The UCSF Health Problem: Endometriosis is a chronic, debilitating condition affecting approximately 10% of reproductive-age women worldwide. It is characterized by the growth of endometrial-like tissue outside the uterus, leading to severe pelvic pain, dysmenorrhea, dyspareunia, bowel and bladder dysfunction, and infertility. Despite its prevalence, endometriosis remains significantly underdiagnosed, resulting in profound patient suffering and substantial economic burden.

The diagnostic journey for endometriosis is protracted, with patients often experiencing a delay of 8-10 years from symptom onset to diagnosis. This delay is compounded by the fact that patients consult an average of 10 physicians before receiving a correct diagnosis. This protracted process leads to years of untreated pain and suffering, impacting patients' quality of life, mental health, and productivity.

The economic impact of endometriosis is substantial. Studies have estimated the annual cost of endometriosis-related healthcare and lost productivity to be in the range of $10,000 to $20,000 per patient. This includes direct medical costs (e.g., physician visits, imaging, surgery, medications) and indirect costs (e.g., lost wages, reduced work productivity, absenteeism). A study published in the Journal of Human Reproductive Update estimated the annual cost of endometriosis in the US alone to be over $69 billion.

Given the significant burden of endometriosis, there is an urgent need for improved diagnostic and management strategies. The current reliance on subjective symptom assessment and non-standardized history taking contributes to diagnostic delays and inconsistencies in care. An AI-driven tool that can standardize patient history collection, assess symptom severity, and predict treatment response has the potential to: reduce diagnostic delays and improve patient access to appropriate care, enhance clinician understanding of endometriosis and improve patient-clinician communication, optimize treatment planning and improve patient outcomes, and reduce the economic burden of endometriosis by minimizing unnecessary healthcare utilization and lost productivity.

How Might AI Help? AI can help by developing a dynamic, interactive questionnaire integrated into the patient's history within APeX EHR. This questionnaire would cover key symptom domains (pelvic pain, dysmenorrhea, dyspareunia, bowel symptoms, etc.), patient medical history, family history, and lifestyle factors. Machine learning algorithms would analyze patient responses to generate: 1. symptom severity score, quantifying the patient's endometriosis burden, 2. risk stratification for specific endometriosis subtypes (e.g., deep infiltrating endometriosis, ovarian endometrioma). 3. personalized recommendations for diagnostic workup, and predicted likelihood of symptom improvement with various treatment options.

How Would an End-User Find and Use It? Within the patient's history in APeX EHR, a "Endometriosis Symptom and Risk Assessment" button would be available. Clicking this button would initiate the interactive questionnaire. Patients could complete the questionnaire in the clinic or remotely via the patient portal. Once completed, the AI would generate a report summarizing the symptom severity score, risk stratification, and personalized recommendations. The report would be displayed within the patient's chart, with visual aids to enhance understanding. Clinicians could use the report to: guide discussions with patients about diagnostic and treatment options, tailor treatment plans based on predicted likelihood of symptom improvement, pend orders for recommended labs and imaging directly from the report interface.

Embed a picture of what the AI tool might look like:

"Endometriosis Symptom and Risk Assessment"

Patient: [Patient Name]

Date: [Date]

Symptom Severity Score: 18 (Moderate-Severe)

Risk Stratification:

- Deep Infiltrating Endometriosis: 60% probability

- Ovarian Endometrioma: 30% probability

Recommended Diagnostic Workup:

- Pelvic MRI with endometriosis protocol

- Pelvic ultrasound

Predicted Treatment Response:

- Hormonal Therapy (GnRH agonist): 70% likelihood of symptom improvement

- Laparoscopic Surgery: 85% likelihood of symptom improvement

[Symptom Severity Graph: Showing patient's pain scores over time]

[Risk Probability Chart: Visualizing risk of different endometriosis subtypes]

[Buttons: “Order MRI,” “Order Labs,” “Discuss Treatment Options”]

What are the Risks of AI Errors? 1. False Negatives: The AI might underestimate symptom severity or miss subtle indicators of endometriosis, leading to delayed diagnosis.False Positives: The AI might overemphasize certain symptoms, leading to unnecessary investigations or treatments, 2. Algorithmic Bias: The AI might exhibit biases based on the training data, potentially impacting care for minority populations.3. Data Misinterpretation: Clinicians might misinterpret AI-generated risk scores or recommendations.

