What the
Machine
Does Not Know.

THE GAP

Women remain underrepresented in much cardiovascular research, and in many of the diagnostic standards built on top of it. The MELD score, used to prioritise patients for liver transplantation, relies on serum creatinine, a marker that runs naturally lower in women. The formula therefore systematically underestimates the severity of female patients' disease, reducing their likelihood of receiving a transplant under equivalent clinical conditions. The same male-default logic runs through cardiac troponin thresholds, post-MI guidelines, and much of what gets coded into the electronic health records that train modern medical AI.

HOW THE MODEL INHERITS IT

In High-STEACS, sex-specific cardiac troponin thresholds increased detection of myocardial injury by 42% in women and 6% in men. A 2024 JMIR study reproduced cardiac prediction algorithms across two open datasets; in its heart-failure experiments, false-negative rates were significantly higher for women in thirteen of sixteen tests. A 2023 JAMA Network Open vignette study also found that changing demographic characteristics could change recommendations made by clinicians and AI chatbots. The REBOOT trial (NEJM, August 2025, n=8,505) found beta blockers, the standard post-heart-attack drug for four decades, raised the risk of death, another heart attack, or heart failure by 45% in women with preserved heart function, with no such effect in men. That is a guideline built on male trial data, actively harming patients it was never tested on.

Sources: FemTechnology & Women At The Table, Invisible by Design: Women's Health as the Blind Spot in AI and Medicine (2025), citing the High-STEACS trial; Straw, Rees, and Nachev (2024); Kim et al., JAMA Network Open (2023); npj Digital Medicine review on the MELD score; REBOOT trial, New England Journal of Medicine (August 2025, n=8,505).

THE GAP

Unpaid caregiving is one of the largest economic contributions in the world, the ILO values it at roughly 9% of global GDP, and one of the least visible in formal data. None of it appears on a résumé. None of it appears in a credit history. None of it appears in a social-security earnings record. To the systems learning from these datasets, decades of labor look like absence.

HOW THE MODEL INHERITS IT

The blind spot compounds. Hiring systems can penalise résumé gaps that reflect unpaid care. The same pattern can enter care administration itself. A 2025 study used gender-swapped versions of records for 617 adult-social-care users. Gemma described men’s physical and mental health more directly and underemphasised women’s needs more often. Because care is allocated by assessed need, the authors concluded that these summaries could influence how urgently or extensively support is provided.

Sources: ILO estimates of unpaid care work; Harvard Business School, Hidden Workers; Rickman / LSE Care Policy & Evaluation Centre (2025), Evaluating Gender Bias in Large Language Models in Long-Term Care; cited in FemTechnology & Women At The Table, Invisible by Design (2025).

THE GAP

Insurance and health pricing can treat observed differences in claims as fixed biological risk. FemTechnology's 2026 research asks whether some of that difference is instead downstream of locatable, correctable failures in the clinical pathway, including diagnostic thresholds built on male cohorts, dosing rules built without sex-specific pharmacokinetics, and trial designs that systematically under-recruit female participants.

HOW THE MODEL INHERITS IT

The IQVIA Institute's 2025 review of 182 medical disorders found female trial participation below parity in 43% of trials between 2015 and 2024. Zucker and Prendergast's 2020 review of 86 FDA-approved drugs found higher pharmacokinetic values in women for 76 drugs; 96% of drugs with female-biased values were associated with more adverse reactions in women. The open question for AI-enabled underwriting is whether downstream cost created by miscalibrated evidence or care pathways will be treated as fixed biological risk instead of a correctable failure.

Sources: FemTechnology, Pricing the Pathway: An Agenda for the Actuarial Profession (2026), citing IQVIA Institute (2025), Women in Clinical Trials; Zucker and Prendergast (2020), Biology of Sex Differences; FDA Drug Safety Communication on zolpidem (2013); Test-Achats, Case C-236/09, Court of Justice of the European Union (2011).