Digital Diabetes Congress 2018

Speaker: Jianying Hu, PhD

AI for health care is all about enabling the journey from data of different types collected from different sources in the health echo system to impact at an individual level to support precision medicine: prevention and treatment strategies that take individual variability into consideration. The computational challenges in this journey can be roughly categorized into three pillars: (1) Pattern Extraction, (2) Insights Generation, and (3) Personalized Interventions. Pattern Extraction is all about data preparation: from preprocessing which focuses on addressing issues such as data irregularity, noise, bias, sparsity, and heterogeneity across disparate data sources to feature engineering and selection to enable automatic identification of most important patterns and features. Insights Generation is the central pillar that requires methodological innovations in many key areas, including methods to automatically assess patient similarity in different clinical contexts; comprehensive multitask learning methods capable of discovering and leveraging interrelationships among multiple conditions and factors; longitudinal modeling approaches that can go beyond traditional point predictions and capture the temporal evolution of a disease and its characteristics at different stages for more actionable insights even before it become symptomatic; methods to improve the interpretability of AI models; and causal inference for robust and scalable effect estimate using real-world data. Personalized Intervention is about taking the insights and making an impact for the individual. A key challenge here is how to understand behavior and take behavioral factors into consideration. Methods are being developed to learn behavioral patterns from data, and to put these learning algorithms in the context of a well-defined behavioral intervention framework to support systematic ways of defining behavioral factors, measures and objectives, and systematic ways of designing, testing and evaluating behavioral interventions.

Tremendous progress has been made in recent years in all these three pillars, with numerous early proof-of-concept demonstrations of the power of AI for health care. Significant effort and investments are now being put forth to accelerate the impact—we are at an exciting time for digital health and AI for health care.

Moderator: Courtney Lyles, PhD

Panelists:

One major topic affecting broad uptake of digital diabetes solutions centers around access. Specifically, this panel at the Digital Diabetes Congress focused on access as it intersects with digital solution functionality, stakeholder engagement, and financial investment.

In terms of functionality, the panel members shared their personal experiences in testing digital solutions with broader groups of patients and expanding the features of the tools. This included examples of content, such as ensuring that the literacy, numeracy, and language of the products were relevant for diverse patient populations. But it also included broader functionality, such as testing across mobile device platforms (eg, both Android and iOS as well as text messaging capability) and adding support services (eg, virtual coaching or other personalized assistance). Finally, the panelists discussed how the usability testing of the digital platforms needed to include diverse users rather than convenience samples that might skew toward those with higher socioeconomic status, younger ages, or other characteristics that limit broad generalizability.

In terms of stakeholder engagement, the panel members were also quick to mention that this broader focus on functionality and patient engagement necessitates deeper stakeholder engagement across multiple domains. In order to truly design, test, and then disseminate digital health tools to more diverse groups of patients with diabetes, it is common to have to establish new collaborations and expertise. For example, user testing in multiple languages might require relationships with community organizations or health care systems serving monolingual patient populations. Similarly, digital health implementation strategies that build on provider or community leader recommendation of a specific platform also might require different types of partnerships—and might represent new roles for digital health companies specifically.

Finally, the panel discussed the need for these type of activities to be measured appropriately and communicated to those making financial investments in the products themselves. In other words, the return on investment for the digital health solution could benefit from consistently measuring and reporting the reach and adoption of the product across diverse end users as well as the usability and patient experience with the tool. For example, if a health plan made an investment in a specific digital diabetes product for all patients within their system, then specific measures for the uptake, sustained use, and patient ratings of the product by key subgroups might improve the access of the digital tool—beyond measuring clinical impact alone among a smaller group of early adopters.

Moderator: Charlene C. Quinn, PhD, RN, FAAN

Panelists:

The adoption of diabetes digital tools and interventions is dependent on translation of research and clinical end points providing evidence of efficacy and effectiveness. In this session, experts discussed the following issues:

Digital tools enable us to assess continuous glucose (CGM) data, which may provide more granular information for treatment adjustments. Concern was expressed that regulatory and payer stakeholders consider current digital diabetes research as observational end points and not effectiveness confirmation. The session was summarized by recognizing that diabetes digital research design, including evaluation of new end points, is crucial to the rapid technology changes.

