Why You Should Join SmarterDx
Copilot for clinical documentation.
If you’re interested in roles with any of the companies we’ve covered (or the unreleased ones we have in our pipeline 👀), please fill out this form to join our talent network.
Welcome to “Why You Should Join,” a monthly newsletter highlighting early-stage startups on track to becoming generational companies. On the first Monday of each month, we cut through the noise by recommending one startup based on thorough research and inside information from partnered venture firms. We go deeper than any other source to help ambitious new grads, FAANG veterans, and experienced operators find the right company to join. Sound interesting? Join the family and subscribe here:
Why You Should Join SmarterDx
(Click the link to read online).
In 2022, 97.6 million Americans lived in a federally-designated Health Professional Shortage Area — any location with less than 1 doctor for every 3,500 residents.
That’s 15 million more than in 2021, the year prior.
With 38% of physicians expressing some desire to retire in the next year and our country’s 65 or older population growing 47% by 2032, this problem is only going to get worse: by 2034, we’ll be short between 54,100 and 139,000 healthcare practitioners, up from the 17,000 we need today.
If you’re in medical school, prepare to never retire.
If you’re a doctor, prepare to see a lot more patients :)
So it’s a growing problem that instead, doctors are seeing a lot more paperwork.
In 2005, physicians spent a third of their time on administrative work and clinical documentation — summarizing a patient’s visit and activities for future reference. These days, it’s much higher - doctors today spend twice as much time here as actually meeting with their patients.
Between 2009 to 2018, the chart notes physicians write and review doubled in length. One review of 100 million notes containing 33 billion words found the average patient record had more than half the words of Shakespeare’s Hamlet. This means a physician seeing 10 patients in a day would be responsible for reading around 85 pages of single-spaced text across 691 notes — all before writing a good amount more in their own documentation afterwards.
All this contributes to physician burnout and takes away valuable time doctors could be spending with their patients.
The trouble is, this paperwork (documentation) is actually quite important.
As the canonical record of a patient’s hospital visit, clinical documentation is used to communicate across visits, healthcare providers, and insurance providers. It’s the fundamental source of a patient’s medical history — making sure it’s thorough and accurate is critical for ensuring continuity and quality of care. Small, secondary/tertiary diagnoses can become important with time and it’s important that they don’t slip through the cracks because someone was too busy to describe them.
If anything, doctors should be spending more time on documentation - about 70% of patient records contain incorrect information.
In one survey that had 22,889 patients read their own records, 21% found errors that were between “very serious” and “somewhat serious”. One patient found errors describing their medications and allergies, another with BRCA-1 saw they were labeled non-BRCA-1, and another got a PT referral for the wrong body part.
So should doctors be spending more time or less time on clinical documentation?
Well, in an ideal world, the responsibility for documentation would be completely separate from doctors.
Imagine walking into court as a defendant and discovering the judge presiding over your case was also responsible for typing out a transcript of your case and checking it for accuracy. The solution wouldn’t be to give the judge a better stenograph machine, no matter how good it is — it’d be to disconnect the responsibility from the judge.
Even when doctors are in a room with a patient, they’re only able to spend 52.9% of the time talking to or examining them: 37.0% of the time, they’re doing paperwork.
Improving clinical documentation and the process around would be one of the highest-leverage ways to can improve our healthcare system. Removing physicians as the bottleneck for every patient’s system of record with a more accurate, more automated approach to clinical documentation would improve workflows for doctors, outcomes for patients, and data integrity for hospitals.
That’s the future SmarterDx is building.
Background
Each time you visit the hospital, every test taken, scan done, and action prescribed gets summarized into a set of clinical documentation for that visit.
This documentation is prepared by your physician and is important to the healthcare process — it’s used as a summary of prior visits so future doctors looking through your file can get an understanding of your medical history without having to look through the raw charts and results themselves. This documentation is also important to the business of healthcare — it’s what eventually what gets formalized into standardized codes for billing purposes.
The challenge is, summarizing all that’s wrong/all that’s been done for a patient can be a lot — the average patient is diagnosed with 7 secondary conditions on top of their primary one and the average Medicare user on a high severity visit is diagnosed with 18. With 70 percent of doctors receiving more information than they can manage, it’s no surprise not every condition for every patient is documented very thoroughly - things slip through the cracks.
