Written by Mark Goad, OMERS Ventures
Technology has long been prescribed as a remedy for healthcare’s ailments.
Programmers in the 80s built expert systems (early AI) like the CADUCEUS to automate disease diagnosis.
IT providers in the 90s proposed networked health information systems that would increase consistency in care and generate robust data to better manage & predict costs.
Startups in the early 2000s claimed the internet was a revolution for healthcare and was going to provide unlimited information, education, virtual interactions with your doc, and second opinions from experts around the world.
In the 2010s, the progress & cost reduction of genome sequencing held promises of personalized medicine that are still developing today.
It is the year 2020 and most of our engagement with healthcare systems look similar to how they did 20 years ago. The cost of delivering healthcare (particularly in the US) has risen faster than inflation for decades but health outcomes for patients are mostly better.
Change in healthcare takes time. For every example of progress, we can point to over-hyped claims that failed to deliver. This assessment of ‘failure’ vs. ‘progress’ depends on the time horizon applied.
Today, we stand at the dawn of a new era of healthcare innovation. The smartphone — a device some of us (figuratively) cannot live without — as a new platform to deliver care. While assessing this new crop of innovative startups, it is important to balance historical adoption rates of technology with our enthusiasm for innovation.
You should read this article if you are interested in understanding:
- what Digital Therapeutics are & the current state of the market;
- a VC’s view of DTx risk & upside;
- DTx commercialization, stakeholders & strategies;
- what we think the DTx work will look like in 5 years.
If you are building in this space — I’d love to talk.
1. Overview of PDTx
Prescription Digital Therapeutics (PDTx’s) are medical interventions which leverage software as the ‘main ingredient’ to prevent, manage or treat medical disorders or disease.
While there are many variants, this article will mainly discuss standalone PDTx apps — as opposed to apps built to support a connected health device (i.e. heart rate sensor, etc.) or associated pharmacological intervention (i.e. drug companion).
PDTx’s are apps that treat diseases just as well as drugs.
For us, PDTx’s are different from other ‘digital therapeutics’ or ‘disease management apps’ as PDTx’s are accredited by governing authorities to be distributed by primary care providers (i.e. approved by the FDA in the U.S. and prescribed by your doctor). Whereas digital therapeutics or disease management apps may have peer reviewed studies and be reimbursable by payers, but lack the clinical rigour imposed by regulatory scrutiny (more on this below).
Most PDTx’s focus on indications with a strong behavioural component — using the smartphone’s components that were originally designed for communication & entertainment to capture, analyze, and transmit medical interventions.
Because the input vehicle is a smartphone, we can breakdown the common PDTx application components along the specifications of modern smartphones:
Input Mechanisms (for collecting patient data)
- Screen: often used to collect patient’s behavioural or cognitive status by administering tests/quizes (mood, cognition, memory, familial recognition, concentration);
- Accelerometer: often used for measures of dexterity or steadiness (stretching, range of motion tests); and,
- Microphone: often used for audio analysis (voice quality, shakiness)
Output Mechanisms (for delivering care)
- Screen: for delivering content (blogs, tips, FAQs, videos);
- Chat (Voice/Text/Video): for patient communications (professional coaching, peer-to-peer support); and,
- Notifications: often used for adherence prompts (inactivity, sleep, diet).
Using these input and output mechanisms, PDTx’s can be used to replace or augment existing treatments like counselling, devices, or drugs.
The industry is rushing to get their apps approved by the FDA (and similar regulatory bodies, globally) via the same rigorous clinical assessments that are required for Class II medical devices [more info below].
The most common indications treated by public FDA PDTx applicants are anxiety, depression, insomnia, pain management (typically via iCBT) and cardio metabolic conditions (typically via dieting/coaching) [see market map below].
The Most Advanced PDTx Companies
All public info from company websites; funding data from PitchBook:
Pear Therapeutics, developed the first FDA approved, software-only PDTx reSET-O (iCBT for opioid outpatients); and reSET for substance abuse disorder (FDA approval pending) with Sandoz, a division of the +$200Bn market-cap pharma giant Novartis. Sandoz and Pear recently cancelled their commercial agreement (more info below) [$139Mn funding].