Moderator: Saleh Adi, MD

Panelists:

This panel convened representatives from large pharmaceutical companies, smaller biotech companies, and device manufacturers.

The panel began by addressing the question of whether health care companies will be primarily consumers of digital health solutions or be major contributors to digital health solutions. The responses indicated that overall companies are just beginning to ponder these kinds of decisions, and that the answers will likely depend on several factors. For example, most devices used in managing diabetes include software that is essential for the function of the device, therefore it is expected that device manufacturers will develop their own software in tandem with developing the hardware. On the other hand, large drug manufacturers may find it more efficient to either acquire digital health products as needed or partner with software companies to develop what they need. There was general agreement that the nature of software development is best done by smaller digital companies which tend to be more nimble and agile, and better suited for the faster pace of designing and building digital solutions than large pharma companies. An added advantage of not building software internally is that independent digital solutions provide more interoperability and prevent further development of vertical silos within large device and pharma companies.

Regardless, panelists acknowledged that digital solutions are already changing the way biotech and pharma companies conduct their business. The simplest example was how using social media has become important for reaching a wider base of subjects and patients and accelerating recruitment for clinical studies. Another important example is the use of digital platforms to collect patient-reported outcomes, using tablets or web-based questionnaires and software specifically designed for data collection.

The panelists then argued that while use of digital health solutions accelerate drug development, approval, and marketing, it also brings an added large volume of data that is collected during clinical trials. This is especially true in the field of diabetes, where use of technologies such as insulin pumps and continuous glucose monitoring (CGM) generates a wealth of various data points that most pharma companies may not be well equipped to utilize to its fullest potential. This is an increasing challenge that drug makers are just beginning to address. The other challenge when a “pharma” solution or a new device are developed with integrated software, the software must be very intuitive and on par with patients’ expectations of matching the quality and speed of any other personal digital devices, such as smartphones and tablets.

When asked about the biggest gaps is digital health in the industry currently, there was general agreement on the need to integrate the diabetes data collected with digital tools into patients’ electronic medical records (EMR). Currently, pumps and CGMs collect extremely valuable information that are the basis for complex medical decision-making and adjusting insulin or drug regimens. These data are largely remaining outside of the EMR. The other challenge identified is the need to increase utilization of human factors and psychology in designing digital technologies to optimize utilization and adherence with diabetes treatments. Interesting comments were then made regarding the importance of designing comprehensive digital solutions that meet the needs of not just “the diabetes” part, but also the often common comorbidities associated with diabetes, such as hypertension and liver and cardiovascular diseases.

The panel then tackled the topic of data ownership. There were differences of opinion, which seemed to stem from specifying which data set is in question. The panelists agreed that data generated by subjects during their participation in clinical trials are owned and controlled by the trial sponsors. However, data generated during routine care should be owned by the patients themselves, who may then elect to donate them for research purposes, or back to the device manufacturers in order to improve their products. In either case, the panelists expressed overall enthusiasm and excitement for the immense potential of big data to gain insights and illuminate the development of better digital solutions, including the potential role of artificial intelligence and machine learning in mining big data and improving algorithms used in closed-loop systems. In the world of diabetes, there is increasing focus on collection of postmarketing data, which reflect real-world performance and should be made available to device manufacturers to reiterate and improve the design of their products. The panel concluded with a discussion around the importance of collecting other contextual data that are not necessarily collected by the digital tools. For example, continuous glucose monitors collect BG data, but these data may be useful only if provided with all the contextual information that influence BG levels, such as insulin doses, food, and exercise information.

Speaker: Bakul Patel, MS, MBA

The FDA is working to align itself with the needs of the digital health industry and recognizes that a new regulatory pathway is needed to enable products to be available to patients faster without compromising safety or effectiveness. When considering software as a medical device (SaMD) new questions will arise, such as how the FDA will evaluate functions that are not part of a medical device but may affect the safety or effectiveness.