As a result, relying on doctors for record-keeping isn’t great for anyone involved:
Doctors are more stressed and have less time to spend with their patients.
Patients suffer worse outcomes — poor records hurt quality and continuity of care.
Hospitals end up having to hire Clinical Documentation Integrity (CDI) teams, who work to manually identify and correct errors in documentation.
SmarterDx has developed an AI-based solution for clinical documentation improvement.
They use the same raw inputs as physicians — over 30,000 data points from lab results to radiology reports to pathology reports to vital signs to respiratory therapy to medication ordered to physical therapy notes — to fill in gaps a physician might have missed. For instance, if a patient admitted with a blood pressure of 170/100 gets prescribed medication for hypertension but the doctor forgets to record this as a secondary condition, SmarterDx can fill in that gap while also citing the clinical evidence to support that correction. So far, SmarterDx has identified inaccuracies in 80% of the charts they review.
The accuracy required for this “second pass” is accomplished via a combination of custom features, traditional machine learning techniques, deep learning, and LLMs - it’s much more than a fine-tuned version of GPT-4. You can think of SmarterDx’s product as a highly domain-specific application of retrieval-augmented generation, a Copilot for clinical documentation that’s able to cite its sources.
The mission is to begin offloading the burden of documentation from physicians to machines, something which would create value for everyone involved:
Doctors worry less about documentation and can spend more time with patients.
Patients receive better outcomes — better records improve quality and continuity of care.
Hospitals save money and don’t have to hire as large of a CDI team.
As helpful as SmarterDx is in improving the accuracy of clinical documentation, its their impact on downstream revenue that’s really gotten hospitals excited.
Since most serious documentation errors are errors of omission (it’s more common to leave things out than to write too much in), correcting them in practice often means collecting on revenue hospitals would’ve otherwise missed (i.e. recording a missed secondary condition or a condition whose complexity was otherwise understated).
Of the errors SmarterDx discovers, roughly 15% end up affecting hospital revenue. SmarterDx then takes a cut of that, which is today how they monetize.
So far, this has been an incredibly effective business model. Each year, the average hospital using SmarterDx has recovered $800,000 in otherwise missed revenue per 100 beds, or $2 million per 10,000 discharges. Based on this, SmarterDx grew revenue 6x in the past year and are on track to 25x over a two year period. The company scaled to over $1 million in contracted revenue before raising their first round of funding last year and will end this year at a $6 million run rate. By the end of the first half of next year, they will be at $15 million.
The company was founded in 2021 by Michael Gao, M.D., Josh Geleris, M.D., and Nathan Perilo. Michael previously oversaw applied AI at New York-Presbyterian, was an Assistant Professor in Medicine and the Medical Director for Transformation at Cornell, and did his Residency in Internal Medicine at New York-Presbyterian after receiving his M.D. from the University of Michigan. After receiving his M.D. from Israel’s Technion, Josh also did his Residency at New York Presbyterian before joining Columbia’s Irving Medical Center as an Assistant Professor. Nathan was previously Director of Quant Libraries and Global Engineering at CitiBank.
Joining them are Floodgate and Flare Capital, who co-led SmarterDx’s $6 million dollar seed round in 2022.
SmarterDx has a strong product, excellent traction, and (as we’ll soon see) a rather stacked team. They’ve got a vision to transform the healthcare system’s system of record and a mission to free doctors from the burden of documentation.
But can they build a lasting business off of it all? Especially given how challenging it is to change anything in healthcare?
Well, we think they have a shot.
Let’s dive deeper into why.
The Opportunity
We’ll begin our exploration of SmarterDx like we do for all of our companies, from a first principle:
Large companies lead massive, growing markets.
Many startups with product-market fit don’t get very big (and aren’t worth joining) because they don’t meet this condition. Thus, we must establish two things:
SmarterDx operates in a massive, growing market.
SmarterDx will become a leader in that market.
Market Size
SmarterDx makes money when hospitals use their software to correct inaccuracies in clinical documentation. In other words, they operate in the market for clinical documentation integrity (CDI) solutions.
