Informatics for sex- and gender-related health: understanding the problems, developing new methods, and designing new solutions

INTRODUCTION

Investment in medical research continues to grow; however, the sex and gender gap of health outcomes persists^1,2 with poor support for the health of cisgender and transgender women,^3–5 intersex people,⁶ and all gender-diverse people (ie, people whose gender identity and sex assigned at birth do not fully align).

Informatics approaches have the potential to identify, address, and mitigate these disparities. For example, methods that elucidate the impact of sex as a biological variable, clinical decision support (CDS) systems, and personal health informatics tools are needed to specifically fit the needs of women, intersex people, and all gender-diverse people. The goal of this issue is to highlight such informatics research.

Included in this issue are 19 outstanding articles that focus on 4 key areas including gender disparities, gender diversity, maternal health, and sex differences (Table  1). The majority of these articles were focused in the clinical informatics domain with 13 clinical informatics papers. The remaining 6 papers were from the fields of consumer health informatics (N = 4) and translational informatics (N = 2). We also organized the included articles by sex- and gender-related health themes focusing on 4 areas: gender disparities, gender diversity, maternal health, and sex differences. Interestingly, the majority of the consumer health informatics papers (3 out of the 4 included in this issue) were studying gender disparities. No consumer health informatics papers focused on maternal health, and therefore, this could be an area that warrants further investigation.

Table 1.

Categorization of Published Articles Included in This Issue by Level of Sex- and Gender-Related Health Theme and Informatics Domain

First author last name	Title	Gender- and sex- related health theme	Informatics domain	Ref
Gender diversity (N = 8)
Ram	Transphobia, encoded: an examination of trans-specific terminology in SNOMED CT and ICD-10-CM	Gender diversity	Clinical informatics	17
Wagner	Transgender and nonbinary individuals and ICT-driven information practices in response to transexclusionary healthcare systems: a qualitative study	Gender diversity	Clinical informatics	22
Kidd	Operationalizing and analyzing 2-Step gender identity questions: methodological and ethical considerations	Gender diversity	Clinical informatics	21
Kronk	Transgender data collection in the electronic health record: current concepts and issues	Gender diversity	Clinical informatics	19
McClure	Gender harmony: improved standards to support affirmative care of gender-marginalized people through inclusive gender and sex representation	Gender diversity	Clinical informatics	18
Marney	Overcoming technical and cultural challenges to delivering equitable care for LGBTQ+ populations in a rural, underserved area	Gender diversity	Clinical informatics	25
Antonio	Toward an inclusive digital health system for sexual and gender minority people in Canada	Gender diversity	Clinical informatics	24
Pho	Online health information seeking, health literacy, and human papillomavirus vaccination among transgender and gender diverse people	Gender diversity	Consumer health informatics	23
Maternal health (N = 5)
Zheutlin	Improving postpartum hemorrhage risk prediction using longitudinal electronic medical records	Maternal health	Clinical informatics	27
Rattsev	Recurrent preterm birth risk assessment for two delivery subtypes: a multivariable analysis	Maternal health	Clinical informatics	33
Singh	Outpatient portal use in prenatal care: differential use by race, risk, and area social determinants of health	Maternal health	Clinical informatics	29
Zheutlin	A comprehensive digital phenotype for postpartum hemorrhage	Maternal health	Clinical informatics	26
Meeker	Neighborhood deprivation increases the risk of post-induction cesarean delivery	Maternal health	Clinical informatics	30
Gender disparity (N = 4)
Lee	Towards gender equity in artificial intelligence and machine learning applications in dermatology	Gender disparity	Clinical informatics	10
Ryu	Microaggression clues from social media: revealing and counteracting the suppression of women’s healthcare	Gender disparity	Consumer health informatics	7
Josephy	A model for building a national, patient-driven database to track contraceptive use in women with rare diseases	Gender disparity	Consumer health informatics	9
Pichon	The messiness of the menstruator: assessing personas and functionalities of menstrual tracking apps	Gender disparity, gender diversity	Consumer health informatics	8
Sex differences (N = 2)
Zhou	Gender-specific clinical risk scores incorporating blood pressure variability for predicting incident dementia	Sex differences	Clinical informatics	34
Gutiérrez-Sacristán	Multi-PheWAS intersection approach to identify sex differences across comorbidities in 59 140 autism spectrum disorder pediatric patients	Sex differences	Translational informatics	35

GENDER DISPARITIES

Four articles address questions pertaining to gender disparities and describe novel informatics approaches to tackle the challenging questions raised. Three of the 4 articles fall under the category of consumer health informatics. Only 1 article is a full research and applications paper and the others are review, brief communication, and perspective articles, indicating that this area is still in active and ongoing development.

