Why health IT has under-delivered, and how we can recover the promise of the 2010s.
One could be forgiven for believing the 2010s should have been a golden age for health IT. Between employers fed up with ballooning costs, pressures from the new ACA plans, and the implementation of MACRA, the environment seemed ripe for sweeping transformation in healthcare delivery.
But our health systems installed Epic and stopped there.  It is not clear that care quality is much better in 2020 than it was in 2010. And far from doing more with less, our providers actually have been doing less with more. According to McKinsey, multifactor productivity in healthcare delivery “decreased by about 420 bps per annum” through most of the 2010s. 
What happened to the Smart Enterprise wave that was promised in healthcare? What happened to precision medicine and data-driven decision support? Where are our AI pathologists?
Reasons behind the stagnation
There is not just one simple reason behind health IT’s relative stagnation. Rather, the confluence of dozens have held back innovation.
EHR software from most major vendors is notoriously challenging to extend. Limited interoperability means new capabilities are challenging or impossible to build. And once installed, replatforming is a step too costly for most health systems to consider.
Security and Privacy Requirements
HIPAA compliance imposes strict requirements on data handling and data storage. Some buyers require HITRUST certification. This adds complexity and cost to development and presents barriers for newcomers to the industry.
Extended Sales Cycles
Health systems frequently lack strong central buyers, requiring the alignment of multiple stakeholder groups to advance even a pilot. In line with security and privacy requirements above any sale will also require lengthy security questionnaires and audit processes after business approval. What might take 6 months in pharmaceuticals, manufacturing, and other industries likely requires 12 to 18 months in healthcare.
Changing workflow for any role in any industry is challenging — but making changes to how a clinician practices medicine can be particularly daunting. Liability concerns can also make practitioners skeptical of anything that replaces or influences their own judgement.
Limitations From Billing Codes
The ontology of billing codes can limit innovation by constraining the way care is delivered. Payers like CMS reimburse providers for an set of specific services, each of which has a distinct billing code. But if a code for a specific service does not yet exist, billing can be made more complicated.For instance, only in 2019 did Medicare begin reimbursing providers for remote review of data from connected devices.  Meanwhile, reimbursement for services delivered by autonomous AI is an ongoing discussion. 
In some cases incentives for providers to harness new technology have been limited. For example, reimbursement can be reduced for a service provided by a nurse, tech, or community health worker instead of a physician. On the flip side, payers do not necessarily reward quality in the same way that top service providers (lawyers, architects, barbers) can earn a premium. Medicare pays an essentially flat rate for a service code, full stop.
In some cases the FDA must approve software used in care delivery, a barrier that is compounded by review processes that are still incipient. IDx’s autonomous diabetic retinopathy screen was the first such approval in 2018,  Nines AI’s automated review of CT images is another example. These are still early days and FDA is still developing institutional know-how.
In some cases, health systems’ monopolistic behavior may also have played a role,  as well as scope of practice laws. These explanations can become political and contentious.
We only know with certainty the outcome: healthcare providers haven’t benefited from information technology nearly as much as they could have. And neither technical capability nor investor appetite is among the reasons why.
The rise of tech-enabled services
Hamstrung working with incumbents, entrepreneurs began building technology and using it themselves. Livongo and One Medical were some of the first movers. Oscar also fits this mold, when applied to payers instead of providers. 
Starting de novo and owning delivery helped these companies circumvent the challenges mentioned above. No integration with closed legacy systems, no transmission of sensitive healthcare data, no sales cycle headaches. Alternative approaches to financing helped get around some limitations of claims. Livongo offers a subscription model to employers, and One Medical asks for a membership fee on top of fee for service reimbursement to reward the premium services it offers patients.
Two successful IPOs later,  the door is open for more technology-enabled health services. The proliferation of at-risk financing models is currently driving significant investment in primary care, with companies like Cityblock, Galileo, and Firefly Health on the vanguard. Mindstrong is re-imagining behavioral health. Cricket Health and Strive are working on renal care, and several players such as Physera, Kaia, and Luna are attacking the musculoskeletal space.
The key for many of these businesses has been to marry top engineering talent with medical expertise and experience to create new models of care facilitated by software. For instance, Cityblock was born of industry veterans from Commonwealth Care Alliance joining forces with engineers from Google. Without engineers, we get rehashes of what exists already with minor operational tweaks. Without medical expertise and industry experience, we get misapplied product focus and weak business models.
Venture investing in services
While conventional wisdom holds that these businesses are more capital-intensive and lower margin than SaaS, tech-enabled services have in fact offered attractive returns for venture capital investors.
Early investors in Livongo realized IRRs comparable to Lyft,  and One Medical likely delivered IRRs over 35% to most investors. See below for a breakdown of estimated IRR by stage:
To be fair, Livongo has no brick and mortar clinics and manages patients remotely, which offers a structural advantage that is not possible for all types of healthcare. But meeting patients face-to-face doesn’t necessarily mean building brick and mortar: many new businesses are focusing on delivering care in the home setting. And as One Medical illustrates, even setting up clinics can pay off if matched to the right business model and executed well.
Where do we go from here?
Building technology for healthcare providers is certainly not a lost cause. Nuance, Phreesia, and to some extent Health Catalyst show us that it is possible, as does Innovaccer’s rapidly growing footprint. In our own portfolio we are excited about Commure’s approach to accelerating software innovation for health systems.
Additionally, health IT vendors will in the near future be required to publish APIs that will allow for easier interchange of data. The Office of the National Coordinator for Healthcare IT (ONC) recently published their final rule on interoperability which details requirements and timelines. This will improve the speed at which software can be developed, distributed, and deployed.
But selling to providers will continue to be challenging, and the model still won’t work for some types of software and for ideas that radically change the way medicine is practiced. In those cases entrepreneurs can often realize the value of their intellectual property through care delivery.
Every specialty and hospital department is ready to be re-imagined. The hospital will be unbundled and replaced with highly-efficient “focused health factories,” a concept first proposed by Regina Herzlinger of Harvard Business School.  Hip and knee replacements, pregnancies, oncology and more will be streamlined; plenty of gas is left in the tank for chronic disease and primary care as well. Under the strong leadership of Adam Boehler and now Brad Smith, novel financing models from the Center for Medicare and Medicaid Innovation are encouraging investment and accelerating progress.
Medicine is filled with tradition, risk-aversion, and regulation. It takes tremendous courage and conviction for teams of physicians and engineers to propose a new way to practice medicine. But if we want Americans to enjoy better healthcare at a radically lower cost, technology is our best hope — not price controls.
There is a decade of progress in information technology that has not yet made it to the clinic. The canonical SaaS playbook didn’t work. We need our brightest doctors to join forces with the technology community to deliver the transformation that was promised.
Thanks to Jack Moshkovich for assisting with IRR analysis, and to Jack and Drew Oetting for valuable input.
 Save for the odd installation of something like Health Catalyst or Nuance, which have achieved meaningful penetration.
 Payers have experienced analogous stagnation alongside providers, still harnessing antiquated systems for care management, population health, utilization review, claims adjudication, and more. Some of the factors behind it are the same: long sales cycles, data privacy, and regulation. In light of this Oscar built a complete insurance stack from scratch to power its unique offerings.
 See: ONEM, LVGO
 These are illustrative estimates only. Each analysis treats all investors in one round as a single unit and relies on best-guess assumptions about pro rata participation.
LVGO — calculates returns based on market cap of $2.50B on 1/26/20, 6 months after IPO on 7/26/19
ONEM — calculates returns based on market cap of $3.31B on 4/28/20; as of writing 6 months haven’t elapsed since IPO