Lessons from Neil Sanghavi, President and Head of Product and Solutions at Atropos on developing publication-grade evidence for healthcare and life sciences
Neil Sanghavi, President of Atropos Health, an organization that provides on-demand evidence from real world data to close evidence gaps in healthcare and life sciences.
Subscribe to our substack for updates and listen on Apple Podcasts or Spotify. Connect with Andrew or Lipsa if you find this post insightful and want to learn more.
Welcome back to the Pear Healthcare Playbook! Every week, we’ll be getting to know trailblazing healthcare leaders and dive into building a digital health business from 0 to 1.
Today, we're thrilled to introduce Neil Sanghavi. Neil began his journey with Atropos Health in 2020, initially focusing on product and launch strategy part-time, before taking on the role of President in 2021. Before joining Atropos, Neil lent his expertise as an advisor to prominent health tech startups such as Turquoise Health, Abridge, Tuva, and Ribbon Health, where he contributed to go-to-market strategies, data strategy, analytics team development, and network contracting models.
In addition to his advisory roles, Neil was the Provider Strategy & Analytics leader at Haven, the joint venture formed by Amazon, JPMorgan, and Berkshire Hathaway. His career also includes strategic positions at the Cleveland Clinic and The Advisory Board Company (now part of Optum), as well as leadership roles in product and commercialization at healthcare startups like Centivo.
In this episode, Neil and I explore his journey into health tech, the role of data in healthcare, and Atropos Health's growth. We discuss their go-to-market strategy, the development of ChatRWD, and the importance of data quality. Neil also shares his vision for the future of Atropos and offers advice for founders.
If you want to listen to the podcast click below:
Being an Advisor to Health Tech Companies
For those interested in becoming advisors to health tech companies, it's essential to begin with a genuine passion for the company's mission.
Instead of focusing on getting an advisory role, it's better to understand what the company is building and why. Formal advisory positions are rare—only about 1-5% of conversations lead to one.
The goal should be to explore whether the company is working on something interesting where you can offer valuable insights. By asking questions and sharing your thoughts, you can build meaningful connections.
"I would be compelled and respond positively if someone came to me with a similar approach at Atropos Health—wanting to understand why we're building what we are and how we're going about it. Even more impactful would be if they shared their thoughts on the space and offered their perspective, asking if it aligns with our vision or differs."
When discussing similar issues or spaces, sharing thoughtful perspectives can lead to impactful conversations.
In early-stage companies, founders often feel isolated, so demonstrating a deep understanding of the industry and providing relevant opinions can be a strong way to connect.
However, it's important to recognize that forming an advisory relationship usually takes time, often months or even years. The focus should be on building genuine relationships and offering value without expecting immediate formal roles.
Starting with Atropos Health
Atropos Health was founded in 2019 starting with the innovative “Green Button” technology developed at Stanford University. The concept originated when Stanford Health Care physicians frequently requested insights from Dr. Nigam Shah, seeking guidance on patient treatment based on data from Electronic Medical Records (EMR).
Recognizing the potential to improve clinical decision-making, Dr. Shah created what became known as the “Green Button” service—a tool designed to analyze Stanford’s EMR data and deliver evidence-based consultation reports to physicians.
The success of this service at Stanford,, demonstrated the tremendous potential of harnessing real-world evidence to guide clinical decisions.
Brigham Hyde—formerly an executive at life sciences firm Eversana—and Saurabh Gombar formed a new company with the Green Button technology.
Thus, Atropos Health was born, aiming to bring publication-grade real-world evidence to the broader medical community and assist clinicians in answering complex clinical questions, such as determining the most effective treatments for cancer patients.
Stanford University’s role in the formation of Atropos Health reflects its best-in-class approach to nurturing innovation and supporting the commercialization of groundbreaking technologies.
"Stanford is truly first class at this—they really know how to incentivize effectively while recognizing that the majority of the work is still ahead. A common mistake big organizations make is thinking they should have the majority control from day one because they’re the anchor customer and own part of the technology. But the reality is, the company is essentially useless on day one—it doesn’t accomplish anything at that stage. Stanford understands this well. They realize that very little has actually been built yet and that their success lies in taking a relatively small piece of the company, allowing it the potential to grow and deliver outsized returns."