If there are 1) lots of inaccuracies in clinical documentation and 2) lots of hospitals willing to pay to address those inaccuracies, then the market for SmarterDx’s technology is large: there’s an opportunity to create a lot of value and an opportunity to capture a significant part of it.
We think there’s reason to believe both of these conditions:
There are lots of inaccuracies in medical documentation.
We’ve already discussed how despite chart notes doubling in length between 2009 and 2018, 70% of patient records still contain incorrect information. These facts might be related: electronic medical records have a serious problem with duplicate text.
One study of 104,456,653 routinely generated clinical notes containing 32,991,489,889 words found 50.1% of it was copied-and-pasted from prior documentation. Another study of 24,000 notes from 460 clinicians at UCSF found more than 80% of it was copied-and-pasted from a prior document. Both studies found the copying to be prevalent across physicians of all levels of training, and that physicians were equally likely to copy from prior documentation they wrote themselves and prior documentation written by their colleagues. All this copying makes it more likely outdated/inaccurate information is accidentally left in and updated/important information is left out.
Handwritten notes are worse: they’re often physically unreadable. Before the rise of electronic medical records, up to 25% of medication errors were related to illegible handwriting, contributing to an estimated 7,000 deaths each year.
POV: you’re giving your pharmacist your prescription. Ultimately, the root cause of all this is simple: physicians don’t have the bandwidth to make documentation as good as they’d like it to be. As we’ll see next, this is something hospitals have had to work around.
Hospitals are willing to pay to fix documentation.
Today, hospitals hire teams of Clinical Documentation Integrity (CDI) specialists who do manual second passes over documentation, also with the goal of fixing errors and recovering missed revenue. These teams have meaningful impacts on patient outcomes: 88% of hospitals confirmed documented quality improvements within 6 months of implementing a clinical documentation integrity program.
CDI as a practice is widespread today, but it only really picked up in the last two decades with the rise of easily-revisable electronic medical records. For instance, while the total number of healthcare job listings increased just 9% from 2007 to 2011, the number of listings for clinical documentation specialists increased by 223%, which made it the second fastest growing title in health informatics. By 2013, 66% of hospitals had already implemented a CDI program, with about half of those who hadn’t planning on creating one by the following year. It continues to grow today: the Bureau of Labor Statistics classifies CDI professionals under “Medical Records Specialists”, a profession which continues to grow roughly three times as fast as the average occupation.
Despite both of these facts, however, hospitals are facing increasing pressure to automate or outsource functions like CDI. There are several reasons behind this:
People are expensive. Each year, administrative expenses account for over 25% of total national health care expenditures. This is both dramatically higher than other industrialized countries and growing faster as a proportion of overall costs than actual clinical expenses.
Hospitals spend 22.6 billion each year maintaining medical records, and revenue cycle (the department CDI teams most commonly fall under) is the number one function hospital administrators are looking for outsourcing solutions in.
There aren’t enough people. Even if they aren’t looking to outsource, some 48% of revenue cycle management departments are experiencing labor deficiencies, with one in four finance leaders needing to fill over 20 positions to be fully staffed. More generally, over 57% of health systems and hospitals have over 100 open roles to fill.
Put this all together - significant inaccuracies in clinical documentation, a willingness to pay to fix them, a general pressure to outsource/automate things - and it’s clear why demand for a tool like SmarterDx has been strong.
Given broader trends like the incoming shortage of physicians or the growing adoption of AI-powered tools in healthcare, there’s reason to believe this demand will grow also.
Market Leadership
So we know hospitals are adopting solutions to improve their clinical documentation. The question now is whether SmarterDx will be their answer of choice.
As a startup, gaining market share is a function of having a great product (good reason to adopt) and great distribution (little reason not to adopt). SmarterDx has strong answers to both.
Product
Compared to the teams of clinical documentation specialists hospitals employ today, SmarterDx’s end-to-end, AI-driven approach offers several benefits:
Greater accuracy. SmarterDx identifies inaccuracies in 80% of the charts they review, providing all the clinical evidence needed to support corrections in each case. These errors range from very minor to very serious, but about 15% of them end up affecting hospital revenue.