We are moving toward this concept of organization based regulation. The main goal of the planned Precertification Program is to create a culture of excellence among manufacturers; one in which they would be expected to consider patient preference information and develop products that best serve patients. We must see if we can offer credit to companies that consistently offer products that perform well in the real world. A good analogy is the TSA PreCheck program. We trust that these companies are making good products based on their history just as we trust that those that voluntarily submit to background checks are likely to have good intentions when flying. This is not a guarantee that they will continue to do so forever. This is why we need postmarket surveillance.

This is where the FDA needs input. Some may say that the proposed plan is a step in the right direction, while others will say that it is not fast enough or not tailored enough. To those who would say the latter: submit feedback and tell us how we can make it faster and better suited to your needs. We want to trust the companies and you want to make sure that the products that are reaching the marketplace are verified. How can we perform better postmarket analysis to ensure that products are doing as well as they claim to be doing? We must create a culture of transparency within these companies.

There are risks when it comes to software and digital products that may not be as visible as they are in other types of products that we regulate. We are working hard to address these concerns while creating a more efficient regulatory process. In order to make the program more customer-centric, we need input from you.

Moderator: Kong Chen, PhD, MSCI

Panelists:

The regulatory landscape for digital health is evolving. The Precertification program by the Food and Drug Administration (FDA) is still under development, but with good progress and promising opportunities. Recent implementations of software as medical device highlight the digital health revolution, which include clinical decision tools and products. This is particularly important for diabetes, a predominantly self-managed chronic disease. Depending on its intended use and potential risks involved, software can be regulated by the FDA and/or broadly by other bodies. Patient safety and good manufacturing processes for effective care are key to medical devices. Challenges exist for regulating apps, but the data privacy and cybersecurity are also regulated. FDA, FTC, IMDRF, state AGs, SEC, and HHS (OCR) all play important regulatory roles. The ultimate goal for these agencies is to work closely to benefit patients with diabetes who are using digital products, and thus promote and protect public health. Security and privacy needs to be incorporated in the design and product development process. Industry and developers should know the different regulatory bodies, and work with regulators collaboratively to minimize risks. To stay within compliance, health care companies should have annual risk assessment of HIPPA requirements, a privacy policy with vendors, and considerations of individual rights (via EU GDPR requirements but these will extend to the United States). For small and innovative start-ups and developers to navigate through these regulatory issues, it is helpful to enlist experts in this field.

Moderator: Korey Hood, PhD

Panelists:

There are many apps and other digital health tools available today that may help with patient adherence. For the purposes of this discussion adherence will be defined as the degree to which a patient or person with diabetes follows a treatment plan. Mobile apps, data from diabetes devices, and electronic medical records are just a few of the components of the digital framework that can help patients with adherence.

The key way that digital tools increase patient adherence is by enhancing the patient experience and helping to track diabetes management behaviors in real-time. This could be as simple as providing direct educational support to the patient or it could be a patient’s CGM sending trends to a physician without the patient having to do anything extra. There are many pros and cons to these digital tools. For example, some apps have been shown in clinical trials to improve A1C, especially among people with type 2 diabetes. However, few randomized controlled trials have been conducted and often by the time a trial is complete the tech has usually been updated. There is also some evidence to suggest that mobile interventions improve hypoglycemia in patients with type 1. Some evidence suggests that simple interventions such as text messages can be helpful with medication adherence, but the literature is mixed on the overall benefit of mobile interventions across disease states.

It is difficult to address the shortcomings of digital health as a whole because it encompasses so many different apps and tools. One issue is that patients who want to know which apps are most effective do not have a credible source for this information and many clinicians find it difficult to make a recommendation because there is little literature on app effectiveness. Future trials should aim to connect specific adherence behaviors to specific features of an app, and test those connections in addition to broader attempts to improve outcomes such as A1c. In order to do this, you have to know what outcome you are looking for, and one of the shortcomings in digital health is a lack of clarity about the desired outcome.

One way to determine desired outcomes is to include patients in the design process. App developers need to figure out how to get patients to stick with their apps, and apps need to be adaptable to remain useful across the disease states. We also need to incentivize physicians to analyze data from apps. Most physicians simply do not have time to analyze this data, so physician input in app design is another critical step to making these apps effective. When an app easily integrates with patient lifestyle and physician workflow, it is much more likely to be successful.