This strategy, which involves taking a relatively modest stake in the newly formed entity while leaving the majority of the equity and revenue streams available for the new company’s future stakeholders, significantly enhances the likelihood of success.
With Stanford as a key partner, Atropos Health was able to start strong, benefiting from direct product feedback, a robust initial user base, and ongoing support that has positioned it to set new standards in personalized evidence.
Neil joined Atropos in 2020 and was drawn to Atropos Health for several compelling reasons. Initially, what captured his interest was the groundbreaking “Green Button” service that originated as an academic project at Stanford. This service addressed a significant leap forward in closing the evidence gap.
When physicians face a treatment decision without clear guidance from traditional resources like UpToDate or medical literature, their options are often limited. They may turn to PubMed, consult a specialist, or seek advice from personal networks. However, these methods rarely provide immediate, evidence-based answers, especially when even specialists are uncertain in their clinical decision-making.
The "Green Button" service revolutionized this process by enabling physicians to ask specific, real-time questions, such as whether Denosumab or zoledronic acid (both bone-strengthening medications) is more effective in preventing fractures in female oncology patients. This then triggers a publication grade observational research study or a real-world evidence study using Stanford's extensive EHR data.
Remarkably, the service delivers a Prognostogram back to the requesting physician within days—a dramatic improvement over the typical six months or more required for such studies in an academic setting.
What particularly impressed Neil was the speed and efficiency of this process. Over 18 months, the service conducted around 150 to 200 consults across 18 different specialties and even had a manuscript accepted by NEJM Catalyst when Neil first engaged with the team.
The idea of offering a service that could answer bespoke clinical questions based on vast datasets was a new frontier in medicine.
The fact that Atropos Health was successfully implementing this vision, running thousands of studies for doctors and researchers globally, made it a compelling opportunity to join and help scale the company.
Leveraging Healthcare Data
In medical practice, the gold standard for evidence-based medicine remains the randomized control trial (RCT). However, not every medical question can be addressed through an RCT due to challenges like obtaining equitable sample populations. These challenges stem from the frequent exclusion of diverse demographic groups such as ethnic minorities, the elderly, and those with multiple comorbidities, leading to a lack of representation that can skew results and reduce the applicability of findings across the broader population (source).
Healthcare decisions stem from varying levels of evidence. In fact there is a hierarchy of the level of evidence below.
Currently, when RCTs or meta-analyses do not offer answers, the standard practice leans heavily towards seeking expert opinions or specialist consultations. These methods, though valuable, sit at the bottom of the evidence pyramid, indicating a reliance on less rigorously tested knowledge.
“If you think about it, the evidence pyramid places RCTs and meta-analysis at the top and expert opinion at the bottom. You go directly from the top of the pyramid to the bottom if you don't have the answer to your question. So, there must be something in the middle, right? Something that's above expert opinion but below meta-analysis and RCTs. That’s what observational research is. And if it's done well, it can be a very useful addition to medical knowledge and practice.”
Recognizing this, Atropos Health aimed to position itself strategically within this continuum by leveraging observational research that offers a valuable middle ground. Observational studies can enhance the medical knowledge base by providing insights where RCTs are not available. These studies draw from real-world data, which, while not as controlled as RCT environments, reflect a broader spectrum of patient experiences and outcomes.
Let’s say a physician is trying to decide between putting an oncology patient on Drug A (Denosumab) vs. Drug B (Zoledronic acid) to prevent fractures since both are bone modifying agents. The physician will submit the question to Atropos. Upon receiving a question, a physician on the Atropos team reviews it for clarity and completeness, asking follow-up questions if necessary. Once the query is fully understood, it is translated into an appropriate epidemiological study design, usually a cohort study comparing two treatment options (e.g., drug A vs. drug B).
To ensure rigorous and unbiased comparisons, Atropos uses advanced techniques such as high-dimensional propensity score matching. This methodology involves selecting features to adjust the data properly, ensuring that patients in different cohorts are as similar as possible across numerous variables. This comparison is not only meticulous but is also automated within Atropos’s technology stack, which can rapidly define and analyze patient cohorts.
