Note: This is the first installment of a three-part series. For an overview of what this series is all about, check out the Prologue.
We see a lot of care delivery startups at Canvas: direct-to-consumer telehealth brands, value-based primary care clinics with government contracts, and everything in between. We spend time with the founders and get to know leaders across product, engineering, and clinical domains. Some teams have raised more than $50 million out of the gate while others are bootstrapped by accomplished clinicians. Still others are daring new efforts by forward-thinking incumbents who see the writing on the wall and know the old models of ambulatory medicine are not long for this world.
Our goal across these conversations is to help teams get their tech stacks right, carefully think through the software they need to build in order to bring their differentiated patient experience to market, and to understand the tooling we need to provide that will help them get there faster, safely. The pattern we see among the most seasoned and ambitious teams is a fierce pragmatism about what they know and need now versus what they aim to learn and experiment with later.
Our role as a software platform is not to supply a collection of software components to operationalize a particular care model, but to be the care modeling system, supplying the UIs, SDKs, and APIs to write, test, deploy, measure, analyze, and update care models across those components for as broad a range of workflows as possible. The scope of impact is massive. It fuels the Canvas team's dedication to our mission of improving human health through clinician-developer collaboration, and we're inspired by our flourishing band of co-travelers on the journey to better healthcare in America.
The Elements of Care Modeling
A care model is an abstract representation of a real system for intervention in human health. As health IT innovator and former deputy U.S. health IT coordinator Jacob Reider highlights often, health and care are two radically different concepts. Health is always a good thing. Care is not. Care is an intervention, the result of which lies on a spectrum between better health and a total waste of time, money, and resources.
Care modeling is the continuous process of (1) defining a care model, (2) operationalizing it by deploying technology and training people, and (3) updating it with feedback on an ongoing basis and with the goal of creating more health among the persons served. Getting started is easy. The second and third aspects of care modeling are the most crucial, but also are the most difficult. Going live and getting better is a hairy beast. Talent and patient-centered culture are preconditions for success. Beyond that, for care delivery startups with a digital-first strategy, software is the most effective means of continuously improving your care model. They are inseparable.
We've found it useful to conceive of care models in the seven dimensions that follow. They map nicely to decisions you'll need to make about technology partners (more on that later). Now, the dimensions.
Patient Sourcing and Intake
Patient Sourcing is how you'll find patients, or how they'll find you. It's tightly coupled with the Pricing and Payments dimension of care modeling (i.e., your business model). You might need to convert employees into a newly rolled-out benefit program; acquire consumers via SEO, Instagram ads, and shiny shopping mall retail locations; or sign deals with managed care plans to send seniors your way. Whatever your sourcing strategy, it will color your intake process and you'll need to design your first interaction in that context, often involving particular data requirements for payment or reporting.
Patient Intake is how you'll register new patients. Traditionally it's filling out forms, eligibility verification, and payment processing. But it can – and should – be so much more than that. This is your first impression. What if it was joyful? Magical even? This is a key moment in your patient relationship management strategy (PRM) and software. Imagine you didn't start cold on a new patient and were instead able to use a simple identifier with Health Gorilla, Particle, and other data services (maybe Zus? I would guess yes, or yes soon) to paint a more complete picture of your new patient right off the bat.
How do you maximize conversion and also triage appropriately, putting the patient's care and safety first? This is an area that deserves sharp clinical rigor in part because it can be tempting to maximize each potential patient’s economic value over their health and safety.
Ongoing Interaction Modes and Utilization Policies
Interaction Modes, sometimes called “channels” of communication, can be grouped into synchronous and asynchronous types. Synchronous modes include the traditional office visit, phone calls, video calls, and nominally live chat, though in practice it seems to slip into the asynchronous category pretty fast. Asynchronous modes include messaging (whether that's texting with patient consent, WhatsApp, Signal, proprietary mobile apps, etc.) and email.
Many commentators seem to like the term "omnichannel" but I find it a touch misleading. It is unwise to support every possible mode of interaction; In some cases doing so carries compliance risk. But the spirit of the term is right.
