Introduction
1 in 10 women of reproductive age in England is affected by one of the most common gynaecological disorders, endometriosis (Endometriosis UK, 2026). Despite such prevalence, the average time to diagnosis has now reached 9 years and 4 months, with 39% of women requiring ten or more GP visits before the condition is even suspected (Endometriosis UK, 2026).
Acknowledging this as a system-level failure, the NHS updated NICE guidance NG73 (2024) to promote earlier ultrasound for women with suspected endometriosis and referral. However, NICE also recognised that such changes are unlikely to address the growing failure of symptom recognition in primary care, which remains the primary driver of diagnostic delay, and risk placing further pressure on already overstretched specialist waiting lists.
Currently 94% of GP practices utilise digital triage tools (NHS England, 2025), but no targeted Al application yet exists for the early recognition of endometriosis symptoms. Although the government has increased funding for Al deployment platforms, the majority of tools do not progress beyond pilot stages (Department of Health and Social Care, 2025), suggesting institutional and political barriers to implementation. The cost of inaction falls hardest on women who wait nearly a decade for answers. This article argues that targeted Al-enabled clinical decision support, embedded within existing primary care workflows, is a necessary and achievable response to the current diagnostic crisis.
Diagnostic Delay: A Multi-level System Failure
The causes of persistent delay in endometriosis diagnosis have been historically presented as a multi-level failure across patient, clinician, and system domains (Fryer et al., 2025).
At the patient level, medical and psychosocial factors contribute, which include low public awareness and the normalisation of pain, particularly among adolescent women. This results in early warning symptoms being dismissed and delays women from seeking medical help (Ghai et al., 2020).
At the clinician level, factors including limited awareness of endometriosis presentation, frequent rates of misdiagnosis with irritable bowel syndrome or pelvic inflammatory disease, and uncertainty in referral decisions further prolong the diagnostic pathway (Ghai et al., 2020).
At the healthcare system level within the NHS, short 10-minute GP consultation times significantly restrict comprehensive history-taking and the detection of non-specific symptom patterns. This is exacerbated by gaps in mandatory gynaecology exposure in GP training, which in practice result in diagnostic uncertainty in recognising endometriosis.
Not all women face the same barriers. For women from the Global Majority, the path to diagnosis is even harder. They are more likely to be misdiagnosed with fibroids in the UK due to ethnicity-driven assumptions (APPG, 2020). The data confirms the scale of this disparity. Black women were approximately 50% less likely to receive a diagnosis than White women (Bougie et al., 2019). Similar patterns have been identified among adolescents, even after controlling for socioeconomic status. This points to persistent medical stereotypes about pain perception and I believe AI, trained on diverse data, has the potential to remove this systemic bias.
NHS primary care as the setting
NHS primary care is structured around a decentralised model, meaning national policy ambitions do not always reach individual GP practices on the ground (Tony Blair Institute, 2025).
Despite well-established digital infrastructure, existing platforms operate as rule-based systems rather than machine learning tools capable of identifying patterns across complex symptom data. This creates the illusion of innovation and obscures the real gap in upstream symptom recognition of endometriosis.
Both the NHS 10-Year Health Plan (DHSC, 2025) and the Renewed Women’s Health Strategy for England (DHSC, 2026) create a compelling call for action – the former placing Al as one of five strategic priorities for the NHS, and the latter explicitly committing to streamlining gynaecological care and faster access for conditions including endometriosis.
Stakeholder mapping
Patients require earlier diagnosis, but without investment in specialist capacity, accelerated referral pathways may further overstretch gynaecology services and generate demand the system cannot absorb.
For NHS England and Integrated Care Boards, late stage endometriosis represents a significant downstream cost. Yet this creates a structural paradox: the system bears the cost of its inaction but lacks the procurement agility to adopt the innovation that could prevent it.
Despite patient safety being the primary priority, 60% of GPs report that 10-minute allocated consultation time is insufficient to properly assess patients. UK GPs have also experienced greater workload increases than counterparts in other high-income countries (Beech et al., 2023).
Critically, GPs have actively called for a guideline-linked diagnostic decision support tool starting from first symptom presentation (de Kok et al., 2024). What appears clear is that the GP workforce is not a barrier to innovation as it could be mistakenly assumed, but a potential driver of it.
While civil society organisations, particularly the APPG on Endometriosis, have been the primary driver of political pressure, their 2020 recommendations were not met and average diagnosis times have continued to rise.
Consequences of inaction
The consequences of continued inaction span clinical, economic, systemic and equity levels.
Clinically, endometriosis contributes to infertility in up to 30% of affected women. At the same time, in the UK 83% of patients reported symptom dismissal by healthcare practitioners prior to diagnosis, with 55% attending A&E, half of whom were sent home without treatment – representing a structurally enabled clinical harm (Endometriosis UK, 2026).
On an equity level, women from ethnically diverse communities wait an average of 11 years for diagnosis, nearly two years longer than the national average (Endometriosis UK, 2026). Such governance failure has allowed diagnosis times to worsen across three consecutive measurement points (Endometriosis UK, 2026).
