Fertility is a critical aspect of women’s health, but one that’s too often shrouded in mystery due to the complexity of the factors at play: lack of education around reproductive health, misinformation about fertility in popular culture, stigma and lack of research around fertility issues, and the interplay of biological factors that make it difficult to make informed decisions. One startup is out to change this.
Rita, a women’s fertility startup, is using machine learning to empower women with data-driven insights when it comes to their fertility in a way that’s affordable and accessible.
“Seeking fertility treatments is often perceived as the only alternative for women over 30,” says Kamila Staryga, Rita co-founder, “but according to our medical experts only between 10-15% of the overall population in reproductive age needs them.”And, on top of that, many women don’t have access to knowledge about their own fertility until they’re actively trying to get pregnant.
“Despite the current narrative, 35 is not a fertility cliff,” says Kamila. “There are things women can do proactively to improve their fertility.” Rita wanted to build a platform that would provide women with insights into the factors that contribute to fertility, not just infertility treatment.
To accomplish this, the company needed a partner who could define the technical approach, evaluate an existing data set, and build an AI-powered diagnostic tool to help women understand the complex factors that contribute to their fertility and empower them to take control.
“A lot of companies ask the question – how do you treat infertility?” says Kamila. “But we wanted to shift the conversation to preventative health measures grounded in data. So we started with – what are the factors that even go into analyzing that? Which was really new ground for the fertility industry.”
To start, Rita wanted to build out a proof of concept (PoC) based on their existing data set – thousands of responses to a detailed survey on women’s health. The goal was to look into this data set and create a fertility assessment tool that women could use to evaluate their fertility based on 102 separate factors related to fertility health. So they needed a partner with a unique combination of technical expertise and experience with health data.
As a startup, Rita also needed to balance cost with speed and quality. So Tribe brought on a fractional expert in health data, natural language processing, and machine learning engineering to design a strategic approach to develop, evaluate, and productionize a model:
- Tribe Talent – Katie Baker is an expert in machine learning strategy and has advised dozens of companies on their 0 to 1 models. Katie has worked as a data engineer in healthcare and has built data pipelines that enable scalable production architecture for numerous machine learning solutions. She has also developed end-to-end models for risk adjusting patients for mortality and hospital readmission.
“We needed a talented machine learning expert who could analyze a messy and incomplete data set,” says Kamila. “Katie is not only a fantastic architect, but she was also passionate about the mission. She went beyond the job description. She never just accepted how we worked, she always asked why.”
The project's primary challenge was to build a platform that can leverage machine learning to provide personalized and actionable insights to women about their fertility status. The platform needed to be easy to use, accessible, and affordable. However, first team had to define a technical approach that would take into account the unique challenges:
- Limited data: The technical team needed to define an approach that would allow them to get signal from a small and incomplete dataset, which posed significant challenges for data analysis.
- Complexity of features: Fertility is a complex issue that involves many factors, such as age, lifestyle, medical history, and genetics. The project team needed to identify the most relevant features and develop an appropriate model to predict fertility outcomes.
- Scalability: While the project would begin with a small data set, it needed to be flexible enough to incorporate additional survey data over time, even as questions changed.
The project started with a survey to collect data on various factors that could impact fertility, such as sleep, water intake, period cycle length, and abnormalities in hormones or medical history. The survey was designed with obstetricians and embryologists who worked in the reproductive health space. The team collected over 1000 survey responses and analyzed the data to determine if there was any signal, or meaningful patterns, in the data.
“Our initial process was just to get a lot of viewpoints on the data and approach to determine what patterns we’re seeing,” says Katie. “And, in the end, we were able to find meaningful patterns in a quite limited data set. From there our thought process was: now what does this look like as a product?”
“We landed on an augmented approach,” says Katie. “Because as we got into product design discussions, it became clear that a simple score or classification wasn’t sufficient from a user experience perspective.”
From there, the team defined the larger technical approach to building a fertility product informed by the patterns they were seeing in the data:
- PoC: The project team worked on two parallel pathways – data exploration and software engineering. The PoC was based on a statistical analysis of the data to help identify the most relevant contributors to fertility such as sleep, water intake, hormones, menstrual cycle length, medical history, and more.
- Model refinement: The project team used a factor-based model to predict the likelihood of fertility and flag any health concerns or need for intervention. The team also developed a metadata-based framework that involved a series of answers to questions. The framework created factors that could be good, neutral, or concerning, and recommended various levels of interventions based on combinations of features. This metadata approach also gave the team flexibility to add more surveys and questions as more data is collected.
- Neural network development: The team used a neural network as the main scoring methodology to produce the likelihood of falling into one of three categories: able to conceive naturally in the first six months of trying, still trying even after a year, or being able to conceive either after six months or with some medical intervention.
- Codebase and API development: The project team developed a framework for interpreting metadata that is entirely metadata-driven to anticipate rapid evolution. The backend API was able to integrate with the frontend that can pass answers to survey questions back to it.
The goal was an MVP which would provide users with insights and personalized recommendations based on their survey responses, along with a classification of their fertility status. The approach taken for the tool was a blended approach that used a factor-based model augmented with a neural network. The framework was metadata-driven, and a backend API was developed to support easy iteration and flexibility around the model.
In the end, the team successfully created a high-performing MVP with a limited amount of data and a high complexity of features. The augmented approach, which was a blend of traditional statistical analysis and a supervised neural network, produced more accurate results. And, because the code base is entirely metadata-driven, the product can keep up with a rapidly evolving data set and will allow the team to expand into unsupervised analysis once more data is available.
During the project, the Tribe team was able to scale up from one to two engineers working five to 25 hours a week depending on the project needs.
“We are not the customer that has endless budget, but we’re also not willing to compromise on the quality of the work,” says Kamila. “Tribe worked with us to make this happen within our budget. When the time came, we were able to pair Katie with a more junior resource to extend our runway.”
“And there were additional benefits to working with fractional talent too,” Kamila added. “Katie was able to see our project through the lens of the work she does for other clients. So she’s not just bringing ideas about something she’s read, but applied, hands-on experience. Because this is such a new domain, this was vital for us.”
In March 2023, Rita founders Kamila Staryga and Alessia Morichi launched the beta version of its first product to the US market: a comprehensive digital fertility health assessment for women who hope to prepare proactively to one day achieve pregnancy. The assessment takes only ten minutes and give users a comprehensive analysis of their entire fertility picture including:
- An analysis of 102 factors that takes into consideration their unique experience and fertility journey and highlights what might need attention
- Personalized and easy-to-understand insights with results grouped by category
- Actionable next steps that user can implement on their own or discuss with their doctor
"The area where I tend to offer the most value for my customers is how to take an idea and get it to production within a scalable software solution," says Kaite. "I work with customers who are thinking about their future growth, and want a deeper understanding of how their ML requirements are going to evolve and how they can scale their solutions."
“Throughout the project, Katie was an indispensable member of our team,” says Kamila. “I trust her completely and even included her as part of our leadership team in our fundraising materials. When we started, we weren’t ready to commit to hiring someone full time, but wanted someone excellent to help us build this product. Working with Tribe was the best way to access that talent.”