We don’t seek to answer that question in this paper.Our aims are more modest. Nonbinary folk and transgender folksdeserve access to technologies of self-knowledge, even as thosetechnologies may be used both by companies and individuals tosuggest harmful body standards.We provide a series of options for making weight scales based onBIA more inclusive, with recommendations for users, and both man-ufacturers and researchers. Most of our recommendations specifi-cally focus on sex/gender, but it is worth noting that overall, BIAalso has a long way to go when it comes to race/ethnicity, whichwe leave for future work to explore in more depth. Although ourrecommendations are designed based on the specific context of BIAsmart scales, they could potentially apply to many areas wherebinary sex/gender is used as part of an algorithmic system.
Figure 2: A portion of the front page of the Withings websiteadvertising a version of their smart scale.
One option for making systems more inclusive of nonbinary peoplewould be the elimination of sex/gender as a variable, using oneequation for all users. Practically speaking, this is not difficult.Manufacturers would not have to acquire new data, only re-runthe regression to find the best fit without sex/gender as a variablein the regression equation(s).However, a drawback of this option is that elimination of sex/ gender as a variable for all users would result in readings that
AccT ’21, March 3–10, 2021, Virtual Event, Canada Kendra Albert and Maggie Delano were less accurate in aggregate, as the inclusion of sex does reducethe mean error of the regression equations [40]. Given the lack ofaccuracy of the technology as a whole, this may or may not behugely significant - however, users who find a sex/gender binaryappropriate for their bodies might be upset to lose accuracy in orderto make the technology more inclusive.
The next option is to add a third option to the menu where onecurrently chooses sex/gender. One method of implementing a non-binary option would be the optional elimination of sex/gender as avariable as listed above for people who select a third option. Therewould then be three options: “male,” “female,” and “sex/gender neu-tral.”This option could be helpful for some intersex people, nonbinarypeople who have not medically transitioned but who would preferpotentially less accurate results to having to volunteer a binarymarker, nonbinary people who have taken some medical transitionsteps, and anyone else for whom the binary sex options are unlikelyto produce accurate results. A cautionary note: having a third optionin the menu does not a nonbinary inclusive app make. As AnnaLauren Hoffman explains in her generative work on data violenceand inclusion, without taking meaningful steps to change the powerdynamics present in the product, inclusion is at best, lip service[35]. For example, when Facebook allowed for self-identificationwith a wider variety of gender identities on their platform, they didnot fundamentally change the binary advertising logic of male orfemale, making their claims of nonbinary inclusion questionable[8]. Thus, adding a third option is only appropriate if there is anunderlying algorithmic change.
Ultimately, the ideal outcome of this work would be for the field totake a step back and consider the role that sex/gender are playingas a “stand in” for things like body fat distribution and anthropo-morphic information. This is exactly the kind of work that HCIresearchers have recommended when considering trans embodi-ment [11, 53]. But it is more difficult in this case than many othersbecause of how pervasive assumptions about sex and gender are inclinical research [54, 75, 79]. The full, complicated role that sex andgender play in BIA equations and beyond are not well understood.Significant fundamental research is necessary to begin to under-stand which additional factors to measure and how to measurethem in cost-effective and reliable ways.A deeper understanding of sex/gender and body compositionwill require “slow science” [76]. With more information about therole that these factors are playing, additional information could beprovided by end users - everything from body shape (i.e., “apple”,“pear”) to hormone status. This information could even be madeoptional to not place an additional burden on those unfamiliarwith the specifics or who want to do the basics. Fundamentally, thisapproach is the most well aligned with the promises that companiessuch as Withings make about their technologies, but would alsorequire the most fundamental research.