This impressive lift in accuracy is accomplished in two ways:
Better inputs. SmarterDx uses the same clinical results physicians review as inputs for their algorithms — over 30,000 ground truth data points including lab results, radiology reports, pathology reports, vital signs, respiratory therapy, medication ordered, and physical therapy notes. Starting with raw clinical data is critical for the accuracy, interpretability, and thoroughness of SmarterDx’s outputs. If you were to train a model solely on existing documentation instead (i.e. how Github Copilot was trained on Github repositories), you’d do a lot worse because of how many inaccuracies the existing corpus contains.
Better models. SmarterDx uses a combination of custom, domain-specific features, traditional ML techniques, and deep learning approaches alongside LLMs to solve for high recall/precision while maintaining high explainability. It’s this hybrid approach that allows SmarterDx to surface clinical evidence when correcting errors — relying primarily on LLMs (even tuned ones) wouldn’t work since LLMs suffer from hallucinations and aren’t readily explainable, both unacceptable traits in medical settings.
It’s worth re-emphasizing how the corrections surfaced by SmarterDx are discovered after human CDI teams and the software tools they use have already made their own suggestions. SmarterDx operates as a final check for all clinical documentation produced.
The impressive domain expertise and engineering underpinning all this constitutes a growing technical moat for the company.
Greater affordability. SmarterDx’s automated solution is more efficient than entire teams of human CDI specialists. We were shown how at one hospital, SmarterDx was recovering more revenue than the entire team of CDI specialists they were working alongside.
Greater accessibility. SmarterDx’s “second pass” approach comes in after physicians and doesn’t directly interrupt their work. This makes their product easier to use and also begins to address the root issue - the burden of documentation-related work on physicians. This is in contrast to weaker, more traditional AI “copilot” systems that still rely on direct inputs from physicians. We dive deeper into these below.
Distribution
Compared to most health tech startups, SmarterDx has found their product much easier to distribute. Consider one hospital’s testimonial:
Why is this the case?
Well, it’s worth considering why hospitals are so difficult to distribute into normally. There are two systemic forces working against any budding healthtech company:
High adoption friction. Hospitals are often using relatively old software (read: no simple APIs) and operate without strong IT departments. The technology used isn’t really standardized across hospitals, and the processes around these technologies — data pipelines, user habits, etc. — are complex and difficult to update. Together, this means that integrating a new product (getting it to work with existing tools/data schemas) and adopting it (training teams to use it effectively) is a high-friction process, both for the hospital and the company.
Low adoption incentives. As businesses, hospitals benefit from monopolistic market positions and thus face less pressure to innovate and invest in new technologies. You’ll never see hospitals browsing for new products to try on Product Hunt. This means that to even be considered, products have to solve major pain points or seriously benefit a hospital’s bottom line.
This combination of high adoption friction and low adoption incentive means it’s harder for new products to gain traction with hospitals than with most markets.
Thankfully, SmarterDx has crafted both their product and business model to address these issues:
Reverse ETL makes their product easier to integrate from a technical perspective. Unlike competitors who often require live data, specific formats, or integration into certain electronic medical record systems (all things hospital IT departments and EMR systems aren’t great about), SmarterDx has no technical barriers to adoption — you can get up and running with a regular flat file drop. The secret is a suite of reverse ETL systems built in-house allowing SmarterDx to take data in whatever format they want before mapping it to a unified patient schema across hospitals, which their models use. From the hospital’s perspective, this reduces integration friction.
Zero upfront cost makes their product easier to adopt from a budget/procurement perspective. The unique accuracy and efficiency of SmarterDx’s system enables them to charge via a revenue split instead of a more traditional subscription model, something which offloads financial risk from hospitals. From the hospital’s perspective, this reduces procurement friction.
End-to-end automation makes the product easier to adopt from a workflow perspective. Since the product sits at the very end of the clinical documentation process, SmarterDx doesn’t require physicians to change their habits or workflows. Since it generates corrections complete with clinical evidence, it’s very easy for CDI professionals to accept the output. From the hospital’s perspective, this reduces adoption friction.
Revenue recovery as a quantifiable value proposition makes it easier for hospitals recognize the value of the product. This both aligns and increases adoption incentives.
