Is AI the Pain-killer for Healthcare?

In the recent years, AI has become the new “hot-word” in the tech world. It seems to be the perfect answer to every challenge, from customer support, office management, to financial advisory. In particular, as people become increasingly unsatisfied with the U.S. healthcare system, many entrepreneurs and investors seek disruptive solutions from the emerging machine intelligence technology.

First Thing First, Why Invest In Healthcare AI Now?

1. Time for an upgrade: Digital Health has been a hot field in the recent 5 years, companies like Fitbit, TelaDoc (NYSE:TDOC) and iRhythm (NASDAQ:IRTC) have already achieved remarkable success and accepted by the public. It is time to bring Digital Health to the next level, and AI will be the key.

2. The infrastructure is ready: As digital health solutions and personal gene testing become more popular in the past decades, healthcare data is increasingly accessible. Moreover, Google (TPU)and Nvidia (GPU) are pushing computing to the edge. Thanks to GPU those innovations, our supercomputers are more powerful than ever, and more importantly, evolving rapidly. Today, researcher are able to analyze a massive amount of data, fast enough, to tackle the most complex computational challenges, such as drug R&D simulation and behavior prediction.

Amazon also released a whitepaper, Architecting for HIPAA Security and Compliance on Amazon Web Services, and HIPAA Faqs to help companies building applications in the AWS cloud meet HIPAA standards.

3. Government & Regulation Support: government and regulations started to embrace the novel Digital Health technology. For example, earlier this year, the National Health Service of the UK started to adopt Sensely’s “virtual nurse” technology; and this passed April, FDA approved digital pathology for primary diagnosis for the first time.

According to CB Insights, Funding in the AI in Healthcare sector increased tremendously from only $30 million in 2012 to $748 million in 2016. As Bill Gross laid out in his Ted Talk, Timing is the dominant factor for success, healthcare AI is definitely on the rise. Nevertheless, to make wise investments, we also need to understand the core values that startups in this sector are delivering. Below, we’ll discuss a few ways that we think AI can transform the healthcare industry.

Virtual Nurse

Probably the two most critical issues of the healthcare system are accessibility and affordability. Doctors are scarce and expensive healthcare resource; however, patients, especially those suffering from chronic diseases, do require day-to-day guidance to achieve optimal health. Moreover, even if there are enough doctors, the majority of the population won’t be able to afford long-term private healthcare services.

Furthermore, as stated by the World Health Organization, chronic disease prevalence is expected to rise by 57% by the year 2020. Emerging markets will be hardest hit, as population growth is anticipated be most significant in developing nations. Hence, effective disease management, as a traditional pain point, is becoming a more significant rising concern.

Yet, with the modern AI technology, it shouldn’t take an educated human to do such repetitive work. AI-powered virtual nurse can handle tasks such as routine physical checks, collect and analyse data, provide standard advice, etc. and free the doctors to work on more critical tasks; therefore, increase the efficiency of the modern healthcare system and lower the cost.

Health Management/Virtual Nurse for chronic diseases is certainly a very sizeable market. According to the Centers for Disease Control and Prevention, in 2010, 86% of all healthcare spending was for people chronic medical conditions; the total medical spending for heart disease and diabetes were estimated to be $193.4 billion and $176 billion. Yet, according to Sensely, which is an industry leader in virtual nurse for chronic disease management, its solution was able to cut 72% of the unnecessary calls and significantly increase the productivity clinics (36Kr reported, in a pilot study, the productivity was increased by 20%).

TAM for AI-based Chronic Disease Management Solutions
Wearables & Patient Monitoring

Very much similar to Virtual Nurse, AI-based Patient Monitoring is another promising sector. Startups in this sector usually offer complete hardware-software integrated solutions: smart hardware (wearable) that collect patients’ data, AI-based engine that analyse the data, and apps that interact with the users/patients to provide guidance. Compared to Virtual Nurse startups, Wearable & Patient Monitoring startups usually target specific health problems, for example, asthma, sleep monitoring, heart disease, or addiction. We saw great synergy between the two sectors with Virtual Nurse platform working as the iOS system that supports different applications and wearables being the apps that make iOS system thrive.

Despite targeting specific health problems, the business opportunity for chronic disease patient monitoring is sizeable. As on of the most prevalent chronic disease, about 1 in 12 people have asthma, and the numbers are increasing every year. According to a research on Asthma control on the European Respiratory Journal, out of 10,428 asthma patients, only 41% are well controlled.

Unadjusted Median All-cause Expenditures by ACQ-5 Score (a measure of asthma control)

Poor asthma control also leads to significant increase in healthcare cost. In a recent observation study of Asthma Control and Outcomes (OSACO), patients received a survey 3 times in one year, which included the Asthma Control Questionnaire (ACQ) and questions on exacerbations. The results shows that medical expenditure is highly associated with control (ACQ-5 score).We see that with proper control, medical cost can be reduced by almost 80%.

Medical Imaging (Pathology / Radiology) Diagnostics

Another field that we believe AI will play an important role in is Medical Imaging Diagnostics. For over a hundred years, the reference method for the diagnosis of cancer and many other critical clinical conditions has been histopathological examination of tissues using conventional light microscopy. This process is known as surgical pathology in the U.S.

Source: Efficiency gains in cancer care (Sectra)

In surgical pathology, patient tissue from surgery, biopsy or autopsy goes through a process that includes dissection, fixation, embedding, and cutting of tissue into very thin slices which are then stained and permanently mounted onto glass slides. The slides are examined by a pathologist under a light microscope. Clearly, this is a labor intensive process which usually takes more than 2 weeks.