First, some advice to transgender, nonbinary, and intersex peoplewho wish to use technology that incorporates BIA but presentsbinary sex options. Based on studies looking at changes in bodycomposition, hormone status is a very significant variable for bodycomposition of the bodily characteristics, perhaps more significantthan other variables that are encompassed by the word sex [21, 73].So we would recommend that if folks must pick a binary category,they pick the one most closely aligned with their hormonal balance -male if in a normal male range for testosterone, female if not. In anycase, because of the study populations used to produce equationsand the black box nature of these algorithms, the actual valueproduced is unlikely to be accurate and should be used primarilyto track change over time rather than for its absolute value. (Thisrecommendation also holds true for any person of any gender usingthe scale.)In Table 1, we lay out additional recommendations for researchersand manufacturers who wish to build more inclusive regression-based technologies. Elaboration on some of these recommendationsand additional recommendations for different contexts can be foundin the following references: [11, 28, 53, 54, 74, 75, 79].In general our recommendations can be summarized as a) ac-knowledge the existence of non-gender normative people, b) makefewer assumptions, and c) explain in more detail the limitationsof technology. Of course, it may be difficult for companies to fullybe honest about measurement accuracy and precision. But if beingexplicit and honest with customers about these errors and assump-tions would make them think twice about purchasing a product,perhaps the best next step is for companies to reconsider theirbusiness model.Of course, including trans people after the fact is not ideal. Par-ticipatory methods that incorporate transgender people in problemdefinition around medical devices - to design, in the words of Haim-son et al., trans technologies - would be preferable to all of thesestopgap measures [15, 28, 32, 67]. However, as practitioners whowork with participatory methods, we understand that such prac-tices are unlikely to arise overnight. Until it’s widely accepted thatdesigning for those on the margins can create better medical de-vices, participatory design may never fully adopted by those whocommercialize them.
It can be easy to assume that the use of “sex” in quasi-medicalapplications is neutral, just another fact about one’s body thatallows for a more accurate complete picture. But, in the immortalwords of dril, “this whole thing smacks of gender” [20]. When thelived realities of nonbinary folks cause us to scratch below thesurface, the lack of careful thought around assumptions that gointo technologies like smart scales becomes clear. Cultural beliefsabout gender are driving the bus when it comes to engagement with“sex differences.” And because of that, sex, even in clinically testedBIA equations, is holding space for too many variables, supportedby too little basic research.When inadequately validated, (anti)-Black box algorithms builton these shaky foundations deceive their users. They harm people
This Whole Thing Smacks of Gender”: Algorithmic Exclusion in Bioimpedance-based Body Composition Analysis FAccT ’21, March 3–10, 2021, Virtual Event, Canada
Table 1: Recommended best practices for trans, nonbinary and intersex inclusion in regression based technologies such asBIA. Table design inspired by [54].Context Marginalizing Practices Inclusive Practices
For manufacturersand researchers Assume sex is purely biological and gen-der is purely social. Review literature on sex/gender and select the appropriate measures forthe specific project. If none are available, consider conducting more basicresearch, and consider and articulate the limitations of the state of theart.Assume a biosocial explanation for physiological differences unless evi-dence clearly suggests otherwise.Ignore the existence of trans, non-binary, and intersex people. Acknowledge the existence of trans, non-binary, and intersex people andhow their physiology or experiences might be different from other users.Acknowledge, if needed, the limitations of the current results and howtrans and nonbinary people can still obtain the best results for them.Analyze results only at the populationlevel. Analyze results for sub-groups and at the individual level.Evaluate results across racial and ethnic groups, implementingrace/ethnicity selection or inclusion of non-proxy variable