Further enhancing their system, Atropos has integrated large language models (LLMs) to automate even more steps of the process, reducing human error and increasing efficiency. The final analysis undergoes a thorough review by a medical director before the findings are returned to the requester. This streamlined, automated process enables Atropos Health to deliver accurate, evidence-based responses quickly, often within a two-day timeframe, and even within hours for urgent requests.
For the real-world example involving oncology patients, a physician may be deciding between Denosumab and Zoledronic acid as bone-modifying agents to prevent fractures. Atropos presents a prognostogram result below showing no significant difference in skeletal-related events between the two drugs.
Without adjusting for confounding factors, an initial analysis might suggest a difference. However, by using advanced techniques like high-dimensional propensity score matching, Atropos can accurately adjust for variables that could bias the comparison, ensuring a more balanced and reliable result. This statistical adjustment highlights that, once properly controlled, there is no meaningful difference between the treatments, offering clinicians a clearer understanding to guide their decision-making.
Clients Atropos Health Serves
Atropos Health primarily serves a unique clientele: physicians who function in three pivotal roles simultaneously or in one of the three. These clients are typically affiliated with academic medical centers where they engage in direct patient care, are involved in developing care protocols, and conduct research. This multifaceted role allows them to uniquely benefit from Atropos Health's services.
Clinical Practice: These physicians are on the frontline, providing care to patients. Their direct interactions with patients often generate specific clinical questions, especially when encountering complex cases not covered by existing guidelines or evidence.
Protocol Development: As part of teams that develop treatment protocols, these physicians look to integrate the latest data into clinical guidelines. They seek evidence not only to optimize patient outcomes on an individual level but also to shape the collective approach to patient care within their institutions.
Research: The third role involves academic research. Initial inquiries made to Atropos often stem from recurring clinical uncertainties. For instance, a physician might notice a pattern or a recurring challenge in treatment efficacy and turn to Atropos to seek data-driven insights. The results from Atropos can provide the empirical evidence needed to challenge or change existing protocols.
When Atropos provides evidence that impacts a physician's understanding or approach to treatment, it often leads to broader discussions within their department or at community meetings. If the findings are novel and significant, they might be considered for publication and could even influence industry-wide guidelines. Despite the common belief that guidelines are predominantly based on large RCTs, they frequently incorporate findings from observational studies and other forms of original research provided by services like Atropos.
To date, Atropos has contributed to 50+ publications that originated from these unique clinical scenarios or research questions. Some of these publications have even been incorporated into new clinical guidelines, demonstrating the significant impact of integrating real-world data and advanced analytical methods into everyday medical practice and broader healthcare protocols.
Atropos Health also provides strategic value to life science companies through GENEVA OS™ (Generative Evidence Acceleration Operating System). This platform is engineered to accelerate evidence generation for life science teams and healthcare organizations by integrating advanced AI technologies with real-world data (RWD).
GENEVA OS offers a comprehensive suite of software, services, and data solutions designed to efficiently meet the diverse and rapidly evolving evidence needs of the industry. Key functionalities include supporting preclinical and clinical research, optimizing clinical trial design, facilitating regulatory submissions, and enhancing strategies in medical affairs and Health Economics and Outcomes Research (HEOR).
Additionally, the platform aids marketing and corporate strategy teams in product comparisons, identifying evidence opportunities, and expanding market reach. By enabling clinical trial emulation and fostering health system collaborations, GENEVA OS empowers life science companies to derive actionable insights, drive business development, and achieve a competitive edge in the marketplace.
Atropos’s GTM Strategy
Atropos Health's go-to-market (GTM) strategy began with a landmark partnership with Stanford University as its first paying customer. The company then successfully expanded its reach to other academic institutions, with Mayo Clinic becoming the second notable collaborator. This progression solidified Atropos Health’s credibility in the medical academic sphere and was a crucial step in demonstrating the value of its real-world evidence platform.
“Going from one top-tier academic site to another was really validating for us. It demonstrated the compelling need and effectiveness of our solution in a highly demanding environment.”
The positive feedback and robust data from these academic collaborations reinforced the company’s strategy to continue its strong push on the provider side.
