Safrn Health

1 in 3 women experience a complex or chronic condition. These conditions often manifest differently in women as compared to men, due to unique biology and higher rates of comorbidities. 

We are starting with endocrine conditions because they often serve as the root of high cost, high acuity complications. (For example: for PCOS, over $4 billion of the $8 billion total spend is attributed to comorbidities, reproductive complications, and hospitalizations which could be reduced with early intervention. And we see similar patterns in other endocrine conditions.) Endocrine conditions are highly complex given they span multiple organ systems, and disease progression is highly influenced by social and behavioral health context. 

Through 200+ interviews, we learned that patients are ill-equipped to connect the dots between multiple care providers and are unaware of the symptoms & questions they should surface to their providers. This hinders their ability to self-advocate. Providers, despite best intentions, have an incomplete picture of a patient’s health since the EHR is missing relevant social context. This leads them to make care decisions without complete info, resulting in billions of dollars of avoidable spend and worse health outcomes.

We’re building an ML platform to help clinicians better manage female-focused chronic conditions before they escalate into avoidable, costly comorbidities. Using LLMs, we capture patient social context and layer it with medical history to assess risk factors and deliver personalized interventions that providers can review with their patients. 

Here’s how it works:

1. Our mobile-first, patient engagement tool uses conversational AI to capture nuanced, relevant patient context (e.g. family history, trends in diet and exercise habits, living situation). 

2. Our model, trained on longitudinal, female-focused chronic condition data & peer-reviewed research, synthesizes patient context and layers it with data we pull from the EHR to perform a risk assessment

3. We generate a risk report for physicians and draft a hyper personalized care plan that they can review with their patients 

Managing risk early on and delivering interventions that women are likely to adhere to ultimately tackles billions of avoidable spend and improves the health outcomes of women struggling with these conditions. 

For patients:

We are starting with women who have endocrine conditions (i.e. PCOS, thyroid disorders, diabetes). Our platform enables patients to become better self-advocates for themselves, allows them to feel heard and engaged in between appointments, and facilitates trust between patients & providers. Our ultimate goal is to improve both patient experience and patient outcomes.

For providers:

We are starting with OBGYN and endocrinology practices. We are delivering value on a few dimension: 

1. 1. Time + efficiency: right now clinicians face burden of manually processing paper intake forms, parsing through dozens of pages of medical notes, and the EHR only captures a fraction of that data – our platform helps streamline that process

   2. Better care: we equip providers with the full patient context to enable them to deliver better care and keep more complex patients in-house

   3. Patient experience: patients often churn when they have poor experience and we help facilitate better patient engagement and trust

For payers:

Currently payers are spending over $50 billion dollars for excess spend on escalations of chronic conditions. By intervening early on, we believe we can reduce this spend and deliver better health outcomes for their member population.

Our team brings passion and expertise to tackle the women’s chronic health space. 

Anjali’s strength is surfacing a clear vision to tackle complex, interdisciplinary problems that affect multiple stakeholders. Anjali’s background is in software engineering, product strategy & UX design. Anjali spent 5 years at Salesforce building user interfaces and delivering insights on customer journeys before getting her MBA at Harvard Business School. On a personal note, Anjali experienced the inefficiencies of the healthcare system from the patient side after many years of navigating undiagnosed PCOS. Anjali’s experience compelled her to apply her tech & product skills to women’s health equity.

Alice brings operational focus to abstract problems and loves building strong communities in areas she is passionate about. Alice’s background is in finance & operations. Alice spent several years investing in and working with companies in the chronic health and women’s health spaces before getting her MBA at HBS. She has also freelanced / worked part-time in marketing, branding, photography, and social media design. On the personal side, Alice was diagnosed with a thyroid condition at the age of 13 and has experienced the pain point firsthand. 

Nora works at the cutting edge of technical capabilities and loves to solve the most meaningful problems in healthcare. Nora’s background is in AI/ML engineering and computational analysis in the health & wellness  space. Nora has extensive experience building ML models most recently at ScienceIO. She has also worked in data roles at Maven Clinic and ClassPass. 