Ryu and Pratt’s research article explores how social media posts can be used to uncover microaggressions towards women in Korea receiving gynecological care.⁷ While clinical providers often are responsible for mid-visit microaggressions, their analysis identified other members of the woman’s support structure (eg, mothers, male partners, and superficially supportive social media responders) that contributed to both pre- and post-visit microaggressions.⁷ Understanding all these types of microaggressions is important for women’s health because pre-, post-, and mid-visit microaggressions can all influence whether a woman seeks medical care. In informatics, especially clinical informatics, the focus is often placed on mid-visit microaggressions and other aspects of the clinical visit while it is taking place. However, this pre- and post-period when a woman is undertaking her daily life is often so vital in actually changing and modifying health behaviors in meaningful ways. Social media can be used along with novel informatics methods to identify the problems and address some of these gaps in informatics.

Switching from social media to health apps, Pichon et al⁸ reviewed the literature on menstrual tracking apps to compare and contrast the literature on menstruators themselves and menstrual tracking apps to assess whether these technologies were responsive to users and their needs. The main gap identified by the extensive review was the need for a human-centered artificial intelligence approach for model and data provenance, transparency and explanations of uncertainties, and the prioritization of privacy in menstrual trackers.⁸ Unfortunately, privacy issues among all apps transcend health. However, when menstruators start to log their intimate care and sexual activity in these apps, the privacy issues balloon to astronomical proportions.

Josephy et al⁹ present a model for building a national linked database to track contraceptive usage among women with rare diseases. This brief communication describes a pilot database focusing on 1 rare disease—cystic fibrosis.⁹ While males with cystic fibrosis are mostly infertile (>98%), females with cystic fibrosis have reduced fertility but are not infertile (and many often go on to have healthy children later in life). Therefore, cystic fibrosis represents an interesting rare disease to study this novel national linked database of contraceptive usage. The authors demonstrate feasibility by linking contraceptive survey variables with clinical outcome variables in a small set of 150 patients. A major and important novelty of this pilot is the patient-focused approach to understanding medication usage (eg, contraceptive usage) among women with rare diseases, and the underlying understanding that their needs may differ from other populations.

Lee et al¹⁰ contribute a perspective on ways that machine learning and artificial intelligence (ML/AI) can be used to study skin cancers, and other dermatological conditions, in different genders. They present interesting arguments for both the reasons why one might want to include sex- or gender-based diagnostic criteria for diseases with gender-based differences (desirable bias) and also undesirable biases from usages of datasets with underrepresentation of certain groups. Skin is a part of one’s body that often defines individuals in so many different ways. Skin is a place where folks express themselves in terms of piercings, tattoos, and other body modifications. Furthermore, makeup is commonly used by different genders as another form of sexual and gender expression. Lee et al¹⁰ provide thoughtful and actionable recommendations for ensuring sex and gender equity in the development of ML/AI for dermatology.

GENDER DIVERSITY

As the 8 gender-diversity papers in this issue point out, transgender, nonbinary, and other gender-diverse people face extensive health disparities,^11–14 and many have been actively harmed by healthcare systems.^15,16 In Ram et al,¹⁷ the first 3 authors describe their own personal experiences of harm to introduce their analysis of the role that informatics can play. In particular, they detail how terminology systems like SNOMED CT and ICD-10-CM—which are so commonly used in electronic health record (EHR) systems—use derogatory and outdated language that harms transgender people, excludes nonbinary people, and can inhibit their access to care. As one approach to improve terminology systems, McClure et al¹⁸ describe the Health Level Seven International (HL7) Gender Harmony logical model that promotes the representation of gender identity, recorded sex or recorded gender, sex for clinical use, the name to use, and affirmative pronouns.