As Atropos Health's reputation grew, the company expanded into the life sciences sector, capitalizing on its ability to produce rapid, publication-grade observational research studies.
These companies are typically focused on developing new drugs or repurposing existing drugs for additional indications.
“We started to get inbound interest from pharmaceutical companies who were impressed with our ability to deliver detailed evidence studies quickly. They recognized the potential to use our platform to accelerate their own research efforts.”
This interest from the pharmaceutical sector underscored the broader applicability of Atropos Health’s solutions beyond academic settings, bridging the gap between clinical research and market needs.
The company's engagement with life sciences started with Johnson & Johnson, leading to broader adoption within the pharmaceutical industry. Currently, Atropos Health serves seven of the top 10 pharmaceutical companies, showcasing significant growth and an expanding customer base in this sector.
This dual focus on health systems and life sciences demands a strategic approach to team building and product development to cater to the distinct needs of each sector without compromising on service quality or technological advancement.
Following a successful Series B funding round, Atropos Health has been able to invest more deliberately in its team and product development, expanding its reach into meeting the needs of each sector.
“We entered the life sciences market carefully because we understood that the buyers in this space have specific expectations and established ways of purchasing real-world evidence solutions. We aimed to integrate into their workflows without completely reinventing our product or disrupting our roadmap."
To further tailor its approach across diverse customer bases, Atropos Health invested in dedicated sales teams for health systems and life sciences. This targeted strategy reflects a deep commitment to effectively addressing the unique challenges and opportunities presented by each sector. Such specialization forms a core part of Atropos Health’s broader strategy To accelerate the generation of actionable evidence to improve healthcare outcomes for everyone.
Developing Products and ChatRWD
Atropos Health strategically builds its products by leveraging a structured team approach, led by Dr. Saurabh Gombar, the Chief Medical Officer and co-founder. The team has a rigorous two to three-week training program, learning to navigate Atropos’s proprietary software and code in a domain-specific language.
The medical team, hailing from various medical specialties, are crucial in translating user questions into viable study designs and conducting the studies, full-time medical directors meticulously review each study before it reaches the requester.
This system not only maximizes efficiency but also maintains high standards of quality across the services provided by Atropos Health.
Building on its robust foundation of healthcare analytics, Atropos Health developed its latest feature, ChatRWD, to enhance rapid decision-making in clinical environments.
“We've been captivated by the potential of large language models (LLMs) and various new analytical models to enhance our business. A unique advantage we possess is our three and a half years of experience running services manually. Consequently, we have amassed an extensive content library. We've conducted thousands of studies for physicians, and independently completed another 8,000 to 10,000 studies based on questions we found intriguing, all created for future use.”
Atropos Health developed ChatRWD by leveraging its extensive repository of over 10,000 completed studies, each enriched with clinical definitions known as visual phenotypes.
These phenotypes, meticulously define conditions such as diabetes and heart failure within electronic health records and claims data. With these validated phenotypes at their disposal, Atropos set out to enhance the speed of their service—transitioning from question to actionable insight.
The development process involved utilizing large language models (LLMs) as the primary technology to interpret a user's query, compose it into identifiable phenotypes, and seamlessly integrate these into predefined study templates.
This methodology mirrors the tasks traditionally performed by physicians, but significantly accelerates the process. Instead of the conventional wait of a day or more to receive preliminary findings, ChatRWD provides these insights within minutes, functioning much like an AI co-pilot.
ChatRWD Performance Against other Models
In a comprehensive study, Atropos Health explored the efficacy of five different LLM-based systems in addressing 50 clinical questions, with a focus on the relevance, reliability, and actionability of their responses. This evaluation involved nine independent physicians who critically reviewed the answers provided by these models. The LLMs tested included three general-purpose models—ChatGPT-4 by OpenAI, Claude 3 Opus by Anthropic, and Gemini Pro version 1.5 by Google—as well as OpenEvidence and Atropos's own ChatRWD.
OpenEvidence integrates an LLM with Retrieval-Augmented Generation (RAG) technology to access a diverse range of medical literature, such as PubMed articles and FDA drug labels. This approach minimizes errors by summarizing relevant literature and presenting conclusions with citations, which enhances the reliability of the information provided (source).
