Digital Health 2015: Perspective from Rock Health

Rounding the corner for the final stretch of Biodesign year 14, SBAA engaged our friends at Rock Health to discuss the digital health funding landscape and the state of predictive analytics as clinical decision support (CDS) tools. Strategy Manager Teresa Wang and Managing Director Malay Gandhi walked us through some numbers, trends, and case examples.

Funding

After a record $4.1B funding for digital health in 2014, we have seen $630M of Q1 funding in 2015, representing 15% higher funding than the preceding 16-quarter average.

The six largest rounds of the year in this sector totaled nearly $240M, representing 40% of the quarter’s funding. Deal distribution has remained consistent over the past few years with seed and Series A rounds accounting for 60% of deals and Series B rounds accounting for 21% of deals. Accordingly, growth and maturity expectations for digital health companies seem to have driven later stage funding anticipation. In total, corporate financing supported four deals led by GE, Qualcomm, and Merck; venture financing supported two deals lead by NEA, HLM, DCM, Easton, RRE, Lux, Thrive, and Sandstone.

Categories representing the largest segments of digital health funding in 2015 to-date include (1) analytics/ big data at $107M, (2) healthcare consumer engagement at $68M, (3) EHR/ clinical workflow at $63M, and (4) digital diagnostics at $55M.

Predictive Analytics

Traditionally, clinical trials capture patient data, and generalized recommendations are derived based on the “average” patient enrolled. Aggregated clinical trial results then make their way to society-endorsed guidelines to be used with best clinical judgment. With software-based clinical decision support, patients provide data to physicians to optimize decision support algorithms for diagnosis or treatment. And with patient-controlled data collection, the patient themselves collect, own, and submit their own data to be used in predictive algorithms for clinical insights.

The airplane cockpit model is commonly referenced as the future of medical practice: largely automated systems with humans in place for manual override. Andy Rink shared his stance on this analogy: “In aviation, if the weather’s bad, you don’t have to fly. In medicine, if the weather’s bad, you still fly.”

Big data and widespread use of analytic tools has catalyzed overall growth in the health care space. While we’re moving closer to new applications and expanded adoption of these potentially powerful tools, it should be noted that more than 15 years after sequencing the human genome, public expectations and reality of the implications are not aligned.

The breakdown of new data streams being used for predictive analytics include (1) clinical at 71%, (2) claims at 42%, (3) patient-generated (wearables) at 42%, (4) patient-reported (surveys) at 26%, (5) research at 15%, (6) molecular at 14%, and (6) clinical trials at 6%.

From these data streams, personalized care will emerge with modeling based on high-confidence algorithms that predict actionable interventions. Context and action remain the core drivers of clinical decision-making applications to effectively detect, report, and intervene based on clinically relevant data. Downstream, tracking outcomes data may then be used to refine algorithms and deliver enhanced tools.

Case Examples

Data Collection and Aggregation:
Apple’s Research Kit and Google X’s BASELINE study harness activity and physiologic metrics to define “normal” and to track progression to “abnormal” with hopes of identifying patterns to feed into predictive models for preventive measures. Platforms like these may serve as the new medium for clinical trials.

Wildflower Health addresses high-risk pregnancy with real-time nurse hotlines and data collection to reduce traditional intervention response times. Over 40% of pregnancies in America are covered by Medicaid, and maternity ranks among the top 3 most costly conditions for providers, so optimizing monitoring and early intervention in these patients translates to targeted resource allocation to avert high risk complications and financial savings. Wearing a monitoring patch for 14 days, identifying 1/1000 patients resulted in the program paying for itself due to excessive cost associated with preterm birth. By emphasizing engagement, patients continued to participate and strengthen the program.

Flatiron Health cleans up data with a primary mission “to organize the world’s oncology information and make it useful for patients, physicians, life sciences and researchers.”
Per their website, they are “building data pipelines, infrastructure and software to provide cancer centers, physicians, researchers and patients with the critical insights they need to make informed decisions.”

Wellframe provides a mobile interface and aggregates post-cardiac event data as a form of cardiac rehabilitation post-discharge, ultimately aiming to predict future cardiac events. Physicians prescribe this service that is billed a “care coordination” and reimbursed under CCM codes (Chronic Care Management).

Lumiata addresses user experience and generates “graph-based analytics” that “encapsulate relationships between tens of thousands of variables.” They “develop high dimension, temporal representations of data, enabling individualized models to be generated from messy and incomplete data.” They “build clinical and financial models at individual and population levels for use by humans and computable systems via apps and APIs in real time.”

Contextualization:
Evidation Health aggregates data from assigned trackers and contextualizes it to the user. They identify what works for each individual user and tailors recommendations to enhance engagement. If Facebook knows what you’re doing, it can make recommendations to get you moving.

Performance Capture:
Health Catalyst rapidly optimizes algorithms by re-training through closed loop models, building long-term defensibility. With access to claims, billing, and outcomes data, they can deduce what works and what doesn’t. The goal would be to recommend specific treatment regimens based on a similar population’s historic response (i.e. which statin might be more effective, etc.).

Closing Thoughts

So what are the implications of getting things right? Successful implementation might reduce the $192B currently spent on overtreatment and the $128B spent on failures of care delivery, and provides value to payers in the form of personalized medical policies and benefits.

IP is less relevant in the digital health sector. Many companies no longer publicize their IP, rather electing to maintain trade secrets. What matters most is who has the best algorithm and can actually apply it.

Beyond regulatory factors, the biggest risk for success is adoption from patients and healthcare professionals. Barriers to adoption include privacy concerns, algorithm transparency, liability concerns, and loss of decision-maker power.

“I’d be comfortable being treated by a robot,” said one attendee. But physicians generally don’t like “cookbook” or deterministic medicine; they like probabilistic medicine. EHR system development is heading to a place where physicians treating patients with incompletely validated regimens may select order sets of randomized treatment. If proper infrastructure were in place to support such a system, 60% of physicians state that they would adopt such a technology.

Ultimately, you need engagement. You need to get to market first. You need to collect data clearly. All of this must be actionable to the patient or physician.