Data for good: A perspective on understanding health inequalities in gender

The context of inequality: Defining disease burden and health inequalities

One of the century's greatest challenges will be to provide universal access to best in class health care to the increasing population across the globe. Life-changing innovations are nearly continuous –?from the most recent RNA based vaccines now being tested far beyond coronaviruses to a brand new class of now popular semaglutide-based drugs that are very prevalent in the combat against obesity. But access to these medications are inconsistent due to pricing, regulatory or cultural challenges leading to increased disease burden.

Disease Burden is defined as the total, cumulative consequences of a defined disease or a range of harmful diseases with respect to disabilities in a community. In general, a population’s disease burden is characterized by certain social, political and geographical features that are generally called social determinants of health.

Health inequalities are differences in the disease burden (often an increased difference) and associated opportunities in order to achieve optimal health experienced by certain groups in the society compared to the larger majority.?

It is perceived that they are often characterized a lot more in underserved populations compared to others. Underserved population refers to a group or subset of people that have disproportionate incidence & prevalence of disease because of certain characteristics like race and ethnicity, gender, sexual orientation/gender identity and disability status.?

Understanding gender based inequalities

A particular and persistent area of disparity is across gender lines. Women’s health equity refers to the state in which all individuals who are women and/or were assigned female at birth have a fair and just opportunity to attain their highest level of health. NIH states that despite scientific and medical advancements, women continue to be disproportionately affected by many diseases and conditions, including chronic pain, depression, autoimmune diseases (see OADR-ORWH), osteoporosis, and dementia.

A recently written Mckinsey article on addressing Women’s health gap describes the barriers women face to accessing healthcare, such as high costs and also the disproportionately in conditions. Often they are diagnosed years later, on average, when compared to men with the same medical condition and while women live longer than men; they spend 25 percent more of their lives in worse health with indications such as headaches, autoimmune disease and depression contributing to make up 47% of that disease burden.

In addition, other studies and articles have identified that? common female-specific conditions, such as endometriosis, remain largely understudied and lack effective diagnostics and treatment options.? Healthcare providers also , often estimate women to have less pain and be more likely to exaggerate it compared to men. Fewer women are prescribed medications like beta blockers or cholesterol-lowering drugs after a heart attack, regardless of severity

This makes one wonder if we could establish a causality behind the reasons for these discrepancies. Some of the most commonly cited reasons include the tendency to train medical professionals on the male body type by default and less understanding of the different symptoms experienced by women. This general lack of education in turn can impact outcomes. And while important work has been done to use data to expose these discrepancies, we still don’t completely understand exactly why there’s such a massive variance between women and men.

Whats being measured out there?

There are some standardized indexes as well that measure these statistics across a variety of other spectrums. For example the United Nations defines a gender inequality index which reflects gender-based disadvantage in three dimensions— reproductive health, empowerment and the labor market—for as many countries as data of reasonable quality allow. It shows the loss in potential human development due to inequality between female and male achievements in these dimensions and ranges from 0, where women and men fare equally, to 1, where one gender fares as poorly as possible in all measured dimensions. It does provide a comprehensive view across countries , however as we will note, a recurring theme is data quality and standardized data capture.

Data2X , an organization focusing on addressing? gender gaps with data has identified current areas in gender-disaggregated and sex-specific data collection. Their work highlights areas where inequalities remain undocumented due to lack of data. The article also provides a nice summary of the efforts and organizations devoted to address these gaps.

These sources collectively point to ongoing challenges in collecting comprehensive, high-quality data on gender inequality, particularly in areas like economic participation, childcare, and the implementation of gender equality laws.

What are the challenges to doing this at scale?

The rise of artificial intelligence and the mass adoption of cloud for powering big data analytics has seen the overall volume of data and insights derived from that data soar to a higher proportion. But that data may not be usable without:

• a proper data definition or a strategy designed to democratize data access?
• allow for enrichment and linkages with other datasets
• secured guardrails to allow for such linkages and accesses

In general, there needs to be an overall plan to ensure that insights and analytics derived from AI and other technologies are doing their part to eliminate bias and represent them in a way that is? traceable to the source in order to infer meaningful outcomes grounded in reality.

However, there are two potential blockers that stand in the way of addressing these inequalities: (i) one being issues associated with data collection and (ii) the other one being privacy.?

We can separate the data collection biases into two broad categories:?

• lack of data standards and
• insufficient data capture.

1. Data Standardization & Capture

Data standardization has been an ongoing conversation where the focus has been around ensuring captured data has key attributes across the different sources and a well defined business metadata or data dictionary that provides meaningful and descriptive information on these attributes, so that they can be re-used for analysis with quality. The concept of high quality data is often interpreted and? called as? FAIR and it originates from a paper published in Scientific Data nearly a decade ago. The absence of firm, comprehensive standards means different studies (no matter how well-meaning) may not be making equivalent measurements. And in many cases, the actual data needed may simply not be collected, which brings us to the second part of the problem which is data capture.?