as appro-priate.For manufacturers Elide the measurement precision andassumptions. Explicitly state the precision of the measurement system, along withassumptions and constants used.Represent sex and/or gender with pic-tograms. Use clear terminology based on the underlying research or known physi-ology to select a term.Be explicit why sex and/or gender are being used so that trans, non-binaryand/or intersex users can choose the best option for them.For researchers Assume “gold standard” has no built-inassumptions. Discuss the assumptions embedded in the gold standard methods them-selves, and how those assumptions influence the results.Assume the collection of sex and/or gen-der information is obvious or straight-forward, and therefore does not need tobe discussed in depth. Include whether reported sex and/or gender was based on self-report,and if so, what options were available for participants to choose between.Were there only two options? Was there a free response option?Explicitly state how sex and gender are being used and what they standin for. Example: “sex is a stand-in for the dimorphic distribution of bodyfat in the human population.”who do not line up with the assumed user base, promising knowl-edge of self but instead merely reproducing the violence of erasingclinical systems [35].It doesn’t have to be this way. Even without additional clinicaltesting or regulation, there are clear steps that manufacturers cantake to mitigate some of the harms caused by these systems. First,they can educate users as to the population-level accuracy of metricslike BIA, rather than advertising body composition analysis as ifit was accurate on an individual basis. Second, as discussed above, they can make clear how the technology does or does not workon transgender or nonbinary people, while also identifying otherfactors (such as race and/or ethnicity) that make results more or lessaccurate. Finally, manufacturers could release as much informationabout their equations as possible, including validation studies, inorder to facilitate cross-validation by independent researchers.Admittedly, these solutions may reify technical expertise andserve to legitimize the ideas of these types of body measurement,
AccT ’21, March 3–10, 2021, Virtual Event, Canada Kendra Albert and Maggie Delano as Greene, Stark, and Hoffman, point out in their work on tech-nological ethics [29]. Ultimately, BIA is just one example of howregression-based body measurements, whether implemented intechnologies like smart scales or described in scientific papers,harm those who are not presumed to be the ideal. And that wholething is worth overturning.
ACKNOWLEDGMENTS
Thank you to Siobhan Kelly for their perceptive comments on lipservice, and to the many essential workers, particularly those atPhilly Foodworks and South Square Market, whose labor allowedus to write (and eat).
REFERENCES
Medicine & Science in Sports & Exercise
29, 6 (June 1997), 781–787.[5] Ruha Benjamin. 2019.
Race after Technology: Abolitionist Tools for the New JimCode . Polity Press, Newark.[6] Brinnae Bent, Benjamin A. Goldstein, Warren A. Kibbe, and Jessilyn P. Dunn. 2020.Investigating Sources of Inaccuracy in Wearable Optical Heart Rate Sensors. npjDigital Medicine
3, 1 (Feb. 2020), 1–9. https://doi.org/10.1038/s41746-020-0226-6[7] Philip E. Bickler, John R. Feiner, and John W. Severinghaus. 2005. Effects of SkinPigmentation on Pulse Oximeter Accuracy at Low Saturation.
Anesthesiology
New Media & Society
19, 6 (2017), 880–898.https://doi.org/10.1177/1461444815621527[9] Whitney Erin Boesel. 2013. What Is the Quantified Self Now?https://thesocietypages.org/cyborgology/2013/05/22/what-is-the-quantified-self-now/.[10] Joy Buolamwini and Timnit Gebru. 2018. Gender Shades: Intersectional Accu-racy Disparities in Commercial Gender Classification. In
Proceedings of MachineLearning Research , Vol. 81. 15.[11] Sabrina Burtscher and Katta Spiel. 2020. "But Where Would I Even Start?":Developing (Gender) Sensitivity in HCI Research and Practice. In
Proceedings ofthe Conference on Mensch Und Computer . ACM, Magdeburg Germany, 431–441.https://doi.org/10.1145/3404983.3405510[12] Marika Cifor and Patricia Garcia. 2020. Gendered by Design: A DuoethnographicStudy of Personal Fitness Tracking Systems.
ACM Transactions on Social Com-puting
2, 4 (Jan. 2020), 1–22. https://doi.org/10.1145/3364685[13] S. H. Cohn, C. Abesamis, I. Zanzi, J. F. Aloia, S. Yasumura, and K. J. Ellis. 1977.Body Elemental Composition: Comparison between Black and White Adults.