Akili Interactive, which develops apps and games which treat ADHD (FDA approval pending), autism spectrum disorder, major depressive disorder, multiple sclerosis, and is currently developing products to treat traumatic brain injury, ICU delirium, and lupus. [$134Mn funding].
Gaia Therapeutics, develops PDTx’s for major depressive disorder (RCT study), borderline personality disorder, generalized anxiety disorder, panic disorder, social phobia, fatigue in multiple sclerosis, epilepsy and others. [Unknown — German company].
Click Therapeutics, develops PDTx’s for major depressive disorder, insomnia, acute coronary syndrome, chronic pain, and migraines. [$17.5Mn funding — with $300Mn co-dev deal with Otsuka Pharmaceuticals in Japan].
Better Therapeutics, develops PDTx’s for type two diabetes, hypertension, and hyperlipidemia. [$23.8Mn funding].
Natural Cycles, develops an FDA-approved contraception app and associated basal body temperature thermometer to assess risk of pregnancy. [$38Mn funding].
2. Why is PDTx Venture-backable?
Four big reasons why we (a software-focused venture fund) like the idea of PDTx:
a) Software can influence behavior.
We know that software influences behavior (shout-out Zuck) and most diseases have a meaningful behavioral component.
b) Bits over atoms.
We are familiar with the near-zero-marginal-cost-of-reproduction economics of software businesses and have seen how scalable these businesses are relative to non-software competitors (in this case — pharmaceuticals). We are also comfortable with the archetypal software startup journey with associated fundraising milestones, metrics, and operating strategies.
Further, as your app’s ML and DL algorithms are deployed to more devices and patients, we have seen accumulating accuracy that these models produce over time, meaning your solution could get better over time — something drugs cannot possibly claim (more on the FDA’s treatment of unauditable algorithms below).
c) “Healthcare is a big business”
VC platitudes be damned, Americans spend a lot of money on prescription drugs on an absolute and per capita basis, making it a great target for disruption via software.
According to the Centers for Medicare and Medicaid Services, the U.S. spent $360.3 billion on prescription drugs in 2019. If a fraction of that spend can be diverted to high-margin software applications that actually work — there will be strong investor interest.
d) Alternatives to PDTx (i.e. drugs) are expensive
Finally, the cost of drug development has climbed substantially as the ‘low-hanging-fruit’ of pharma has been captured over the last few decades and R&D and regulatory costs continue to rise.
According to a study by Tufts Center for the Study of Drug Development (CSDD) published in the Journal of Health Economics, the average cost of developing a drug that gains marketing approval by the FDA reached $2.6 billion, up from $802 million in 2003 (CAGR of 9.5% over 13 years).
3. A framework for measuring PDTx progress
After talking to +50 teams building in the space, we have traced the evolution PDTx startups follow, with associated funding milestones. Note: there are certainly companies that buck the outline below — I am stating the general trends we see — not the mandatory or required progression.
Level 1: Validation
Financing Stage: Seed
The first wrung on the journey of a PDTx startup is building an MVP of your app and testing it with a small group of users to prove it has an impact on a specific indication. Most startups we engage with have a product and have run studies with small (10–50) groups of users with positive results.
These teams are usually out actively fundraising for a sizable ($10-$25Mn) Series A with this evidence to convince investors to fund the FDA clinical trial process.
The needle to thread: teams have limited money to run massive, complex studies — but investors need solid clinical results to be enticed to take on FDA/regulatory risk to fund the business at this stage.
Essentially, you are asking investors to extrapolate the results of exploratory trials to the FDA’s rigorous design & control thresholds.
The best teams we see at this stage have designed their early studies with robust controls that are at least analogous to the studies used by more mature FDA applicants.
Plenty of academic literature exists to help us classify and rank studies based on size, design, quality of evidence, and risk of bias.