Competitive Landscape
SmarterDx isn’t the only business leveraging AI to help hospitals improve the integrity of their documentation and revenue cycles. Now that we have an understanding of SmarterDx’s unique strengths as a product and company, let’s see how they stack up against their competitors.
Below, we consider a few categories of software products, from most competitive to least competitive.
Documentation Improvement
There are a large number of other companies also selling software to help hospitals improve their documentation. Let’s review some major players in the space.
3M is a public company that sells many different industrial, worker safety, healthcare, and consumer products. Among their offerings are Engage One and CDI Collaborate, two software tools that help CDI teams with their work. Engage One offers AI-enabled “nudges” (reminders) as doctors write their documentation, and CDI Collaborate is a cloud-based CDI workflow application.
Iodine is a unicorn that sells AwareCDI, an AI-enhanced CDI workflow platform. The primary value prop is a built-in AI system called CognitiveML which ranks files based on the likelihood they contain errors. This helps CDI teams prioritize the charts they review, improving efficiency. AwareCDI also offers a query system making it easier for CDI specialists to ping physicians for feedback.
Nuance Communications is a Microsoft-owned, formerly public company offering CDE One, another cloud-based CDI workflow application. Similar to Iodine, CDE One also leverages AI to generate prioritized worklists for CDI teams.
Unlike SmarterDx, these products are primarily focused on workflow management and collaboration, things that help CDI teams organize their work. When they do incorporate AI, it’s generally a weaker version than what SmarterDx has deployed: instead of directly offering corrections and supporting evidence like SmarterDx has deployed, they operate indirectly through prioritized lists or “nudges” telling physicians to write extra details down.
The upshot is that compared to these products, SmarterDx goes a step further and automates the end-to-end process independently of human input, only requiring confirmation at the end. This makes them far more powerful with regards to accuracy and efficiency.
Coding Automation
If you look up medical billing and AI, there’s a good chance a medical coding startup will be the first company to pop up. These businesses help automate the process where documentation is converted to the standardized codes used to bill patients. Some players here include:
Fathom Health - AI to help automate medical coding. Last raised a $46 million Series B from Alkeon Capital and Lighspeed.
CodaMatrix - AI to help automate medical coding. Last raised a $55 million Series A from SignalFire.
Nym Health - AI to help automate medical coding. Last raised a $25 million Series B from Addition.
These aren’t competitors, but it’s worth explaining how they differ from CDI companies like SmarterDx.
Documentation is what ultimately gets converted to medical codes, meaning companies like SmarterDx operate a stage before the companies above. Since you can’t code or bill for something that was never documented, these success of these companies ultimately relies on the integrity of the clinical documentation they’re converting.
Revenue Cycle Management
There are a large number of companies offering software to help hospitals manage the broader function of revenue cycle. These aren’t really competitors either, but some offer features related to revenue recovery (SmarterDx’s value proposition) as part of their platforms. We’ve included a few here for thoroughness.
Vitalware is a revenue integrity platform for hospitals. They help hospitals avoid losing revenue to poor charge capture performance (missed charges, incorrectly coded services, delays in billing) and changing reimbursement guidelines. They also help hospitals manage their chargemasters.
Cerner is another revenue cycle management platform for hospitals. They help manage scheduling, financial clearance, and registration for patients. They also automate many of the billing workflows hospitals go through.
Craneware is another revenue cycle management platform for hospitals. They help hospital administration manage pricing, revenue, cost & decision support, and chargemaster updates.
Execution
Changing how the healthcare system thinks about clinical documentation means challenging a large number of traditional healthcare workflows and software providers head-on.
That requires a strong team.
Fortunately, SmarterDx has such a team, one with a lots of experience at the intersection of AI and healthcare. Consider a sampling of their roster:
Michael Gao - Co-Founder, CEO
Michael previously oversaw applied AI at New York-Presbyterian Hospital. Before that, he was an Assistant Professor in Medicine and the Medical Director for Transformation at Cornell. He holds an M.D. from the University of Michigan.
Josh Geleris - Co-Founder, Head of Product
Josh was previously an Assistant Professor at Columbia University Irving Medical Center. He holds an MD from Israel’s Technion and did his residency at New York Presbyterian Hospital.
Nathan Perilo - Co-Founder, Director of Technology
Nathan was previously the Director of Quant Library and Global Engineering at CitiBank. Before that, he was a consultant with Infusion Development.