A more accurate way to think about Interaction Modes in your care model is to lay out the set of modes you want to support, and then think about how you will link them for an intermodal experience. Caregiving through one mode should carry context supplied through any other mode. That's the cost of supporting multiple modes — it's not enough to offer them, you have to connect them lest your patients balk at the waste.
A technique for managing the cost of many Interaction Modes is setting Utilization Policies (not to be confused with utilization management in the managed care context). While they appear simple on the surface, Utilization Policies are an endlessly complex area of operations management and have a major impact on the Care Team Composition and Sourcing dimension of care modeling.
Utilization Policies answer questions such as: What synchronous interaction modes will be available on demand, if any? In what states is a clinician licensed and with which payers are they credentialed? Will your caregivers be staffed through Wheel or another agency or will they be employed by your organization? How far ahead of time will you allow capacity for synchronous interactions to be reserved? Does that change depending on the reason for visit? Do new patients have the same access to different interaction modes as established patients? Are specific caregivers assigned to patients or do patients have control?
As you can probably see, you won't be able to fully determine all Utilization Policies in advance of operations, and you shouldn't try. Focus on the 80/20 — get the major use cases well-defined, then elaborate as exceptions occur.
Diagnostic Range and Inputs
Diagnostic Range describes the breadth of diagnoses your clinical practice manages. Your target patient population is the primary determinant of complexity in this dimension. On one end of the spectrum, specialty telehealth clinics are often relatively focused, e.g., a handful of migraine diagnosis codes along with others to track notable comorbidities. On the other end, full-scope hybrid primary care and urgent care models might manage thousands of diagnoses.
Depending on your Diagnostic Range, you may want or need to include Inputs from internet-connected remote sensing devices like continuous glucose monitors if diabetes is in range, pressure cuffs for hypertension, weight scales for heart failure, fall detection from vendors like SafelyYou for dementia, potentially even therapeutic monitoring like Propeller Health for asthma and COPD. All of these streaming Inputs have big implications for your care model and enabling technology.
For some care delivery startups, like Hinge with its sensor technology for MSK, this dimension is the bedrock of their competitive advantage. New devices and technologies (digital biomarkers?) will enter the picture with increasing frequency, each giving rise to big new opportunities in digital care delivery.
Furthermore, each condition managed is likely to add complexity to the Content and Automation dimension of your care model as well, especially over time as diagnostic guidelines evolve and content for guiding inquiries and assessments must be maintained. Do you also need educational materials for your patients? How many languages and what form factors? Will you use a service like Healthwise or UpToDate?
Evolving Diagnostic Range and Inputs is a very natural dimension of progression for care models. The best practice is to move methodically and gain consensus from clinical leadership on the maturity of the model for your current range before considering expansion.
Scope of Interventions and Safety Framework
The Scope of Interventions dimension encompasses both diagnostic interventions (labs, imaging, consults) and therapeutic interventions (all forms of medications, instructions and exercises, referrals and enrollments). As your Scope of Interventions broadens - and your care model gets more complex - you’ll need to apply more rigor in your safety framework. Safety is beyond the scope of our write-up, so I'll point you to a great reference from IHI: A Framework for Safe, Reliable, and Effective Care. Visibility into safety issues, learning from them, and fostering a culture of safety is not optional. Safety should drive your technology decisions especially around visibility and feedback/learning when things go sideways.
Virta is a great example of a company competing and winning on its Scope of Intervention. Traditional interventions for diabetes, as an example, include lifestyle advice but rely heavily on pharmaceutical therapies. Virta entered the market in the mid 2010s focused on lifestyle interventions, specifically diet, rather than medication therapy. They developed purpose-built technology to maximize the effect of nutrition therapy and minimize the need for ongoing medications. Coupled with best-in-class clinical rigor and safety as well as continuous innovation in other care model dimensions like Interaction Modes and Payments, Virta is a resounding success.
Other successful new entrants like Ro and hims & hers have relatively narrow Scopes of Intervention for men’s and women’s health. Instead of lifestyle interventions they have historically focused on medication therapies and have optimized their teams and technology specifically to produce those interventions.