The Hierarchy of Evidence Pyramid, ResearchGate
We have seen a few teams get a massive head-start here by leveraging relationships with existing healthcare providers to grow their validation patient population at a significantly reduced cost. One team was working to develop a PDTx for depression and was already engaged with a group of psychotherapy clinics via a chat and patient management application to get detailed data on patients and outcomes to optimize their solution. This is an incredibly attractive way for teams to de-risk this stage of product development and is an early signal you are able to navigate the complex ecosystem of healthcare stakeholders.
Assuming studies are designed with sufficient quality of evidence and controls for bias, investors are next looking to the delta in cost & performance relative to the existing standard of care for that particular indication. Whether or not you believe the alternative needs to be 10x better, relative performance will matter to patients when they are presented with more established alternatives. Establishing clear performance improvements relative to existing standards of care is not an easy task — but the best teams we see have very clear data that shows the efficacy of their solution.
Level 2: Efficacy
Financing Stage: Series A
Typically after the Series A is raised, these startups apply for either a De Novo or 510(k) approval from the FDA.
The De Novo program is for new devices (apps) that have no substantially equivalent predecessor (as determined by this test) while 510(k) clearance is for apps that have sufficient similarity to existing solutions.
Typically, we see PDTx startups first apply for the more rigorous De Novo program for their first solution/indication and leverage their technology platform to build subsequent solutions in adjacent indications which they can route via the less strenuous 510(k) path.
Click Therapeutics, one of the leading PDTx startups, detailed their microservices architecture (from their public website) which categorizes services into behavioral and medical to show how they can build multiple apps for multiple indications by routing new content onto existing, modular service components.
modified from Click’s website, June 2019.
Machine Learning & Deep Learning
A theme dominating discussions with PDTx’s is the FDA’s view on unauditable ML/DL algorithms.
Important concepts include (1) interpretability: being able to discern what the algorithm is doing, without necessarily knowing why, and (2) explainability, which is being able to literally explain what is happening and why an algorithm is making a specific recommendation.
This matters to the FDA, because depending on the type of ML/DL algorithm deployed, developers may not be able to fully explain why their app is making a specific recommendation.
Traditional programming involves the application of rules to data to provide an answer. For PDTx, that would looks like this:
- Patient provides data (i.e. I’m feeling depressed today, subsequent details of severity, duration of symptoms, etc.)
- Application applies rules (i.e. based on measures of severity & duration — either take deep breaths, go for a walk, meditate, or if more concerning, seek medical help).
Hubert Wang, Introduction to TensorFlow for AI, ML, and DL
In ML/DL, specific outcomes and the preceding subject data are supplied and the algorithm builds the rules (whether or not fully-explainable) that connect the answers to the underlying data. Using a simple regression or random-forest analysis — it is simple to attribute the relationships between the data and explain the app’s recommendation. However, as researchers continue to deploy more complex, multi-level neural networks — these connections become increasingly obfuscated.
The FDA has generally communicated (via PDTx conferences and in conversations with startups) that it will govern ML/DL algorithms using the existing principals of efficacy and immaterial changes. Essentially, if the models produce effective recommendations at scale in well-controlled studies, they will continue to be approved. Regarding the re-enforced, self-learning (i.e. changing nature) of ML/DL algorithms, regulators have communicated as long as the changes are gradual and within reasonable windows of efficacy that the solution was initially approved, they will remain supportive.
Aglie Development as a Precursor
The concept of software updates and patches are not new to the FDA, which in 2013 worked with software engineering teams during the industry’s macro shift from waterfall to agile development methodologies, allowing teams to use short-term development sprints that add incremental functionality and performance improvements to their FDA regulated software without requiring review every time new code was pushed to production and patients.
We believe the FDA will continue to manage ML/DL based applications under these guiding principals.
Level 3: Integration
Financing Stage: Series B
The next milestone we see are startups who have completed, or are nearly finished, their FDA pre-market approvals and are beginning to test go-to-market strategies in preparation for launch. These teams are usually raising their Series B financing, in the range of $20–$50Mn.
The needle to thread: convincing payers this PDTx is worthy, convincing providers it is effective, convincing patients it will work for them.
Go-to-market strategies for PDTx vary widely, with different strategies for different disease indications and provider/payer/patient paradigms.
