The promise of artificial intelligence in medicine is tantalizing: algorithms that can predict disease, personalize treatments, and democratize expertise. Yet, for all its potential, AI has a fundamental vulnerability—it is only as good as the data it learns from. In many areas of healthcare, particularly women’s health, progress has been stalled not by a lack of algorithms, but by a profound lack of clean, structured, and longitudinal data. This is the “gender data gap” in action, a systemic blind spot that leaves AI with little to learn from.
Before machine learning models can revolutionize care, a foundational layer must be built. A company must first architect a scalable system to collect, structure, and validate high-quality clinical data. This is not a story about AI, but the story of what comes before AI: the construction of the data engine. A fascinating case study in building this engine is Winona, a digital health platform for menopause care. By examining their tech-enabled, human-driven model, we can see a blueprint for how to solve the data scarcity problem and, in doing so, lay the groundwork for a future of truly intelligent, personalized medicine.
The Systemic Challenge: A Data Desert in Women’s Health
For decades, menopause has been treated as a monolithic event rather than the complex, multi-year transition it is. The conversation has been shaped by a 20-year-old study (the WHI) that, as subsequent analysis has shown, was misinterpreted for younger women. Compounding this, as of 2023, nearly 70% of OB/GYN residency programs had no formal menopause education.
The result is a fragmented care landscape and a corresponding data desert. Patient data, when it exists, is often unstructured, siloed in disparate EMRs, and lacks the granular detail needed to see subtle but critical patterns. You cannot train an AI on data you do not have. This is the core problem Winona’s model is, perhaps unintentionally, solving. Their primary mission is patient care, but the strategic byproduct of their platform is one of the most valuable, structured datasets being assembled in modern women’s health.
Architecting a Scalable Data Collection and Validation Model
The power of Winona’s model lies in its integration of technology, expert human oversight, and a unique partnership with the pharmaceutical supply chain. It’s a closed loop that transforms every patient interaction into a structured, validated data point.
1. The Digital Front Door: A Structured Intake Process
The patient journey begins with a comprehensive online intake process. This is the platform’s digital front door and its primary data collection tool. By guiding a patient through a detailed questionnaire covering medical history, lifestyle, and a wide array of symptoms, the system is not just gathering information for a doctor; it is creating a standardized, structured patient profile from the very first touchpoint.
Unlike a free-form conversation in a doctor’s office, this digital intake ensures that key data fields are consistently captured for every single patient. This consistency is the bedrock of any high-quality dataset. It ensures that when the data is analyzed at scale, it’s possible to make meaningful, apples-to-apples comparisons across a population of tens of thousands of women.
2. The Human-in-the-Loop: Expert Validation at Scale
The collected data is then routed to Winona’s national network of over 30 board-certified physicians. This is a critical step that elevates the platform beyond a simple software service. The telehealth consultation serves two purposes: providing expert, empathetic care to the patient, and acting as a crucial validation layer for the data.
The physician reviews the patient’s structured profile, uses their clinical expertise to ask clarifying questions, and makes a formal diagnosis. In the context of data architecture, the doctor is the “human-in-the-loop,” ensuring the information is not just complete, but clinically accurate. They provide the essential nuance and context that a simple questionnaire cannot. This validated data—linking a patient’s self-reported symptoms to a physician’s expert diagnosis—is immeasurably more valuable for any future analytical model. This model of personalized telehealth is what allows this high-level data validation to occur at a national scale.
3. The Bespoke Output: Connecting Diagnosis to Personalized Formulation
Here, the model’s elegance becomes clear. Winona’s physicians are not limited to a small formulary of mass-market drugs. Through partnerships with licensed 503A compounding pharmacies, they can prescribe bespoke medications with precise, personalized dosages.
When a doctor prescribes a custom-formulated cream or capsule, the platform logs that specific formulation and ties it directly to the validated patient profile. This creates a powerful, three-part data record for each patient:
- Input: The initial structured symptom profile.
- Validation: The physician’s diagnosis and clinical notes.
- Action: The exact, bespoke medication that was prescribed.
Over time, patient-reported outcomes provide the fourth and final piece: the result. This creates a complete, longitudinal data loop: [Symptoms] -> [Diagnosis] -> [Treatment] -> [Outcome].
The Strategic Byproduct: A Foundational Dataset for the Future of AI
While Winona’s operational focus is on delivering care, the strategic asset they are building is a rich, proprietary dataset. This is the “crown jewel” their internal document refers to. It is a clean, structured, and clinically validated collection of tens of thousands of patient journeys.
This is precisely the kind of “training data” that is a prerequisite for developing meaningful AI and machine learning models in healthcare. With a dataset of this quality and scale, one could eventually:
- Identify Novel Symptom Clusters: Analyze the data to find non-obvious correlations between symptoms that predict a patient’s response to specific hormonal combinations.
- Predict Treatment Efficacy: Build models that could predict, with increasing accuracy, the optimal starting formulation for a new patient based on her unique profile, reducing the trial-and-error period.
- Power Proactive Care: Identify leading indicators for more severe health risks associated with menopause, like osteoporosis or cardiovascular issues, allowing for earlier, preventative interventions.
The companies that will lead the next wave of AI in medicine may not be the ones with the best algorithms, but the ones with the best data. By focusing on a tech-enabled, human-validated system for delivering personalized care today, Winona is simultaneously building the engine that will power the predictive care of tomorrow. Their model serves as a powerful case study for any health tech innovator: before you can run the AI, you must first build the railway.