Apps and other mobile interventions can be very effective tools for managing diabetes and increasing adherence. From increasing time in range, predictive low glucose alerts, and other technologies, digital tools for diabetes are proven to have positive clinical effects as well as increase patient adherence to treatment.

Speaker: Dennis Boyle, MS

The key to making a great product lies in what is called cognitive psychology design. This is rooted in immersing yourself in the situation and experiencing what people with the problem are actually experiencing. With respect to designing diabetes devices, this includes having patients, the consumers, on your team. What are they thinking and feeling? How are they interacting with the device? Sometimes what people say is different than what they do, and we have to be cognizant of that with design. If a product is designed for an older population, then include them in your design team. If you want to design a wearable technology, then make sure the people testing it think it is comfortable and doesn’t intrude on their daily activities. On this note, make sure the device is quiet, small, and unplugged.

Many successful designs are popular because they are foolproof. If you design an injectable that is impossible to stick yourself with, then more people will want to use it. It is also important to be mindful of people’s habits. For example, people are used to removing the cap of a pen and having the writing element be underneath the cap. What has happened with some designs, such as EpiPen, is that the cap is on the opposite side of the needle and as a result many patients stick themselves in the thumb instead of the thigh. Old habits die hard, so why not design around them?

One of the keys to successful designs is simplicity. Sometimes a word is worth a thousand icons. Signage can be helpful, but it is not always enough and can at ties lead to more confusion than clarity. One of the best features in design are behavioral motivators. They are a great way to keep patients involved with a device, especially if there a community element that users can tap into.

Proper UI/UX design for AI helps increase patient adherence and helps physicians manage the flood of data, decreasing burnout from data entry. If you want better designed products, then have patients and users on your design teams. For the academics, promote good design by highlighting products that you feel have good design. This is how design will improve across devices.

Moderator: David Kerr, MBChB, DM, FRCPE

Panelists:

The two headline topics for this session were (1) digital communities and (2) digital therapeutic platforms. Digital communities for diabetes can be found across multiple social media platforms. They allow people with diabetes to access, using the internet, others living with diabetes and also family members and caretakers. A key success of a digital community is uninterrupted and continuous access without structured inclusion or exclusion criteria for people interested in joining. Notably, digital communities can take multiple forms and focus on a variety of diabetes-related topics. This supports the concept of segmentation of the online diabetes community according to need. Therefore, people with diabetes can join multiple digital communities and this membership can change over time according to need. A continuing controversy is related to the participation of diabetes care professionals in an online community—currently the emphasis is for professionals to be aware of these communities as a resource rather than to actively participate in them. Metrics of success, at present, are focused on individual outputs rather than clinical end points, for example, testimonials and increased awareness of new devices and therapies with participants acting as their own controls to decide if their experience was beneficial. The free and self-governing nature of these communities also allows individuals to choose which ones to participate in and the duration of their participation. Although privacy and risks of misinformation being posted are concerns, at present these do not appear to create barriers for the type 1 community. However, there is recognition that more needs to be done for individuals with type 2 diabetes, which at present are less likely to contribute to online communities.

Digital therapeutic platforms have a number of features that differentiate them from smartphone applications in that they focus on a specific long-term medical condition such as diabetes, have specific outcomes that are measurable (eg, HbA1c or time in range), bring a number of constituents together (eg, people with diabetes, clinicians, pharmaceutical and medical device industries), and importantly are focused on safety. It may be useful to consider them as Class II medical devices. Other important aspects are embedded medical intelligence features (eg, algorithms for dose management) and add-ons such as access to coaches and the ability to reorder supplies or make appointments. Another differentiator is that, increasingly, digital therapeutic platform have trial evidence of their value. A key feature of such a platform is the requirement for participation by HCPs to set parameters for therapeutic interventions, for example, target glucose levels, the use of evidence-based dose titration schedules, upper limits of dose escalation. The metrics of success for a digital therapeutic platform are likely to be different for different stakeholders—people with diabetes, providers, payers, and industry partners. However, the challenge going forward is how to increase adoption of these platforms, which is likely to require new approaches to reimbursement for providers. It is also possible that with a move toward value-based outcomes the adoption of the platforms may be accelerated.