Integration with Providers
Providers include anyone who provides care: doctor, chiropractor, clinical psychologist, optometrist, nurse practitioner, dentist, clinical social worker, etc.
We have talked to a lot of providers about PDTx — from primary care physicians, psychologists, and dentists. Their concerns are persistent across focus areas and can be bucketed by constraints on time & concerns over efficacy:
- I have a lot to do (their schedules are insane) and I didn’t ask to read your BDR’s crappy outreach email.
- I don’t have time to learn about new apps.
- Even though it’s enabled by your app, how am I supposed to check-in with my patients when they’re not in an appointment? What is a reasonable cadence? What reminders will you provide?
- No EMR? No go. Some physicians draw the line if they have to log in to a separate web-app to monitor or report on performance.
- I don’t trust your 10-person, VC-funded clinical trial — show me something more robust.
- You told me EMR’s (aka technology) were going to change the way I treated patients, they haven’t delivered on any of that value I was promised — why should I trust you now.
- I don’t trust unauditable deep learning algorithms.
For us, integrating with providers represent an important hurdle for any PDTx startup. Leading companies at this stage have practicing representatives from their domain advising on this engagement strategy. Repeatedly in these conversations, and similar to other industries, the importance of winning over key opinion leaders in specific domains is paramount so they can validate the solution for other providers.
Integration with Payers
Payers pay for healthcare. In the U.S., most Americans have insurance (91%) and are covered by their employers (67%).
Payers are employing ‘measured optimism’ regarding PDTx. They see the cost of delivering healthcare first-hand that is forcing them to charge more while delivering the same or less value. PDTx looks like an amazing way to meet their members where they already spend time (on their phones) and deliver meaningful no-drug interventions.
However, most PDTx preventative solutions are just too nascent to truly prove they will deflect or reduce future health care spending (i.e. panic attacks leading to emergency room visits) and most treatment solutions suffer from the same lack of clinical data to establish they can manage harmful diseases, especially in complex co-morbidity scenarios which drive the majority of costs for payers.
Efficacy vs. Effectiveness
Payers play the important role of selecting which PDTx’s they will reimburse for their members. From payers, we hear the selection criteria will involve the important distinction between efficacy and effectiveness. Efficacy refers to the performance of an intervention under ideal conditions (i.e. in a controlled trial environment) while effectiveness describes performance under normal circumstances of use (i.e. real life). Many payers we spoke with expressed skepticism of claims being made by PDTx’s from their initial studies and talked about how they may require additional clinical trials to address their concerns before supporting a PDTx.
Adherence (=? Retention)
The biggest gap between efficacy and effectiveness we see in PDTx’s today is that retention mirrors what we see in some consumer applications: a massive cliff between signup and week 1 and a steady, unrelenting decline from there. More often we see teams try to hide poor engagement data when we would rather see it benchmarked to the existing standard of care. Here’s why: drug adherence is empirically awful. From the CDC:
In the United States, 3.8 billion prescriptions are written annually. Approximately one in five new prescriptions are never filled, and among those filled, approximately 50% are taken incorrectly, particularly with regard to timing, dosage, frequency, and duration. Whereas rates of nonadherence across the United States have remained relatively stable, direct health care costs associated with nonadherence have grown to approximately $100 — $300 billion of U.S. health care dollars spent annually. From: Improving Medication Adherence for Chronic Disease Management — Innovations and Opportunities (2017).
A 20% drop from signup to week 1!! 50% adherence thereafter! The reasons for nonadherence are critical in this evaluation:
CDC: Improving Medication Adherence for Chronic Disease Management
The top three reasons SHOULD be addressable given the inherent properties of PDTx: forgot, ran out, left at home (a big reason why we think Providers/Payers are excited — but haven’t seen enough data to support this). Even when we see strong adherence, we often have this lingering unease as to whether the Hawthorne effect (where people behave differently when they know they are being watched) is present in small scale trials with people more regularly touching base with clinicians thereby resulting in artificially high engagement. Addressing these concerns head on with study design constrains can help assuage these investor concerns.