In one of our Editor’s Choice papers on transgender data collection, Kronk et al¹⁹ also describe how the design of our EHR systems overall fuel these harms and provides concrete suggestions for improving systems. They provide explicit examples of a more respectful intake form for people to select their pronouns and detail changes in terminology over time, describing why historical terms can be harmful towards transgender people now. One of the paper’s suggested solutions to these problems is to incorporate a 2-step gender question for intake forms, which includes separate questions for sex assigned at birth and gender identity and allows for inference of various gender-diverse categories with increased community acceptability.²⁰ Another paper by Kidd et al²¹ in this issue describes the use of such a 2-step gender question in clinical research to illustrate that how the data are represented impacts both study results and interpretation of those results. They point out the lack of standards on what gender information is collected and how it should be used in research. Their study shows that combining all the gender-diverse youth into 1 category—as opposed to grouping based on sex assigned at birth or nonbinary vs binary gender identities—influences the concluding odds ratio of suicide ideation and obscures important differences among gender-diverse categories. Thus, choices we make in how to represent gender in clinical research influence our ability to understand challenges before we can ultimately address the health disparities faced by gender-diverse people.

Several papers identified specific information practices by gender-diverse people. Based on extensive interviews and focus groups in North Carolina, Wagner et al²² describe the health information barriers that transgender and gender nonbinary people face and detail how they use information and communication technologies through a variety of information practices to help them overcome the barriers and biases in healthcare systems. Pho et al’s²³ paper dives deeply into the information seeking practices of gender-diverse people regarding human papillomavirus (HPV) vaccination and discovered that people were much more likely report HPV vaccination if they visited a social networking site like Facebook, potentially due to peer validation.

Our final 2 papers provide detailed case reports. Antonio et al²⁴ describe an effort to modernize the capture of gender, sex, and sexual orientation data in Canadian digital health systems. They present their co-created plan with 7 main actions. In our other Editor’s Choice paper, Marney et al²⁵ nicely round out the selection of gender-diversity papers by describing a detailed case for how an integrated health system in a rural and largely conservative area was able to make changes to their EHR system as well as social changes to better support the needs of lesbian, gay, bisexual, transgender, queer, or questioning (LGBTQ+) community.

Finally, we acknowledge that none of the authors on this editorial identify as gender-diverse, and we do not have the lived experience that many of the gender-diverse authors of these papers have. We have tried to use the vocabulary expressed in these gender diversity papers throughout our editorial, and we have encouraged authors of all papers in this issue to carefully consider their usage of sex- and gender-related terminology. Nonetheless, we also note that the vocabulary that best supports gender-diverse people is evolving and subjective.

MATERNAL HEALTH

Another important aspect of sex- and gender-related health is maternal health. Two studies investigated postpartum hemorrhage using EHR data. The first developed a phenotype algorithm to identify this important and difficult to capture phenotype from EHR data.²⁶ The follow-up study described a hemorrhage risk prediction algorithm²⁷ that was based on the developed phenotyping algorithm. From a machine learning perspective, the prediction algorithm’s performance may appear somewhat modest; however, their algorithm outperformed other postpartum hemorrhage prediction tools.²⁷ This demonstrates the importance of machine learning and other prediction algorithms on this important clinical task. Importantly, implementation of these novel prediction tools into CDS represents a separate and very challenging step towards improving postpartum hemorrhage rates in the future. However, this additional step would require that no EHR-specific or algorithm biases had been introduced into the risk prediction algorithms. Biases, including race and gender-specific biases, are common in machine learning algorithms²⁸ and other risk prediction algorithms and careful planning must be taken before implementation of any prediction tool into the clinical pipeline via CDS. Unfortunately, none of the articles included in this issue described novel CDS tools applied to the maternal health space. Therefore, this could represent an area in need of future research.