TheAmerican Journal of Physiology
Journal of Personalized Medicine
10, 1 (March 2020), 17.https://doi.org/10.3390/jpm10010017[15] Sasha Costanza-Chock. 2020.
Design Justice: Community-Led Practices to Buildthe Worlds We Need . MIT Press, Cambridge.[16] K. D. Côté and W. C. Adams. 1993. Effect of Bone Density on Body CompositionEstimates in Young Adult Black and White Women.
Medicine and Science inSports and Exercise
25, 2 (Feb. 1993), 290–296.[17] Kimberle Crenshaw. 1989. Demarginalizing the Intersection of Race and Sex: ABlack Feminist Critique of Antidiscrimination Doctrine, Feminist Theory andAntiracist Politics.
University of Chicago Legal Forum
Contexts
16, 1 (Feb. 2017), 80–80. https://doi.org/10.1177/1536504217696082[19] Paul Deurenberg, Jan A. Weststrate, and Jaap C. Seidell. 1991. Body Mass In-dex as a Measure of Body Fatness: Age- and Sex-Specific Prediction Formulas.
British Journal of Nutrition
65, 2 (March 1991), 105–114. https://doi.org/10.1079/BJN19910073 [20] dril. 2012. "’This Whole Thing Smacks Of Gender,’ i Holler as i OverturnMy Uncle’s Barbeque Grill and Turn the 4th of July into the 4th of Shit".https://twitter.com/dril/status/213849618415484929.[21] Jolanda M H Elbers, Henk Asscheman, and Jacob C Seidell. 1997. Reversal of theSex Difference in Serum Leptin Levels upon Cross-Sex Hormone Administrationin Transsexuals.
Journal of Clinical Endocrinology and Metabolism
82, 10 (1997),4.[22] Kenneth J. Ellis. 2000. Human Body Composition: In Vivo Methods.
PhysiologicalReviews
80, 2 (Jan. 2000), 649–680. https://doi.org/10.1152/physrev.2000.80.2.649[23] Anne Fausto-Sterling. 2000.
Sexing the Body: Gender Politics and the Constructionof Sexuality (1st ed.). Basic Books, New York, NY.[24] Lynne M. Feehan, Jasmina Geldman, Eric C. Sayre, Chance Park, Allison M. Ezzat,Ju Young Yoo, Clayon B. Hamilton, and Linda C. Li. 2018. Accuracy of FitbitDevices: Systematic Review and Narrative Syntheses of Quantitative Data.
JMIRmHealth and uHealth
6, 8 (Aug. 2018), e10527. https://doi.org/10.2196/10527[25] John R. Feiner, John W. Severinghaus, and Philip E. Bickler. 2007. Dark SkinDecreases the Accuracy of Pulse Oximeters at Low Oxygen Saturation: TheEffects of Oximeter Probe Type and Gender.
Anesthesia & Analgesia
International Journal of Communication arXiv:1908.06165 [cs] (Aug. 2019). arXiv:1908.06165 [cs][29] Daniel Greene, Anna Lauren Hoffmann, and Luke Stark. 2019. Better, Nicer,Clearer, Fairer: A Critical Assessment of the Movement for Ethical ArtificialIntelligence and Machine Learning.
Proceedings of the 52nd Hawaii InternationalConference on System Sciences
Proceedings of the 2020 CHI Conference on Human Factorsin Computing Systems . ACM, Honolulu HI USA, 1–13. https://doi.org/10.1145/3313831.3376669[33] Foad Hamidi, Morgan Klaus Scheuerman, and Stacy M. Branham. 2018. GenderRecognition or Gender Reductionism?: The Social Implications of EmbeddedGender Recognition Systems. In
Proceedings of the 2018 CHI Conference on HumanFactors in Computing Systems . ACM, Montreal QC Canada, 1–13. https://doi.org/10.1145/3173574.3173582[34] Graham Harman. 2010.