A corroborating datapoint for benchmarking adherence is digital health company Livongo, which IPO’d in July, and publishes enrollment data for its corporate and health system customers. Livongo is best known for its diabetes management application which helps patients manage their blood sugar levels with a bluetooth-enabled blood glucose meter and associated app that provides personalized recommendations (I refuse to say ‘health-nudges’) for diet, exercise, and related behavioural changes. In its S1, Livongo marked its 12-month enrollment for new customers at 34% of total recruitable patients. Most PDTx startups we speak to speculate that this rate is relatively low because Livongo primarily acquired its 148,000 members through their employers — who have a weaker relationship when it comes to healthcare than a primary care physician who would prescribe a PDTx. In most cases, it is too early to establish what a ‘good’ enrollment and adherence rate is between PDTx’s and we prefer to use existing drug adherence as the baseline.
Adherence is a key value proposition for DTx and PDTx startups. Providers, patients, and investors are excited about the smartphone as a delivery vehicle for healthcare given we can’t seem to live without our devices. However, early data from the industry suggests that this does not necessarily translate into continued usage and adherence — making this point a key aspect of our early diligence efforts.
Marketplace Approach with many PDTx’s
However, once a PDTx solution can pass this effectiveness bar (which varies by payer and indication), payers have publicly stated that their intention is to have many solutions for each indication. Most payers have created or are in the process of building directories of approved applications that will allow members to self-select their desired intervention, as advised by their provider.
The most advanced public offering is Blue Shield of California’s Wellvolution platform. We expect most payers to employ a similar ‘centralized’ distribution strategy because if the app is accessed via their platform, it makes tracking engagement, usage, and subsequent billing more streamlined for the payer but puts the burden of integration on the PDTx.
We also have seen pharmacy benefits managers begin to curate formularies (lists of medicine) for digital products that payers can select for their members. Both CVS Caremark and Express Scripts have announced they will offer lists of pre-vetted solutions that have been screened for “safety, effectiveness and usability”.
Level 4: Efficiency
Financing Stage: Series C
Now that a PDTx has been approved by the FDA, integrated with providers & payers — the go-to-market race begins. We remain concerned that some teams forecast this to be an ‘easy’ step and envision monopolistic markets in which they are the only PDTx for their indication — with the specific rationale that the FDA approval process is a definitive and enduring moat… but is FDA approval a competitive moat?
The FDA and Drug Patents
Hear me out. If we believe that software can influence behaviour and most diseases have a key behavioural element, we should expect to see a significant increase in the number of PDTx startups over the next few years. Further, recession predictions aside, in a world awash in (venture) capital, there should be sufficient funding for most PDTx teams to pursue FDA approval. The FDA and payers have both indicated their desire to have multiple solutions for each indication — allowing the consumer to select the intervention that best suits their unique needs. While we expect the first apps through the gate will have a relative advantage compared to the 350,0000 non-FDA-approved, health & fitness apps, we expect most widespread, behaviour-based indications to have multiple PDTx and disease management alternatives — and we are already seeing this with multiple competitors in behavioural health, diabetes, and hypertension.
In pharmaceuticals, economic moats are built by patents to guard market access before generic versions of the drug are sold at a far lower cost because competing pharmaceutical companies did not bear the cost of discovery. While the true length (and value) of drug patents depends on who you ask (industry watchdogs say too long; pharmaceutical companies say too short) a similar regime does not appear to exist for software businesses.
We generally do not get excited by software patents as moats because we often see multiple approaches that solve the same issue at the application layer (very different for infrastructure and middleware) — and filing a software patent complaint is usually a long and expensive process with uncertain outcomes. Applying this lens to PDTx, we see multiple apps treating the same indication which would not necessarily infringe on each other’s patent protections as there are many ways to influence an individual’s behaviour and treat a disease.
Fundamentally, we see patenting PDTx software as a much weaker economic moat than patenting molecular composition.
We predict that success for a PDTx startup will at first be correlated to the substantive, proven efficiency of the intervention, such that it is driven by payer and provider buy-in and pushed down to consumers (patients). We expect long-term market success to be more pull-through, relying on a combination of distribution, positioning, and product efficacy — more similar to the success formula for consumer apps than pharmaceuticals (see section #5 below).
