ARK Disrupt Issue 109: Deep Learning, Health Apps, Autonomous Vehicles, and Cancer Detection

Please enjoy ARK Disrupt Issue 109. This blog series is based on ARK Brainstorming, a weekly discussion between our CEO, Director of Research, thematic analysts, ARK’s theme developers, thought leaders, and investors. It is designed to present you with the most recent innovation takeaways and to keep you engaged in an ongoing discussion on investing in disruptive innovation. To read the previous issue, click here.

1. Deep Learning Predicts Hospital Stay Outcomes With 95% Accuracy

Follow @jwangARK on Twitter

University researchers and GoogleGOOG have built a deep learning model that can predict the outcome of a patient’s stay in a hospital with up to 95% accuracy. It outperforms current machine learning models in predicting mortality rates, 30-day readmission rates, lengths of stay, and diagnoses upon discharge.

While these results are impressive, hospitals have been reluctant to embrace such models…until perhaps now. The biggest impediment has been converting patient records into a machine-readable format: according to the authors of the study, that step represents 80% of the work.

Remarkably, the new model presented in the study requires minimal data processing: all of the patient’s records – charts, notes, prescriptions, etc. –  are converted into a timeline structure easily read by a recurrent neural network. While conventional models work only with clean, structured data, this deep learning model should be much easier to implement in clinical settings.

The proposed deep learning model also provides explanations for its predictions. Unlike early opaque “black box” models, this one is open and transparent, highlighting the data that influenced the decision-making process.

Effectively, hospitals are data centers, but little of the data today is captured or analyzed in real time. Deep learning will convert data streams into potentially life-saving alerts. Given how quickly healthcare costs have ballooned, we believe that a number of unicorns are evolving to capitalize on the predictive health care space.

2. Apple Will Launch a New Health Records App this Spring

Follow @juliahARK on Twitter

Despite massive advances in technology in recent years, consumers have little access to their own health records, a problem that AppleAAPL aims to solve. This week, it announced that its spring update for iPhones and iPads “will include a new ‘Health Records’ feature. The update will import and store consumer medical information such as allergies, conditions, immunizations, lab results, medications, procedures, and vitals.” Patients will have access to their own data and will be notified when their health records are updated.

Apple is working with Electronic Health Record (EHR) companies such as athenahealthATHN, CernerCERN, and Epic on this product, taking advantage of the Fast Healthcare Interoperability Resource (FHIR) protocol to draw information into the application. To launch this feature, Apple is working with 12 U.S. hospitals, including Johns Hopkins, Cedars-Sinai, and Penn Medicine. Important to note, Apple will not have access to the data, as health records will be encrypted.

As noted in Healthcare IT News, Apple has a meaningful opportunity to succeed in the personal health record (PHR) space, thanks to software development kits (SDKs) like HealthKit, ResearchKit, and CareKit, as well as its widely accepted range of devices and the biometrically secure nature of its ecosystem. As a result, Apple Watch could become an integral part of an Apple-based health-tracking ecosystem.

3. What Happens When Autonomous Vehicle Systems Get Confused?

Follow @tashaARK on Twitter

When they commercialize, what will happen when autonomous vehicle systems get confused? ARK believes that a remote network of third party operators will direct or re-direct them. Already, NissanNSANY, WaymoGOOG, and GMGM have gravitated toward this strategy, while startups Phantom Auto and Starsky Robotics are evolving operator networks for both passenger cars and trucks.

Based on our recent conversation with management, Phantom Auto uses image recognition from 360-degree camera feeds and sends videos to remote operators over cellular networks. While different sensor suites are emerging, if camera feeds begin to dominate, then companies like TeslaTSLA should be in a good competitive position: Tesla has collected more video data from its customers’ cars than any other company jockeying for position in the autonomous vehicle space.

This strategy should gain more acceptance in one to two years with the debut of 5G networks, which will offer roughly 10 times the bandwidth and a 10-fold improvement in latency relative to existing cellular networks. Virtual reality (VR) also could become an important part of this ecosystem, helping operators to monitor video feeds and direct autonomous vehicles when they run into uncertain situations.

ARK looks forward to following Phantom Auto as it demos its solution next week and as autonomous vehicle systems evolve during the next few years. Needless to say, we’ll keep you in the loop!

4. GRAIL Updates Analysts on Its Progress in Liquid Biopsies to Detect Early Stage Cancer

Follow @msamyARK on Twitter

Spun out of DNA sequencing giant, IlluminaILMN, GRAIL is developing a minimally invasive blood test for early stage cancer detection. In January 2017, Illumina accelerated GRAIL’s path to independence by reducing its ownership from 51% to less than 20%. Now GRAIL is ramping to become Illumina’s largest customer.

Just a year later, at the Global Precision Medicine Conference this month, GRAIL provided an update on its clinical roadmap for liquid biopsies. As illustrated below, the holy grail for cancer survival is early diagnosis.

Aravanis AM, Lee M, Klausner RD. Cell. 2017; 168(4):571–574, 9.

In its update, GRAIL announced that it plans to start a clinical trial for nasopharyngeal carcinoma in Hong Kong, the city with the highest prevalence of that strain of cancer. By the end of 2018, it expects to reach full enrollment with 15,000 patients, 10,500 already diagnosed with the disease and 4,500 healthy patients. GRAIL will monitor those patients for the next five years, calibrating its liquid biopsy assay along the way. GRAIL emphasized that the trial is “very much a discovery study” and will help “build a reference set of cell-free DNA profiles.”

That said, the study also will influence a breast cancer clinical trial that GRAIL anticipates will lead to a working diagnostic test in the next few years. Along with the Mayo Clinic, GRAIL will enroll 120,000 patients in that trial.

Liquid biopsies rely on the detection of cell-free DNA that circulates in the bloodstream after a cancer cell dies. Typically, these DNA fragments are only 170 base-pairs in length and linger in the blood stream for only two hours. Consequently, GRAIL will have to sequence with 60,000 coverage, that is sequencing the same DNA sample 60,000 times, 2,000 times more than normal. It also will sequence bisulfites, which are linked to methylation and will help determine the cancer’s tissue of origin.

While these trials are in very early stages, if successful they could change the course of cancer patient outcomes dramatically. Sequencing technology is at the foundation of these breakthroughs and is in the very early days of what could be explosive growth.

ARK's statements are not an endorsement of any company or a recommendation to buy, sell or hold any security. For a list of all purchases and sales made by ARK for client accounts during the past year that could be considered by the SEC as recommendations, click here. It should not be assumed that recommendations made in the future will be profitable or will equal the performance of the securities in this list. For full disclosures, click here.