Wayne Grodsky - Chief Commercial Officer
Wayne was previously Senior Vice President at PointClickCare (last valued at $5 billion), Chief Growth Officer at Access TeleCare (acquired for $300 million), and Enterprise Growth Officer at ZirMed (acquired by Waystar).
Henry Su - Head of Finance and Operations
Henry was previously VP of Business Operations at Embark Trucks (IPO’d at $5 billion) and a Management Consultant at McKinsey.
Sandra Frykman - Head of Sales
Sandra was previously Vice President of Sales at Elemento Health and Vice President of Sales at Vivian Health. Before that, she was Head of Sales at Incredible Health.
Brendan White - Director of Growth Operations and Enablement
Brendan was previously Director of Growth Operations and Enablement at PointClickCare (last valued at $5 billion). Before that, he was Director of Sales Operations at inContact (acquired for $940 million).
JP Korelc - Regional Vice President of Sales
JP was previously Director of Business Development at OncoLens. Before that, he was Director of Health Systems Sales at Current Health and Regional Vice President of Sales at Access TeleCare (acquired for $300 million).
Janita Adams - Regional Vice President of Sales
Janita was previously Regional Vice President of Sales at Incredible Health (last valued at $1.6 billion). Before that, she was Regional Vice President of Sales at Access TeleCare (acquired for $300 million).
Ning Liang - Principal Software Engineer
Ning was previously Co-Founder and CTO of HealthSherpa, a $100M ARR healthtech startup. Before that, he was a Software Engineer at Twitter, Rapleaf, and Microsoft. He holds a BS in Mathematics and an MS in Physics from Yale.
Alexander Yankov - Principal Software Engineer
Alexander was previously a Staff Software Engineer at Google. Before that, he was a Founding Engineer at Alkymi, a Machine Learning Engineer at HyperScience, and Software Engineer at Goldman Sachs.
Jason Zeiber - Principal Product Designer
Jason was previously a Senior Principal Product Designer at Clari (last valued at $2.6 billion). Before that, he was a Product Architect at DealPoint (acquired by Clari), a Director of UX at Bumped, and a UX Lead at HP.
Marie Klosterman - Lead Data Scientist
Marie was previously Director of Data Science at Olive (last valued at $4 billion). Before that, she was a Data Scientist at Optum (UnitedHealth Group).
Ruth Nancy - Director of Product Management
Ruth was previously Director of Product Management at Olive (last valued at $4 billion). Before that, she was Director of Product Management at HealthVerity (last valued at $720 million).
Scott Fleming - Research Engineer
Scott was previously a Graduate Student Researcher at Stanford Medical School. Before, he was a Machine Learning Research Intern at Apple and a Research Scientist Intern at Verily. He holds a BS in MCS, an MS in CS, and a PhD in Biomedical Data Science, all from Stanford University.
Richard van Dys - Senior Software Engineer
Richard was previously a Senior Software Engineer at Truepill (last valued at $1.6 billion). Before that, he was a Senior Platform Engineer at Moovweb.
Bates Jernigan - Senior Software Engineer
Bates was previously a Senior Software Engineer at Carbon Health. Before that, he was a Software Engineering Manager at SpaceIQ and Full Stack Developer at Speed Digital.
Steven Reed - Software Engineer Steven was previously a Senior Software Engineer at Mutiny. Before that, he was a Senior Staff Software Engineer at Paypal.
The team today is 17 people and growing swiftly.
If you’ve ever wanted to work alongside the founding CTO of a $100M ARR healthtech startup, a global director for quant engineering at Citi, or a Stanford PhD in machine learning, now’s your chance.
Conclusion
American physicians will face major challenges in the coming years, but more paperwork doesn’t have to be one of them.
SmarterDx is building a future with better work for physicians, better outcomes for patients, and better data for the healthcare system.
They’re hiring.
In case you missed our previous releases, check them out here:
And to make sure you don’t miss any future ones, be sure to subscribe here:
Finally, if you’re a founder, employee, or investor with a company you think we should cover please reach out to us at ericzhou27@gmail.com and uhanif@stanford.edu - we’d love to hear about it :)
Great company!