We are innovating through the following ways: 

1. Physician champions: we’re building alongside physicians from the start and enabling a strong feedback loop so we can quickly integrate their modifications to our analysis or care plan back into the model

2. On the data and ML strategy, we are securing proprietary data from health systems to boost our model’s accuracy

3. We are capturing social and behavioral context from patients and training our model with that data when most other competitors are just looking at data within the EHR
   

Nearly ~75% healthcare spend is due to chronic conditions and we believe this is a huge problem to tackle. Even though we are starting with women with endocrine conditions, our mission is to become the trusted partner for women's chronic conditions at every stage of life, including reproductive health, pregnancy, post-partum, menopause and beyond. We believe that by tackling health at the root, we can transform the way disease progression occurs for millions of people over time.

We are testing a range of foundational models from closed source + prompting to open source + fine tuning approaches. LLMs are progressing quickly, and it is important that we build our infrastructure in a modular way in order to test newer LLM features and/or models. We will want to assess and fine-tune our models specifically for each task in our product offering (ie intake chat feature, vs health summaries, vs care plan creation vs health coach), and acquire training/fine tuning data where possible.

Health Intake LLM: Health intake is static - standard questions that don’t change or update as a patient answers the questions. There are no follow ups or deeper/contextual questions asked depending on a patient’s answer. Intakes also don’t usually ask more behavioral/lifestyle questions, or at least with the nuance that may be useful to doctors. These intakes are also long and not layered into other health data, meaning a physician doesn’t have a summary of important points or a full picture of health. Using LLMs for this type of patient intake allows for a dynamic and engaging approach which should result in more data for providers.

Initially, feedback on our conversational/LLM powered features will come from pilot partnerships with provider practices and their opted-in patient-user base. We may also do patient facing user testing with friends & family or direct recruitment to get feedback on our patient facing conversational features. As we measure data and feedback from our pilot programs, we will incorporate this data as part of our LLM development and fine-tuning approaches.

Risk Model: with a target audience of OBGYNs, they often don’t have the same expertise as endocrinologists in terms of endocrine conditions, how they might present, and what risk factors and or comorbidities exist b/c of that condition. By training a risk model in partnership with endocrinologists, we help extend care to OBGYNs and provide a tool for them to give better care to their patients w/ these conditions. By empowering OBGYNs, we are helping increase the number of providers who can can for, and assess risk appropriately, for these kinds of conditions, while also extending practice/appointment/service opportunities and efficiency for partner-providers. 

Initially we can work on finding some publicly available datasets on endocrine conditions and risk factors, but we are also securing private data partnerships from hospitals and/or research centers that have complete and relevant data on our target patient population, and risks that we would like to identify. These partnerships are necessary to develop a classification model that can actually be useful in determining risk factors, based on health indicator + lifestyle data. 

We will build and certify a HIPAA and SOC2 compliant tech stack. Many of these details include - encryption of data, secure production environments with policies of least privilege, signing BAAs and ensuring that any 3rd party technology is also HIPAA and SOC2 compliant, regular pen testing of our product and endpoints, and ability for users to consent to data sharing, and revoke and wipe access as requested.

Measuring bias will be important in our data sets - for instance, if we secure a data partnership for training our risk model from an academic hospital with a majority wealthy population, we will need to supplement this data with data from a public hospital serving a different demographic group. In the meantime, we can use data augmentation techniques, build partnerships with populations and demographics that will help to fill out our dataset, and also use our models in a supervised environment where the limitations are understood. For a model like risk classification, it will be important to 1. assess missing/biased data 2. Recruit and fill out these data pieces to build a more inclusive model, by working with partners, or populations directly.

In the next year, our goal is to build out our conversational intake and risk model products for providers and launching this to at least 10 clinicians  across OBGYN, endocrinology and fertility clinics.

In five years, we hope to be able to demonstrate improved health outcomes and cost savings for health systems and payers in the ecosystem. 