Economically, menstrual health conditions including endometriosis cost the UK economy an estimated £11 billion annually through absenteeism and lost productivity (Endometriosis UK, 2026). And I see this as a direct call for action to adopt AI-enabled recognition tools that could shorten the diagnostic journey and prevent years of avoidable harm.
Describing and evaluating the innovation
A machine learning algorithm trained on patient-reported symptoms achieved a sensitivity of 91-95% in external validation, representing real potential as a screening tool for endometriosis in primary care (Bendifallah et al., 2022). It is worth noting, however, that this study was conducted in France rather than within the NHS.
If such a tool were built into existing primary care systems, GPs could receive risk-based prompts at the point of consultation – not diagnoses, but flags indicating whether a patient’s symptoms align with the NICE NG73 (endometriosis) referral criteria. The advantage of this approach is that it works within existing workflows rather than creating extra clinical workload.
In the NHS context, this means embedding AI tools directly into EMIS and SystmOne, which are the electronic health record platforms used by 98% of GP practices in England. Such embedding is already possible within the NHS infrastructure. The most practical example is Klinik, a CE-marked AI triage tool currently integrated within both platforms for clinical triage purposes.
That said, the evidence has real limits. No study has yet shown these tools outperform human clinical decision-making, most datasets are small and retrospective, and none have been tested prospectively in diverse populations. Rushing to deployment without addressing this risks repeating a familiar pattern of promising innovation that never reaches practice.
Multi-stakeholder impact of the innovation
A recent survey found that 79% of women reported improvement in quality of care after receiving a diagnosis, with 98% reporting psychological harm due to symptom normalisation prior to it (Endometriosis UK, 2026). Early Al-enabled recognition could slow disease progression before it affects organs, with particular benefit for the 65% of women whose symptoms began at age 17 or under and who are currently least likely to be believed. However, risk-stratification tools carry the risk of false positives, which may generate anxiety in patients who are subsequently not diagnosed.
For GPs, Al-enabled triage offers workflow gains most likely when embedded within existing systems rather than when added as separate tools (Alamoudi et al., 2026).
For NHS England and ICBs, Al offers potential to reduce downstream surgical costs, though algorithmic bias risks mean models trained on majority populations may systematically deprioritise women from minority ethnic or lower socioeconomic backgrounds (Alamoudi et al., 2026).
Barriers to implementation
Two significant barriers stand between the evidence and implementation.
The first is clinical inertia, where decision support tools have historically been developed with poor workflow alignment, generating alert fatigue that causes GPs to dismiss them regardless of their clinical value (Beasley et al., 2021). Co-designing these tools with frontline clinicians is a necessary condition for adoption (de Kok et al., 2024).
The second is algorithmic bias: if training data underrepresents women from ethnic minority or socioeconomically deprived backgrounds, the tool risks perpetuating the very diagnostic inequalities it is designed to address (Alamoudi et al., 2026). We should ensure that dataset diversity is a prerequisite for equitable deployment.
Conditions for successful embedding
Addressing these barriers requires simultaneous action across three conditions.
First, Al-enabled tools should be piloted at Primary Care Network level with pre-specified evaluation before scaling (Tony Blair Institute, 2025).
Second, GP training on endometriosis-specific decision support must precede deployment (RCGP and Nuffield Trust, 2025).
Finally, all Al diagnostic tools must be monitored for performance and adverse events following deployment (MHRA, 2024).
A structural analysis: the Quintuple Helix
Applying the Quintuple Helix model reveals an obvious structural gap between evidence and implementation across all five helices simultaneously.
At the government helix, a strong policy mandate exists through the NHS 10-Year Health Plan and the Renewed Women’s Health Strategy, yet implementation is devolved to 42 ICBs, meaning progress varies significantly across geographic areas.
At the academic helix, the evidence base for machine learning-based symptom screening is expanding, but all existing studies were conducted outside NHS primary care populations and no real-world validation has yet been achieved
At the civil society helix, organisations such as Endometriosis UK and the APPG have been the primary drivers of political pressure, however their 2020 recommendations were not met and diagnosis times have continued to rise.
At the socio-ecological helix, the inequalities already documented in endometriosis diagnosis risk being magnified if Al tools are trained on non-representative datasets. This makes equity simultaneously the greatest challenge and the real test of whether the innovation has succeeded.
Conclusion
The increasing average time to receive a diagnosis of endometriosis in the UK has never been longer, with over 1.5 million women in England living with the consequences. This is not a failure of science or technology, but a failure of system design. The NHS 10-Year Health Plan and the Renewed Women’s Health Strategy provide a clear mandate for action, an emerging evidence base for machine learning screening tools exists (Bendifallah et al., 2022; Sivajohan et al., 2022), and the digital infrastructure is ready to be used. With diagnosis times worsening despite two rounds of NICE updates, what remains missing is the coordinated will to translate policy into practice. Successful embedding can be achieved only by addressing all five helices simultaneously: PCN-level piloting with diverse populations, followed by MHRA post-market surveillance, and supported by RCGP-led workforce preparation.
What remains is the will.