Moderator: Patricia Salber, MD, MBA

Panelists:

This all-female panel was designed to explore different aspects of the glass ceiling as it pertains to the digital health and related companies. Panelists included female founders and CEOs and a general partner in a venture capital firm. We started the panel talking about difficulties female founders face in raising capital. There was general agreement that there is bias, unconscious or not, that works against female founders. This may be compounded if the female founder is also a woman of color. One panelist felt that women were evaluated on what they have actually delivered whereas men were evaluated on the hypothesis that they will be successful. Another panelist, a serial entrepreneur, said that when she first started raising money, there were rarely any women at the meetings she attended, but the venture partner noted that in her firm the health care investor team is currently 50% women.

When asked what advice they would give to other women digital health entrepreneurs, one panelist emphasized the importance of being prepared for meetings with funders, researching who you will be meeting with beforehand because that is one element of fund-raising that is completely within your control. Another pointed out the importance of really getting to know your investors, spending time with them and learning from them. A male audience member felt it was critical that women find some male mentors to support them. One suggested when a woman’s idea is ignored in a meeting (as many in the room had experienced) and a man later repeats the idea, that he make of point of prefacing his remarks with “as Julie just said.”

By the end of the panel, there was some agreement that things are changing for the better when it comes to the glass ceiling, but there are still significant challenges facing female founders.

Moderator: David Klonoff, MD, FACP, FRCP (Edin), Fellow AIMBE

Panelists:

Should we be concerned about the cybersecurity of digital diabetes devices? All of the panelists agreed that cybersecurity is a concern. Whenever there are vulnerabilities in a technology, there is a “bad guy” out there waiting to exploit that vulnerability for a variety of reasons. We may not think that someone would want to hack into a diabetes device, but at the end of the day criminals will do anything if there is a way to monetize their actions or hactivists may want to cause chaos.

It is, however, important to quantify how concerned we should be. We have to be careful not to overdo the concern with the general patient population because the average user doesn’t necessarily need to be concerned about cybersecurity on a daily basis. What they should be concerned about is the lack of transparency from industry about what capabilities products have in terms of being able to protect data.

To address these concerns, the Diabetes Technology Society created the DTSec standard and DTMoSt guidance. Both were bottom-up initiatives that began with individuals who wanted to help people with diabetes by ensuring the devices they use are secure. An audience member thanked the creators of DTMoSt for doing so, but wondered who should be responsible for creating these standards? Panelists agreed that it needs to be a collaborative effort like that of the developers of these standards that included input from the FDA, NIH, DHS, NASA, FBI, NIST, US Air Force, US Army, US Navy, doctors, nurses, industry, and so on.

To address the role of government in setting cybersecurity standards, the consensus was that government should have a role in creating standards but can’t own it. Instead standard development organizations like UL and IEEE should manage the standards, and the government should play a role in mandating transparency. Many panelists noted that it is not difficult for industry to incorporate things like encryption and other security features, but they should come standard in all devices and not as a bonus feature.

The controversy surrounding cybersecurity for diabetes devices as it relates to the DIY movement was a highlight of this panel. Managing type 1 diabetes is itself very difficult and patients risk dangerous hypoglycemic episodes and DKA. Everyone wants safe and secure devices, but how do we balance cybersecurity standards with getting these patients the tools they need to live healthier lives as fast as possible? If DIYers had a system that worked for them, then they would not have to hack into their own devices. All panelists agreed that if security compromises usability then it becomes irrelevant.

Cybersecurity is a responsibility shared by device manufacturers, government, health care providers, insurance plans, and patients.

Speaker: Casper de Clercq, MS, MBA

Historically health care in the United States has been “fee for service”; we have freedom of choice but with minimal integration or continuity of care. This transactional system has effectively eliminated chronic care coordination by primary care physicians (PCPs). In some European countries we see more emphasis on outcome or value-based rather than transactional care. The UK National Health Service, for example, disburses funds to clinical commissioning groups of PCPs who coordinate care for their assigned patient populations. How might adoption of digital technology help inform and coordinate better health care?