Nevertheless, since early 2000s, digital pathology has always been a more efficient solution for data sharing and analysing. What has been really holding the industry back was regulation: FDA has only started to approve digital pathology for primary diagnosis in April 2017. More importantly, digitalized data allows for the application of AI, or machine learning tools, to analyse images and provide diagnostic suggestions.

In fact, AI can handle the histology job not only much faster but also more accurate than humans. A research was conducted in 8 U.S. states from 2011 to 2014, covered 6900 individual case diagnoses. It shows that compared with the consensus-derived reference diagnosis, the overall concordance rate of diagnostic interpretations of common pathologists was 75.3%. There is still substantial room for improvement if we consider that AI can potentially achieve over 95% accuracy.

Computational Medicine

The last sector we want to cover is Computational Medicine. Fist of all, what is Computational Medicine? In a nutshell, it is characterized by modeling, simulation, and visualization of biological and medical processes in computers with the goal of simulating real biological processes in a virtual environment. Need to say that though this has been a hot subject in academia since the 1990s, Computational Medicine is still far from mature. Nevertheless, we are excited to see innovation in this field in the next 5 to 10 years. Here are the reasons:

1. Due to technology advancement, data is becoming increasingly accessible. The DNA sequencing cost has decreased by almost 100,000 times since 2001. Companies like Human Longevity has created an open research database of over 10,000 genomes. Moreover, the infrastructure tech is also quickly maturing. For example, DNAnexus provides a cloud-based data analysis and management platform for DNA sequence data. Leading startups like Sentieon also offer analysis tools that process genomics data with high computing efficiency.

2. . Thanks to industry pioneers like Nvidia and Google, computational power is increasing exponentially each year, which enabled a new framework for drug discovery. As one researcher commented that, with GPU, he can now do his life’s work, in his lifetime.

3. The market is substantial. According to the PhRMA 2016 biopharmaceutical research industry profile, the average time to develop a drug takes 10 to 15 years, the average cost is $2.6 billion, and only 20% of the marketed drugs return revenues that match or exceed R&D expense. The fundamental R&D approach has remained the same for many decades, and the annual R&D cost is increasing with 7% CAGR since 1995.


So, what does a promising AI healthcare startup looks like?
A 10x Better Solution To A Major Pain Point

For any startup, after identifying a pain point, the first challenge that all have to overcome is “from 0 to 1”. An entrepreneur can either become a pioneer in a whitespace or competing with existing solutions offered by matured companies or even industry giants like Apple and Johnson &Johnson. Either way, to get trial users for early adoption, the new solution has to be very attractive to customers, enterprises and/or consumers.

As early-stage investors, we spend a lot of time researching from this approach. First, the pain point needs to be solid and “pain-enough”. For example, the pain point Sensely identified was chronic disease control, well, here is some data partially shows why this is a good start point.

High blood pressure may be the most common but seemingly mild chronic disease. According to the cdc.gov High Blood Pressure Fact Sheet, about 1 in every 3 adults in the U.S. have high blood pressure; however, only about half (54%) of them have their condition under control. Furthermore, the shocking fact is: high blood pressure was a primary or contributing cause of death for more than 410,000 Americans in 2014 — more than 15% of the annual American registered resident death (number of firearm death in 2014 was 33,599).

As we can clearly see the pain point, treating chronic diseases effectively requires time that doctors don’t have in daily calls and brief office visits (would be too expensive as well), and a degree of daily self-management, which often overwhelm the patients. Well, AI can be a perfect solution. Potentially, a personal AI health advisor can handle simple tasks like daily symptom check or help the patients to form healthy living habits just as good as humans. Even better, it works 24/7 and doesn’t need to be paid.

Defensibility: Data and Partnerships

Healthcare is a quite crowded vertical for AI applications, and competition is a major challenge for everyone. In 2016 along, CB Insight recorded over 90 deals in the sector. So, how to become the winner?

First, a promising AI healthcare startup should be able to collect a considerable amount of high-quality data. Because all AIs are more or less unintelligent during their infancy, yet they rely on the massive amount of data to “educated” themselves to become intelligent and continue evolving overtimes. Therefore, a question that we usually ask the early-stage entrepreneurs is “how do you plan to get the first batch of high-quality data to train your AI reasonably smart?” Being much smarter than others is the No.1 defensibility for all AI startups.

Moreover, partnerships and connections are important when considering the business model of an AI healthcare startup. Unlike many other industries, the healthcare industry is heavily influenced by government sector and regulators. And, often, large companies have monopolies over the distribution channels. Having unique access to certain resources can be an irreplicable advantage.

A Team with Multidisciplinary Expertise

Most importantly, it takes a very versatile team to run a successful AI healthcare startup. Above all, the team shall have expertise in healthcare service and has an empathy first value. In addition, having expertise in AI, which is a combination of data and computer science, is also crucial. Besides, sales and product development expertise are just as important to healthcare AI startups as well.

Business Model

Finally, the business model is fundamental for any startup. Though many may think that early-stage startups usually have a long way to profitability, we believe entrepreneurs should plan for a sustainable and scalable business model in the very beginning. On one hand, this shows that the founding team has a good understanding of business and finance(otherwise, we should help them to develop such expertise). On the other hand, profitability should be in the DNA of all companies.