Mitigation strategies include rigorous validation of the AI algorithm on diverse patient populations, clear communication of the AI's limitations and the importance of clinical judgment, continuous monitoring of AI performance and user feedback.

How Will We Measure Success?

This project's success will be evaluated based on the following framework, addressing real-world uptake, meaningful impact, safety, and equity/fairness:

1. Real-World Uptake (Process Metrics): Measurements using data already collected in APeX: include frequency of tool utilization by clinicians (number of assessments completed), time taken to complete the assessment questionnaire, integration of AI-generated recommendations into patient care plans (e.g., orders for recommended labs/imaging), and changes in referral patterns to the Comprehensive Endometriosis Center. Other measurements ideally needed: cinician satisfaction surveys regarding tool usability and integration into workflow, patient feedback on the clarity and helpfulness of the AI-generated report.

2. Meaningful Impact (Health Outcome Metrics): Measurements using data already collected in APeX, such as reduction in the time from symptom onset to diagnosis, changes in the utilization of diagnostic procedures (e.g., number of MRIs, laparoscopies), changes in the utilization of treatment modalities (hormonal therapy, surgery), patient-reported outcome measures (PROMs) for pain, quality of life, and symptom severity, reduction in complications related to endometriosis. Other measurements ideally needed include longitudinal data on symptom control and disease progression, cost-effectiveness analysis of the AI-driven approach, and patient reported measures of improvement in specific symptoms.

3. Safety and Equity/Fairness: Measurements using data already collected in APeX: analysis of potential disparities in tool utilization and outcomes across different patient demographics (e.g., race, ethnicity, socioeconomic status), tracking of adverse events related to diagnostic or treatment decisions influenced by the AI tool. Other measurements ideally needed: validation of the AI algorithm's performance on diverse patient populations to ensure equitable outcomes patient feedback on perceived fairness and trust in the AI-driven assessment.

Evidence for Continued Support/Abandonment: Continued Support: Significant reduction in time to diagnosis, demonstrable improvement in patient-reported outcomes, and high clinician/patient satisfaction would warrant continued support. Abandonment: Evidence of significant algorithmic bias, lack of clinical uptake, or adverse patient outcomes would necessitate re-evaluation or abandonment.

Describe your qualifications and commitment: This project addresses a high-priority area within UCSF Health, specifically the significant delays and disparities in endometriosis diagnosis and management. The AI-driven tool has the potential for large-scale impact by improving patient outcomes and streamlining clinical workflows.

Jeannette Lager, As the Section Chief for Minimally Invasive Gynecologic Surgery (MIGS) at UCSF, and formerly the Medical Director and Interim Chief of the Urogynecology and MIGS division, I have clinical and leadership expertise directly relevant to endometriosis care. My role as Associate Director of the Comprehensive Endometriosis Center provides me with deep insight into the patient population and the challenges they face. I have a strong understanding of the clinical workflows and can design the AI tool to integrate seamlessly into existing practice.

My co-investigator, Zaineh Khalil, is a highly experienced Nurse Practitioner with a specialized focus on endometriosis care. Her extensive clinical experience, coupled with her deep understanding of patient needs, makes her an invaluable asset to this project. She has been actively involved in quality improvement initiatives within the MIGS and Urogynecology clinic. Her expertise in navigating the complexities of endometriosis management, combined with her understanding of clinical workflow, will be instrumental in ensuring the tool's practical application and successful integration into the clinical setting.

Furthermore, I have a strong track record of collaboration with UCSF administration and faculty, ensuring engagement and support from critical operational and clinical champions. Together with my co-investigator, we are positioned to effectively lead this project and ensure its successful implementation.