ChatRWD leverages a language model with an agent, creating an agentic system that combines a natural language interface with an evidence generation platform accessing medical record data. In this system, the LLM acts as a co-pilot, aligning clinical intent with a purpose-built agent dedicated to evidence generation. This process starts with Chain-of-Thought prompting, which transforms plain English queries into the structured PICO format (Population, Intervention, Comparison, Outcome), identifying the appropriate study design and relevant entities. Subsequently, ChatRWD performs a semantic search within a curated library of phenotypes and generates Temporal Query Language (TQL) code for cohort selection. This is followed by a statistical analysis on Atropos's specialized platform. The final results are then concisely summarized, providing clear and actionable insights. The data source for ChatRWD is Eversana’s Electronic Health Record Integrated Database, which includes comprehensive records from 63 million patients across various U.S. healthcare settings, including detailed medication, laboratory, procedure, and diagnosis data (source).
Currently, general-purpose LLMs such as ChatGPT-4, Claude 3 Opus, and Gemini Pro 1.5 seldom deliver answers that are both relevant and evidence-based, achieving success rates between 2% and 10%.
In contrast, systems using retrieval augmented generation (RAG) and agentic models, such as OpenEvidence and ChatRWD, have shown significantly better performance.
OpenEvidence yielded relevant, evidence-based answers for 24% of the questions, while ChatRWD achieved a higher rate of 58%.
Furthermore, reviewers frequently identified ChatRWD as providing the best response 60% of the time, with OpenEvidence following at 46%. However, no single LLM consistently emerged as the best across all reviewers.
Currently, ChatRWD is being utilized primarily by existing Atropos Health customers who already subscribe to other features, such as the 'Green Button' service, and are actively involved in its beta testing. This phase also includes participation from a select group of external stakeholders, including investors and partners, who provide essential feedback to refine the product before its broader release.
ChatRWD's performance in generating relevant, evidence-based answers is set to be further leveraged through a strategic licensing model that mirrors the 'Green Button' system. This model, which is based on user numbers and query volume, will be adopted as the solution transitions from beta testing to general availability.
This approach promotes scalability and affordability, enabling a diverse user base from large healthcare organizations to individual researchers.
Ensuring the Most Robust Data Quality
Atropos Health has introduced a sophisticated methodology termed Real Word Data Score™ and Real World Fitness Score™ “data fitness scoring.”
“There are far more datasets available today than ever before. However, this abundance makes it challenging to determine which dataset is most suitable for a specific question. To address this issue, we have developed a methodology to evaluate the fitness of a particular dataset for a given research question.”
Data fitness scoring incorporates various parameters to assess the suitability of a dataset for specific research needs. This evaluation considers whether the dataset contains the necessary phenotypes, meets the patient criteria, and includes relevant outcomes. For instance, if a study requires data on patients with diabetes who are taking metformin and have a hemoglobin A1C level above seven, the fitness scoring checks whether the dataset includes the necessary lab results.
If a dataset lacks the required lab data, such as hemoglobin A1C levels, it indicates that it may not be suitable for that particular study. Researchers can run their designed studies on multiple datasets and receive a fitness score for each, which helps in determining the reliability and trustworthiness of the results. Each report generated is assigned a numerical value (100 is the highest), reflecting the fitness of the dataset for the specific query.
This scoring system is particularly useful on a relative basis, but it also offers absolute value; for instance, a score below 50 often signals significant issues with the dataset for the intended research question. This insight allows researchers to select an alternative dataset more aligned with their study requirements.
In the near future, the application will enhance this process by automatically suggesting the most appropriate dataset based on the researcher's query. It will identify which datasets are unsuitable and recommend those that are likely to yield higher quality results, streamlining the research process significantly
Future of Atropos Health
Looking ahead, Atropos Health envisions its development across three key horizons.
The first horizon focuses on expanding its presence within the top 20 pharmaceutical companies and leading academic medical centers across the U.S. over the next 6 to 12 months.