An example of data capture with clinical trial data capture and drug rejections

While we emphasize the need for sound data capture and standardizing on attributes, here are examples of the impact of not doing this with respect to studying inequalities in general.

• one recent study found that a quarter of clinical trials in America did not consider sex differences, meaning evidence for certain interventions may not translate to the same benefits in women as in men.?
• It is estimated that women comprise only 29-34% of participants in industry-sponsored early-phase trials.
• As nicely summarized in this article, studying only one sex when developing new treatments misses a whole spectrum of information on how a drug works primarily because males and females have many differences in how drugs work in our body.
• It is noted that females do end up with more adverse events than males alone thus making the need for not only recruiting more females in studies but also to ensure that the sex de-aggregated data is used in studying drug efficacy and safety.
• The Mckinsey article claims, since 1980, women have a likelihood of 3.5X? of drug withdrawals compared to men.

This means that even if one were to study inequality , the absence of attributes defining the data(what we call metadata) to determine the study of it, is going to make it difficult.

2. Data Privacy

The second major concern about a data-driven approach is that patient privacy will be impacted. For example, there are particular sensitives for people either assigned female at birth or identifying as female, as there have been data breaches in the past.This is a legitimate question, but one that can be overcome with proper procedures. We need to build on data sharing safeguards to ensure research is also done in an ethical and compliant manner without compromising on patient privacy and without ability to trace them back to the actual patient even when we link across different sources (namly re-identification). Recent breaches in technologies and cyber leaks have led to general concerns in the storage and aggregation of the data but policy and guidelines in addition to technology rigor have to a large extent ensured that these concerns are addressed with increased discipline.

The Solutions: A recap of the initiatives in public sector

There are initiatives in place that target at bridging this gap. For example, in UK, the NHS has initiated measures targeting a particularly challenging part of the health care system –?primary care data. According to this article Imperial College London, University College London and Boots came together last year with funding from Cancer Research UK to examine over-the-counter medication data from loyalty cards for medicines pain and indigestion medications. The analysis was able to identify ovarian cancer up to eight months before current diagnosis processes. More than 4,000 people die in the United Kingdom every year from ovarian cancer, and when the disease is found has a huge impact on health outcomes. For e.g. just 13 percent of people with Stage Four ovarian cancer will survive, compared to 93 percent of those Stage One cancer.

In the US , ?the Department of Health and Human Services (HHS) and the Patient Centered Outcomes Research Trust Fund (OS-PCORTF) initiative have a focus on studying maternal health and monitor health outcomes that are unique to women. However, patient-centered outcomes research (PCOR) requires robust data to monitor, understand, which are obtained from different sources and hence have difficulties in aggregation across the diverse data entities. Under the OS-PCORTF, the Assistant Secretary for Planning and Evaluation (ASPE) has funded multiple projects that are helping to address these challenges with better tools to collect, standardize, link, share, and analyze women’s health data. The 2020-2029 OS-PCORTF Strategic Plan charts a course for the future and includes a priority area under Goal 1 around maternal health. In an effort for data standardization and enhancing the quality of data collection, they also have set up a Coordinated registry network (CRNs) , in which multiple registries align their data capture and sharing--as a mechanism for increasing the data and analytic tools available for women’s health research.

Age of AI

With the rising use of AI , it is but nevertheless to consider and unlock analytics that can bring value. Delivering into these topics would be a separate blog itself and the aim would be to follow up on this with technological approaches and concepts that can be leveraged to address this gap like differential privacy, tokenization, NLP for data extraction, RAG based LLM solutions. However, while we talk about AI? and generative capabilities (like chat gpt),? there are a few binding principles that cannot be compromised the main one being data quality and metadata descriptions.?

AI can help in many ways including providing better data capture with NLP from multiple source, being able to match patients across different health center visits to create longitudinal profiles, perform anonymization at scale however there is still no AI strategy without data strategy.

In short , to recap, before we consider AI, we need to ensure:

• Metadata becomes key: definitely every data collection needs to have a definition of the source of generation. Every generative AI solution talks about reduction in hallucination and providing attribution to the source. While customers can leverage RAG or Graph based RAG , the important aspect of it again comes back to capturing information regarding the quality?
• Privacy concerns shouldn’t be compromised: analytics auditability and governed access of data should be implemented.? Attempts to re-identify PHI data should be flagged and technologies exist to ensure this compliance.?

Voices from the field: a panel to discuss the skilling and analytical challenges

In March 2024, Women in Data ,in partnership with Snowflake, gathered a group of senior experts for a conversation on “understanding & tackling women health inequalities”. Besides myself, representatives included Natalie Cramp from WID , Jennifer Visser-Rogers from CRO Phastar, Janet Broome from Snowflake and? Mridula Sori and Camellia Williamson from NHS. The topics focused on understanding various perspectives from the field representing different organizations on bringing the gap. The panel addressed various questions? including?