Prince of Networks: Bruno Latour and Metaphysics . re.press.[35] Anna Lauren Hoffmann. 2020. Terms of Inclusion: Data, Discourse, Violence.
New Media & Society (Sept. 2020), 146144482095872. https://doi.org/10.1177/1461444820958725[36] Lotte Holm and M. Møhl. 2000. The Role of Meat in Everyday Food Culture: AnAnalysis of an Interview Study in Copenhagen.
Appetite
34, 3 (2000), 277–283.[37] Dragana Javorina. 2020.
Investigating the Validity of the Fitbit ChargeHRTM forMeasuring Physical Activity in Children and Youth with Mobility-Related Disabili-ties . Thesis.[38] D. F. Keane, P. Baxter, E. Lindley, U. Moissl, S. Pavitt, L. Rhodes, and S. Wieskotten.2017. The Body Composition Monitor: A Flexible Tool for Routine Fluid Man-agement across the Haemodialysis Population.
Biomedical Physics & EngineeringExpress
3, 3 (2017), 035017. https://doi.org/10.1088/2057-1976/aa6f45[39] Os Keyes. 2018. The Misgendering Machines: Trans/HCI Implications of Auto-matic Gender Recognition.
Proceedings of the ACM on Human-Computer Interac-tion
2, CSCW (Nov. 2018), 1–22. https://doi.org/10.1145/3274357[40] Sami Khalil, Mas Mohktar, and Fatimah Ibrahim. 2014. The Theory and Fundamen-tals of Bioimpedance Analysis in Clinical Status Monitoring and Diagnosis of Dis-eases.
Sensors
14, 6 (June 2014), 10895–10928. https://doi.org/10.3390/s140610895[41] M Klaver and G T’Sjoen. 2018. Changes in Regional Body Fat, Lean Body Massand Body Shape in Trans Persons Using Cross- Sex Hormonal Therapy: Resultsfrom a Multicenter Prospective Study.
European Journal of Endocrinology
The BritishJournal of Radiology
82, 974 (Feb. 2009), 123–130. https://doi.org/10.1259/bjr/80083156[43] Ursula G. Kyle, Ingvar Bosaeus, Antonio De Lorenzo, Paul Deurenberg, MarinosElia, Jose Manuel Gomez, Berit Lilienthal Heitmann, Luisa Kent-Smith, andJean-Claude Melchior. 2004. Bioelectrical Impedance Analysis-Part I: Review ofPrinciples and Methods.
Clinical Nutrition
23, 5 (Oct. 2004), 1226–1243. https://doi.org/10.1016/j.clnu.2004.06.004
This Whole Thing Smacks of Gender”: Algorithmic Exclusion in Bioimpedance-based Body Composition Analysis FAccT ’21, March 3–10, 2021, Virtual Event, Canada [44] Ursula G. Kyle, Laurence Genton, Laurie Karsegard, Daniel O. Slosman, andClaude Pichard. 2001. Single Prediction Equation for Bioelectrical ImpedanceAnalysis in Adults Aged 20 –94 Years.
Nutrition
17, 3 (2001), 6.[45] Ursula G. Kyle, Laurence Genton, Daniel O. Slosman, and Claude Pichard. 2001.Fat-Free and Fat Mass Percentiles in 5225 Healthy Subjects Aged 15 to 98 Years.
Nutrition
17, 7-8 (July 2001), 534–541. https://doi.org/10.1016/S0899-9007(01)00555-X[46] H. C. Lukaski. 2013. Evolution of Bioimpedance: A Circuitous Journey fromEstimation of Physiological Function to Assessment of Body Composition and aReturn to Clinical Research.
European Journal of Clinical Nutrition
67, S1 (Jan.2013), S2–S9. https://doi.org/10.1038/ejcn.2012.149[47] Deborah Lupton. 2016.
The Quantified Self . Polity, Cambridge, UK.[48] Deborah Lupton. 2018.