The second horizon aims to extend the service to more traditional healthcare providers, enabling them to access a comprehensive content library and evidence-based resources even if they are not conducting research daily.
For example, a doctor could ask how a patient would be treated at the Mayo Clinic and receive a data-driven answer without sending the patient there. This capability is already available today, showcasing the practical applications of Atropos Health's tools in everyday clinical settings.
The third horizon involves international expansion, where Atropos Health plans to partner with global data vendors to bring its innovative solutions to healthcare professionals worldwide. This expansion is already underway, with the company having secured investors who represent international data sets and entities.
This multi-phase approach positions Atropos Health to significantly impact both domestic and international healthcare landscapes.
Advice for Founders
What are the characteristics Neil believes need to be present for a successful company?
In the earliest stages of a startup, three key characteristic are essential for building a successful company:
Curiosity: Hiring people who are genuinely curious and excited about the work is often the biggest accelerator of success. Curiosity needs to be intrinsic—driving continuous exploration and problem-solving. This mindset helps transform a single idea into a company that truly makes a difference.
Bias for Action: Curiosity alone isn’t enough; it must be paired with decisive steps to satisfy that curiosity and move the company forward. It's about hiring individuals who are not just thinkers but doers, willing to take action and push through barriers.
Integrity: Working with high-integrity people who can be trusted to do what’s right for both the company and society is critical. Integrity ensures that decisions are made with the company’s best interests at heart, creating a foundation of trust and ethical behavior.
When curiosity, action, and integrity are combined, good companies can be built regardless of the industry. Surrounding a worthwhile project with people who embody these qualities significantly increases the likelihood of achieving meaningful results.
What are strategies to hire the right person for your company?
Traditional interviews may not always be the best method for assessing a candidate's abilities, as candidates can prepare and provide answers that align with what the interviewer wants to hear.
Instead, task-based assessments, where candidates are given specific assignments related to the job, offer a more accurate evaluation of their skills and problem-solving abilities. It's much harder to fake the execution of a real task.
For sales roles, this might involve pitching a product, while engineering roles could require a technical challenge. On the product side, candidates who engage critically with the company's product and ask insightful questions are often a better fit.
“You have two options when you interview for a product role at Atropos: you can either pick one of our products to analyze or choose a product you are familiar with—both are acceptable options. If someone chooses our product, we can quickly determine if they have thought critically about it and can instantly assess whether they understand what we are trying to achieve and whether they will be able to get up to speed and contribute to what we are working on.”
Once the interviewee is deemed a good fit, establishing a contractor relationship can be an effective strategy.
In this arrangement, the individual works for the company for a brief period, such as a month or two, on a paid basis while being evaluated for a potential long-term position.
This approach offers a genuine sense of what it's like to work together, whether in a remote or in-person setting, and allows both the company and the job seeker to assess whether they are the right fit for each other.
What personal habits or routines have helped you stay productive and keep your balance throughout this journey?
A key lesson for staying productive while balancing family life and building a company is the importance of maximizing efficiency within the limited time available. This can be achieved by staying hyper-focused during work hours and integrating work tasks into daily routines.
For Neil, this means understanding which tasks truly require sitting in front of a computer and which can be done while performing other activities.
"There are many ways to get information without just sitting in front of your computer reading articles. Podcasts, for example, are a great resource that you can listen to while walking the dog, cooking dinner, or doing other tasks. I try to compartmentalize each of my tasks by asking, 'Is this something I actually need to do while sitting at my computer, or can I do it while waiting for a dance class to finish or while conducting a reference check for a candidate?'"
By regularly evaluating how time is spent and identifying opportunities to multitask or streamline activities, operators can maintain high levels of productivity without sacrificing personal or family time.
This approach not only helps in managing responsibilities effectively but also ensures that critical tasks are accomplished, even during a busy schedule.
Interested in Atropos Health? Learn more on their website, Twitter, and LinkedIn
A note from our sponsor: Banc of California
Looking for guidance, connections, resources, opportunity? Banc of California’s banking products and services are built to support your evolving needs as you navigate the challenges of growing a successful business. As you continue to scale, our team will be with you every step of the way. Ready to take your business to the next level? Learn more:
https://bancofcal.com/