There is tremendous opportunity to address population health through digital solutions which simplify and inform many of the tasks and decisions for those with a chronic disease. For example, the time between clinician visits can be punctuated with multiple smaller interventions and nudges. Many small clinical studies have shown digital interventions to be effective. As investors we have seen hundreds of digital health start-ups; however, we expect three significant hurdles to be addressed before seeing broad adoption in value-based care.

No population is homogenous and many digital health studies have inadvertently selected for motivated recruits. Based on commercial experiences at several start-ups we have seen dispersion of behavior clustered across three broad groups. One in three people, will manage their disease more effectively if they receive education and digital feedback from a measuring device. However, for the next third of individuals, we know that compliance is intermittent and they likely need human support and continuous encouragement. For the remaining third of patients, improving their chronic disease is often intractable to digital solutions because they may require behavioral therapy, need food and housing addressed, and so on.

Digital solutions are not yet mature enough to provide automated, no-cost, personalized guidance and encouragement to change behavior at scale. Fitbits, watches, connected monitors (like BGM) and gamification can all nudge an individual in a healthier direction but longer term adherence remains a challenge. To have a significant impact on a majority of the population including those with severe disease, coaching and technology-enabled (human) services are still required for effecting lasting change.

The most significant hurdle to the success of digital value-based care is the reimbursement landscape, particularly for diabetes. There are many promising digitally enabled services which can be provided for $50-100 per month to effect a 1% drop in HbA1c. Even in the near term, cost-effectiveness should be favorable for digital solutions since today’s therapeutics can cost $300-400 per month for the 1% drop in A1C. However, payors including Medicare have been slow to pay for digital solutions. The fundamental barrier to adoption is that payors are accustomed to large, compelling pharmacoeconomic studies for therapeutic drugs. Until digital innovators demonstrate efficacy through larger randomized clinical trials, their solutions will continue to be limited to sponsoring employers rather than insurance companies.

Moderator: Julia Hu, MBA

Panelists:

This panel addressed funding in digital health, discussing the current funding scene and critical aspects of companies that are more likely to be funded.

In the last few years, over $24 billion have been invested in digital health. 2017 itself boasted $6 billion invested, the most mega deals ever—7 fundings of over $100 million each—and an average investment of $16-17 million. Panelists were particularly excited about the scale of investments. In addition, over 750 new investors in the past three years have invested in digital health. They present an interesting opportunity; however, the panel warned that start-ups should prioritize value-added investors.

Many experienced investors will only take on companies that have delivered on promises. For example, the panel noted that while outcomes results may seem on face positive, many companies’ results are not peer-reviewed, and if they are, then the trials are not randomized. These problems reduce the credibility of the results. Furthermore, it’s difficult for consumers to identify real outcomes, so the panel emphasized the need for consumer education in the future.

Digital health companies must be rigorous in proving clinical outcomes and persistent in their pursuit of funding.

Moderator: Robert Gabbay, MD, PhD

Panelists:

The future of digital health and diabetes care is predicted to look considerably different in the coming 3 to 5 years. Data will be more expansive and individualized to provide insights into self-care, provider interventions and population management. Further development of accessible apps and portable devices will allow for instantaneous and automatic monitoring, which can aid in reducing the burden of disease. Enhanced personalized assessments of one’s own health metrics can empower people to take a more active ownership of their health data. Wireless networks will allow data to be captured in a central repository and potentially mined to enable mass personalization. Providers will have streamlined digital data presentations and analytic tools that will enable them to care for larger populations and address some of the provider shortages challenges. Researchers and analysts will have enhanced tools to understand causal mechanisms and health behaviors among the population. We are headed toward an ecosystem where a patient owns and control their data and can opt to be part of clinical research or postmarketing surveillance anytime and anywhere. Ultimately, digital advancements and the innovation of future technology will lead to more patient centered and highly personalized resources that in turn will improve the delivery of diabetes care and overall quality of life.