Fat . Routledge.[49] Deborah Lupton. 2018. How Do Data Come to Matter? Living and Becomingwith Personal Data.
Big Data & Society
5, 2 (July 2018), 2053951718786314.https://doi.org/10.1177/2053951718786314[50] Deborah Lupton. 2020. Data Mattering and Self-Tracking: What Can PersonalData Do?
Continuum
34, 1 (Jan. 2020), 1–13. https://doi.org/10.1080/10304312.2019.1691149[51] A. D. Martin, M. Z. Daniel, D. T. Drinkwater, and J. P. Clarys. 1994. AdiposeTissue Density, Estimated Adipose Lipid Fraction and Whole Body Adiposity inMale Cadavers.
International Journal of Obesity and Related Metabolic Disorders:Journal of the International Association for the Study of Obesity
Obstetrics & Gynecology
Quantified: Biosensing Technologies in Everyday Life .MIT Press, Cambridge, MA.[56] Jennifer C. Nash. 2019.
Black Feminism Reimagined: After Intersectionality . DukeUniversity Press, Durham.[57] Gina Neff and Dawn Nafus. 2016.
Self-Tracking . MIT Press, Cambridge, MA.[58] Joe Norris. 2008. Duoethnography.
The SAGE encyclopedia of qualitative researchmethods
BMC Medicine
14, 1 (Aug. 2016), 118. https://doi.org/10.1186/s12916-016-0670-y[60] Frank Pasquale. 2015.
The Black Box Society: The Secret Algorithms That ControlMoney and Information . Harvard University Press, Cambridge, MA.[61] Ritva Prättälä, Laura Paalanen, Daiga Grinberga, Ville Helasoja, Anu Kasmel,and Janina Petkeviciene. 2007. Gender Differences in the Consumption of Meat,Fruit and Vegetables Are Similar in Finland and the Baltic Countries.
EuropeanJournal of Public Health
17, 5 (2007), 520–525.[62] Morgan Klaus Scheuerman, Stacy M. Branham, and Foad Hamidi. 2018. SafeSpaces and Safe Places: Unpacking Technology-Mediated Experiences of Safetyand Harm with Transgender People.
Proc. ACM Hum.-Comput. Interact.
2, CSCW,Article 155 (Nov. 2018). https://doi.org/10.1145/3274424[63] Morgan Klaus Scheuerman, Jacob M. Paul, and Jed R. Brubaker. 2019. HowComputers See Gender: An Evaluation of Gender Classification in CommercialFacial Analysis Services.
Proceedings of the ACM on Human-Computer Interaction
3, CSCW (Nov. 2019), 1–33. https://doi.org/10.1145/3359246[64] Morgan Klaus Scheuerman, Kandrea Wade, Caitlin Lustig, and Jed R. Brubaker.2020. How We’ve Taught Algorithms to See Identity: Constructing Race andGender in Image Databases for Facial Analysis.
Proceedings of the ACM onHuman-Computer Interaction
4, CSCW1 (May 2020), 1–35. https://doi.org/10.1145/3392866[65] J. E. Schutte, E. J. Townsend, J. Hugg, R. F. Shoup, R. M. Malina, and C. G.Blomqvist. 1984. Density of Lean Body Mass Is Greater in Blacks than in Whites.
Journal of Applied Physiology: Respiratory, Environmental and Exercise Physiology
56, 6 (June 1984), 1647–1649. https://doi.org/10.1152/jappl.1984.56.6.1647[66] K. R. Segal, M. Van Loan, P. I. Fitzgerald, J. A. Hodgdon, and T. B. Van Itallie. 1988.Lean Body Mass Estimation by Bioelectrical Impedance Analysis: A Four-SiteCross-Validation Study.