2 full-time 

1 part-time

1 clinical advisor

1 intern

We have been working on this solution for 8 months.

We are a fully female-founded team with diverse racial backgrounds. We are very committed to promoting equity and inclusion in our hiring practices as it is one of our core philosophies, personally and professionally.

Our operating plan for the next year is as follows: 

1. Data acquisition and model training ~ 9mos (can happen in parallel with other stages) 

   1. We are in productive conversations with Brigham and early conversations with Mayo Clinic (through their accelerator program) to get access to proprietary data that we can train our model with. 

   2. Both systems offer creative ways to get access to their data - through an equity share or cash component totaling to a $300K convertible note. 

2. Patient Pilots with Clinician ~6mos 

   1. We plan to roll out our conversational intake form to patients through our clinical advisor at the University of Washington over the next 6 months. 

   2. Here, we are primarily assessing patient engagement, integration of insights into clinical workflows & form factors that result in the most meaningful outcomes to guide product development 

   3. The main costs here include compensating our physician champion and software/compute costs for patient product development. We think this would cost $50-$100k.

3. Designing and launching a patient study with the Connors Center at Brigham Hospital ~9-12 mos

   1. In conversations with Connors Center currently about study design and data capture 

   2. Testing 2 key pieces

      1. Whether our ML platform predicts risk factors for women with the same or improved accuracy as a provider 

      2. Whether our conversational intake captures meaningful patient nuance that cannot be captured through patient intake forms and that would result in a shorter time to diagnose, more personalized guidance, more patient buy-in into the plan

      3. If through 1 and 2, we see a meaningful change in clinical outcomes → decreased risk for comorbidities, decreased hospitalizations during pregnancy, maintenance of current state vs. escalation of chronic condition over time 

   3. We estimate this study would cost $50-$100k based on similar projects they’ve conducted in the past

4. Clinician MVP ready ~ 12 mos

   1. We are in talks with several OBGYNs/Endocrinologists who have been providing feedback on the clinician value prop, and plan to convert these informal advisors into pilot partners once we have further developed the clinician outputs

   2. This MVP is dependent on our ability to fine tune our ML model based on the data partnerships we secure and the learnings we get from running a study with the Connors center. While we can begin fine-tuning our model based on patient and provider feedback, the study results may take 12-16 months to feed into our model. 

   3. The bulk of our costs here will be engineering costs (for 2-3 hires) and training costs, totaling ~$300-$500k  

Our long-term revenue model is selling to payers. Currently payers are spending over $50 billion dollars for excess spend on escalations of chronic condition. Even if we could address 1%, of those spend for payers and can demonstrate a 3x ROI, that puts us at $160 million in revenue. In order to get these contracts, we know we have to demonstrate ROI, we are starting with selling to clinics + health systems and working closely with physicians to build real-world evidence. 

Through initial benchmark, we think we could charge $500-$600 a month.

We are currently bootstrapped and are operating at a loss of ~$400k per year attributed to personal costs and general and administrative expenses (mostly comprised of software fees for the business). 

Next year, this number would increase to ~$750k when we take into account the cost to structure a pilot as well as technology and infrastructure costs.

We are seeking $100k for the purpose of structuring a pilot and training our model. We have come up with this number through conversations and benchmarking with other companies that have completed similar pilot projects. We would spend~75% of the money on R&D and ~25% on personnel. The R&D spend would be used towards technology and infrastructure investments, including data acquisition, compute, storage and training. The remaining money left would be used to compensate physicians for their time helping train and provide feedback on our model.

We are excited about the Cure Residency given its alignment with our company mission and it's focus on the cutting edge of AI and healthcare. While we are excited about the potentially large impact that our solution can bring to the healthcare world, we understand that the healthcare ecosystem is complex and difficult to navigate. Thus, we are excited for the mentorship and access to expertise that can help pressure test our idea, accelerate our growth, and help us pivot quickly if we’re working on the wrong things