• main barriers organizations face when adopting a data-driven approach to address alth inequalities
• causes of discrepancies between men and women in how they exhibit health syndromes
• underrepresentation of women in? clinical trials and whether the mechanism of action on women are well understood
• efforts from? NHS that have been undertaken to help mitigate this care gap and the things the NhS and NHSE are doing regarding re-training, dev and recruitment to ensure more women come into data and there is sufficient upskilling

Some of the key themes from these discussions stood out and here is a snapshot:

1. There are systematic, interconnected gaps in technology, skills and data sharing that when combined are the primary cause of adopting a data-driven approach. We see very real challenges in receiving the funding to update archaic infrastructure (which is still for some people done on paper), despite the many benefits. Best practice frameworks and targeted grants can help organizations ‘leapfrog’ development and take advantage of innovation.?
2. There have been efforts to improve the underlying infrastructure for data and technology to provide high-quality, timely data for service improvement, research and innovation
3. The NHS is working on developing its analytical workforce and better harnessing their skills to improve data use. There are initiatives to improve data validation, cleansing, and matching, particularly for demographic data. There's a push for routine development and deployment of open-source innovations developed in collaboration with end user?
4. While we don’t have the full picture, there’s growing evidence that women and men respond differently to drug therapies. We’re learning that ‘women’ are a much more heterogeneous population than men with unique biological life events.? For e.g: Forty-nine percent of people with cardiovascular disease are women, yet only 41.9 percent of participants in cardiovascular research are female. Although 51 percent of cancer patients are female, only 41 percent of cancer trial patients are female. If we’re not studying conditions in women properly, how can we expect to understand the impact that they will have on women’s experiences, including in the workplace.
5. We need more women taking part in clinical trials. Many women have major time pressures – household responsibilities along with jobs mean repeated visits to clinics can be impossible. Digital-based trials can increase accessibility and are more patient-centric. They can play a crucial role in encouraging women to participate.
6. Digital health tools offer the opportunity to tailor interventions according to the unique health care needs of women with real-time data. External control arms make it easier to ensure balanced participation in trials. Advances in personalized medicine are likely to have a bigger impact on women due to the heterogeneity of the population.
7. However even in larger organizations some of the barriers include Data sharing & Privacy concerns , lack of funding for Innovation adoption? and increased time in upskilling workforce at scale to keep up with newer data analysis technologies takes time and effort

Mridula Sori , summed it up well, when she said

The data is everything. Data quality is paramount. Data is vast and complex. You need the right data." and the folks at NHS are doing our best to get the best data and get the flows right".?

Conclusion and Call to Action

What we want to do is use the momentum from pilots and experiments across the world to inspire more comprehensive change. Larger, even more impactful initiatives are ready. The sheer amount of health inequality is a double-edged sword –?representing both a major failing and an incredible opportunity. We now have the tools to make significant progress on this longstanding woe and it’s something we’re confident we can do, together.? There are many initiatives internally and across government to address such care gap and look at inequalities representing data. The central theme to all this will be collaboration, upskilling and technology innovations to bring economies of scale faster. We summarize this blog by highlighting the following themes:

1. Data disaggregation is crucial: Only about 50% of interventions report sex-disaggregated data. Collecting and analyzing data by sex, ethnicity, and gender is essential to accurately represent women's health burdens and the impact of different interventions.
2. Women's diversity must be recognized: Women experience more biological variations across their lifecycle than men, leading to different reactions to interventions and medications. More robust testing and data collection across the entire spectrum of women's lives is needed.
3. Clinical trial participation remains inadequate:Despite women being more affected by certain conditions, clinical trials are still disproportionately conducted on men or male animal models. Creative solutions, like remote trials, are emerging to increase accessibility for women, though they present new challenges in data verification.
4. Data quality and trust are significant barriers:Improving data quality through extensive cleaning and connecting of existing health records is crucial. Building trust among citizens for data sharing is essential, requiring clear communication and demonstrating rapid value from the data collected.

In the next part of the series, we will dive into technologies that can be leveraged to solve the data gaps for both AI and privacy aspects

For more active contributions on this topic, please visit the data for good initiative forum for detailed discussions

References

https://www.ncbi.nlm.nih.gov/books/NBK425844/

https://orwh.od.nih.gov/womens-health-equity-inclusion

https://mcpress.mayoclinic.org/women-health/working-toward-gender-equity-in-womens-health-care/

https://www.sciencedirect.com/science/article/pii/S2772501423000076

https://ourworldindata.org/grapher/gender-inequality-index-from-the-human-development-report

https://www.go-fair.org/fair-principles/

https://www.clinicaltrialsarena.com/features/underrepresentation-women-early-stage-clinical-trials/?cf-view

https://pubmed.ncbi.nlm.nih.gov/34649925/

https://www.dataiq.global/award-winner/dataiq-awards-2023-winner-best-use-of-data-for-not-for-profit-or-non-commercial-purposes-imperial-college-london-boots-and-tesco/#:~:text=In%20a%20world%2Dfirst%2C%20Imperial,months%20sooner%20than%20current%20processes.

https://aspe.hhs.gov/collaborations-committees-advisory-groups/os-pcortf

https://link.springer.com/referenceworkentry/10.1007/978-1-4020-5614-7_297#:~:text=Definition,aspects%2C%20and%20costs%20to%20society.