Partnerships with Pharma
Three of the leading PDTx’s have announced partnerships with pharmaceutical companies in an apparent exchange of digital-learning for commercialization expertise.
Pear Therapeutics, the developer of the first-ever software-only FDA approved PDTx (December 2018) reSET-O, a prescription iCBT app for opioid drug addiction, partnered with — and subsequently cancelled — their go-to-market partnership with Sandoz, a division of Swiss pharma giant Novartis.
Click Therapeutics, the developer of the behavioural & medical microservices platform for PDTx’s outlined above, partnered with Japan’s Otsuka Pharmaceutical to co-develop a new PDTx to treat depression with up to $300Mn in payments available if commercial (GTM) milestones are achieved.
Akili Interactive, the developer of digital medicines (some are video games) for cognitive diseases has partnered with Japan’s Shionogi pharmaceutical (best known for developing Crestor), who has purchased the exclusive rights to the clinical development, sales and marketing of Akili’s ADHD and autism treatment apps in Japan and Taiwan.
We generally see pharmaceutical companies engage reticently with PDTx startups. On one hand — the transition of value from atoms to bits has decimated many other industries and most big pharmaceutical companies have venture arms and corporate development groups to flag incoming disruption to their drug portfolios. Most of these groups are taking a balanced approach to build vs. partner vs. buy — building what is closest to their core, or has the best chance of extending their marquee patents which generate a disproportionate share of their revenue and earnings. When speaking to one pharma executive — I posited that pharma hiring software engineers is similar to the automotive industry hiring for these roles: “not the engineer they usually go for”. There is sufficient operating budget to build these teams — but definitely a deficit in organizational design and incentives that will levy a ‘strategy tax’ on these efforts. Most pharmas appear to be happy to monitor competing PDTx solutions while only the three above have jumped in to the market with partnerships. Of note, we have not seen any big acquisitions/investments (like VW Group’s $2.6Bn investment into Argo) … yet.
Level 5: Engagement
Financing Stage: Series D+
The final level of PDTx startup we foresee, but have not seen in the market today, is one that goes beyond transactional GTM partnerships with existing pharmaceutical companies and engages patients with product, acquisition, and retention techniques we see in leading consumer, social and gaming applications. I generally detest the word ‘gamification’ — in my view it has been co-opted by tired enterprise PMs seeking better engagement by draping game mechanics over existing, crappy applications.
However, game mechanics, social mechanics, viral loops, and feedback cycles are not optimizations done after the fact, but built into the heart of the product experience from day 1. While some VR-focused PDTx’s have hired some game designers (typically building virtual worlds on engines like Unity) we see far fewer companies incorporating modern product management principals into their applications.
From one side — it makes sense. Delightful product experiences will not appease the FDA.
But if you agree with our thesis that multiple PDTx’s will serve the same indication, available from the same providers, payers, and regulators — then you should be similarly concerned that the industry is deferring investment (or at least mindshare) on what we view as an essential element of PDTx long-term differentiation. As we have seen from the legacy enterprise software providers — it is incredibly hard to add these product elements after the fact.
Want to chat?
PDTx’s represent a tremendous opportunity to deliver medical interventions at a global scale with nominal margin cost. The VC-backed startups building in the space employ a wide range of regulatory, product development, and go-to-market strategies — and the playbook looks quite different from a traditional enterprise SaaS company.
OMERS Ventures leads Series A-C investments, usually with a USD$10-$30Mn check from our Toronto, London, & Bay Area offices. Existing investments in the future of health include musculoskeletal (MSK) health pre-hab and rehab solution PeerWell and League, which is building a new OS for employee health benefits with major employers across North America.
We’re looking for teams that can outperform the general path I’ve outlined above and are typically on stage #3(clearing regulatory risk), have some early data of their GTM integration with payers and providers, and share our thesis of a multi-competitor world so are building incredibly engaging PDTx apps.
If you are intrigued/enraged by this article — my email is firstname.lastname@example.org