The American Journal of Clinical Nutrition
47, 1 (Jan.1988), 7–14. https://doi.org/10.1093/ajcn/47.1.7[67] Norman Shamas and Afsaneh Rigot. 2019. Looking At The Margins: IncorporatingHarm Reduction Into Tech - from Radical Networks 2019.[68] Heather Shattuck-Heidorn and Sarah S. Richardson. 2019. Sex/Gender and theBiosocial Turn.
The Scholar and Feminist Online
15 (2019).[69] Anna Shcherbina, C. Mikael Mattsson, Daryl Waggott, Heidi Salisbury, Jeffrey W.Christle, Trevor Hastie, Matthew T. Wheeler, and Euan A. Ashley. 2017. Accuracy in Wrist-Worn, Sensor-Based Measurements of Heart Rate and Energy Expen-diture in a Diverse Cohort.
Journal of Personalized Medicine
7, 2 (June 2017), 3.https://doi.org/10.3390/jpm7020003[70] John Shepherd, Bennett Ng, Markus Sommer, and Steven B. Heymsfield. 2017.Body Composition by DXA.
Bone
104 (Nov. 2017), 101–105. https://doi.org/10.1016/j.bone.2017.06.010[71] Michael W. Sjoding, Robert P. Dickson, Theodore J. Iwashyna, Steven E. Gay,and Thomas S. Valley. 2020. Racial Bias in Pulse Oximetry Measurement.
NewEngland Journal of Medicine
Berkeley Women’sLaw Journal
18 (2003), 15–37.[73] Cassandra Spanos, Ingrid Bretherton, Jeffrey D Zajac, and Ada S Cheung. 2020.Effects of Gender-Affirming Hormone Therapy on Insulin Resistance and BodyComposition in Transgender Individuals: A Systematic Review.
World Journal ofDiabetes
11, 3 (March 2020), 66–77. https://doi.org/10.4239/wjd.v11.i3.66[74] Katta Spiel, Oliver L. Haimson, and Danielle Lottridge. 2019. How to Do Betterwith Gender on Surveys: A Guide for HCI Researchers.
Interactions
26, 4 (June2019), 62–65. https://doi.org/10.1145/3338283[75] Kristen W. Springer, Jeanne Mager Stellman, and Rebecca M. Jordan-Young. 2012.Beyond a Catalogue of Differences: A Theoretical Frame and Good PracticeGuidelines for Researching Sex/Gender in Human Health.
Social Science &Medicine
74, 11 (June 2012), 1817–1824. https://doi.org/10.1016/j.socscimed.2011.05.033[76] Isabelle Stengers. 2018.
Another Science Is Possible: A Manifesto for Slow Science .John Wiley & Sons.[77] Susan Stryker. 2013. (De)Subjugated Knowledges.
The Transgender Studies Reader (2013), 1–17.[78] Shumei S Sun, W Cameron Chumlea, Steven B Heymsfield, Henry C Lukaski, DaleSchoeller, Karl Friedl, Robert J Kuczmarski, Katherine M Flegal, Clifford L Johnson,and Van S Hubbard. 2003. Development of Bioelectrical Impedance AnalysisPrediction Equations for Body Composition with the Use of a MulticomponentModel for Use in Epidemiologic Surveys.
The American Journal of ClinicalNutrition
77, 2 (Feb. 2003), 331–340. https://doi.org/10.1093/ajcn/77.2.331[79] Cara Tannenbaum, Robert P. Ellis, Friederike Eyssel, James Zou, and LondaSchiebinger. 2019. Sex and Gender Analysis Improves Science and Engineering.
Nature
European Journal of Clinical Nutrition
73, 2 (Feb. 2019), 194–199. https://doi.org/10.1038/s41430-018-0335-3[81] D.E. Wittkower. 2016. Principles of Anti-Discriminatory Design. In .IEEE, Vancouver, BC, Canada, 1–7. https://doi.org/10.1109/ETHICS.2016.7560055[82] Gary Isaac Wolf and Martijn De Groot. 2020. A Conceptual Framework forPersonal Science.
Frontiers in Computer Science