Two studies by different study teams investigated ways that race and social determinants of health can alter maternal health outcomes. One study investigated the relationship between outpatient portal usage during the prenatal care period and various clinical and social factors.²⁹ The authors found that non-Hispanic Black and African American women or patients who also lived in lower socioeconomic neighborhoods used the outpatient portals less than their counterparts.⁶ The authors also found that those with high-risk pregnancies were more likely to use the outpatient portals than those with normal risk pregnancies.²⁹ Importantly, these findings demonstrate that disparities in terms of prenatal care are occurring in different communities based in part by access and interactiveness of tools, such as outpatient portals, among those communities. Another study found that neighborhood deprivation (a composite variable including many different indicators of social determinants of health) was significantly associated with increased risk of postinduction cesarean delivery (an adverse outcome following labor induction).³⁰ The authors found that after adjustment for confounders, those living in the most deprived neighborhoods were at 29% increased risk of postinduction cesarean delivery (aOR = 1.29, 95% CI 1.05–1.57) compared to the least deprived.³⁰ This demonstrates the importance of understanding the impact of disadvantaged neighborhoods on health outcomes,^31,32 including adverse delivery outcomes.³⁰ Coupled with the work by Singh et al²⁹ on outpatient portal usage, it demonstrates that certain neighborhoods with lower socioeconomic status and/or higher deprivation levels access outpatient prenatal care portals differently and this could be another factor that is affecting the maternal health outcomes observed in Meeker et al.³⁰ More work is needed to understand the important and complex interaction of race, ethnicity, gender, and social determinants of health on maternal health outcomes and access to care.

Rattsev et al³³ developed a 4-step analytical framework based on a clinical phenotyping tool to assess the risk for recurrent preterm birth (ie, preterm birth among a patient with a prior history of preterm birth). They found that models stratified by delivery subtype performed better than a naive model (concordance 0.76 for the spontaneous model, 0.87 for the indicated model, and 0.72 for the naive model).³³ These practical features may be useful in future analyses aimed at assessing risk of recurrent preterm birth using EHR populations.

SEX DIFFERENCES

Two articles in issue focus on sex differences without exploring the role of gender on the clinical outcome of interest. Zhou et al³⁴ explored the role of sex as a biological variable on clinical risk scores using blood pressure for predicting dementia. While Zhou et al used the phrase “gender-specific clinical risk scores,” these clinical risk scores were focused on male and female patients as identified by the EHR and therefore likely more accurately represent sex as a biological variable and its ability to impact blood pressure variability. These blood pressure variability scores then affected the risk of dementia development in the 2 populations (male and female).³⁴ Another study investigated the effect of sex differences across comorbidities among autism spectrum disorder (ASD) pediatric patients.³⁵ ASD is a disease that typically affects males more frequently than females. This work was interesting in that they identified 27 different comorbidities that were more frequently co-occurring in females with ASD (vs males).³⁵ Many were neurological and mental disorders while some were less intuitive and included endocrine metabolic diseases, digestive disorders, and diseases of the sense organs (eg, strabismus).³⁵

FUTURE DIRECTIONS/VISION

The papers in this issue point out the critical need for further informatics research to understand and better support the special health needs of cisgender and transgender women, intersex people, and all gender-diverse people. Most informatics research still makes inaccurate and simplified assumptions that conflate sex and gender and force people into binary categories of male and female—ignoring the increasing prevalence of people who are intersex³⁶ and rarely explicitly accounting for transgender, nonbinary, or other gender-diverse people. Even within this issue, the 2 papers that focused on sex differences excluded intersex people and failed to account for gender identities. This is due to the underlying datasets used for analysis, which conflate sex and gender and lack representation for intersex and gender identify, and thus impede researchers’ ability to accurately represent sex and gender in their studies. Thus, as the gender-diversity papers illustrate, we need to modify our informatics systems at fundamental levels so that they represent all people and that all people’s health can be better understood and supported.

AUTHOR CONTRIBUTIONS

All authors split the editorial duties for papers submitted to this issue. NE and WP developed the initial set of 4 topical categories and assigned each paper to them. WP drafted the gender diversity section, and MRB drafted the gender disparities, maternal health, and sex differences sections. MRB and WP edited the entire manuscript. NE, WP, MRB approved the final version of the manuscript.

CONFLICT OF INTEREST STATEMENT

None declared.