Every type of intervention requires a value chain to produce it and monitor it safely. You can get a pretty good sense of your software and partnership needs by mapping your interventions to the tools that will help you deliver them. For example Truepill, Sitka, and Ribbon can improve the quality of service you offer for medication and referral interventions, respectively.
Care Team Composition and Sourcing
With clarity on your Diagnostic Range and Scope of Interventions, you can then explore different configurations for your care team. The complexity here is substantial. Different roles have different licensing requirements, and different levels of experience fit different caregiver organizations. One organization may need a geriatric nurse practitioner licensed in Florida, with 10 years’ experience. Another may look for a recently credentialed California-licensed marriage and family therapist.
You'll be balancing license levels, experience, cost, capacity, and compatibility with your payment (and compensation) models, not to mention the softer side of cultural fit. And you might want help credentialing from a partner like Medallion. Or you won't and you'll trust a third party sourcing partner and use temporary clinicians, or a combination of permanent core team and flexible capacity through a third party. There are diverse and strong opinions on this topic.
Regardless of the approach you take, you'll need to balance productivity with license risk. It's good for quality and cost if care team members are practicing at the "top of their license" as folks like to say. But demand has so outstripped supply that new entrant and incumbent provider orgs alike are asking their clinicians to take on more license risk, potentially pushing the limits of productivity and delegation to the brink of safety norms in some cases. Clinicians have deep-seated alarm bells for this, and also a battle-tested intuition on how to triage safely. But the systems they operate in — the companies they work for — can either enhance that intuition or inhibit it with their care modeling choices.
That brings us to the next dimension of care modeling, which asks you to think through the role Content and Automation will play in your future.
Content and Automation
In the context of care modeling, content falls into two buckets: patient-facing content which includes questionnaires, instructions, and educational materials; and care team-facing content including diagnostic guidelines, micro-templates, clinical quality measures (CQMs), and treatment standards of care. Some of this content is relatively static, such as physical exam templates, templates for reviews of systems (ROS), guided histories of present illness (HPIs), and stock language around procedures. Other content is highly dynamic, difficult to build and manage, but can safely unlock automation in the clinical setting.
Advancing Content and Automation might be the single most important reason to adopt the perspective that care modeling is part and parcel of care delivery. A system that enables you to power automation within your care model will drive leading operational cost and quality metrics. Without a well-defined care model that either is or is embedded in the system for care delivery, there is no hope of increasing access to care or reducing variance and bias at the level we need in our communities.
Healthcare Pricing and Payments
If there's one thing we've learned over the last 40 years, it's that incentives matter. How much care is delivered, how it's delivered, who delivers it, and where it's delivered are all massively influenced by how it's paid for and by whom. In general, you have four options:
- D2C: direct-to-consumer
- SIE: self-insured employers
- FFS: fee-for-service reimbursement with commercial or government payers
- VBC: value-based contracting with commercial or government payers
Many folks start with D2C and go from there. Setup costs are lower, but patient acquisition is not easy. Success in building a patient base can give you more power in FFS contracting or help you meet patient panel size thresholds sooner in VBC. (Aside: you often need > 1k attributed lives for VBC in Medicare Advantage and > 10k attributed lives for VBC in Commercial.) That power-accumulation sequence is the notion behind the B2C2B playbook some entrepreneurs are executing.
The care modeling and software implications of evolving business models are substantial. Working with Commercial insurers is a massive undertaking relative to D2C and even SIE. You'll need to handle credentialing, eligibility, claims, denials and appeals, remittances, and quality measurement and reporting. Dealing with Medicare and Medicaid is that times ten. So if your business plan takes you into the commercial payer or government territory, pay special attention to future requirements and avoid torching too much capital on tooling that can't grow with you.
Part II: Strategies for Enabling Software
In the next installment of this series, we discuss a framework to help you think through which aspects of the care model you need to build yourself and which dimensions can be accelerated through technology. It is a high-level cheat sheet to help you plan infrastructure investments and select your technology partners.
Part II: Strategies for Enabling Software ->