AI and Patient Centric Healthcare
The rapid maturation of AI technologies has driven immense interest in AI, with AI and machine learning (ML) companies receiving $3.6bn of VC investment in 2016, and healthcare has consistently been the top industry for AI deals. According to CB Insights, AI in healthcare has seen $1.8bn in funding across 270 deals since 2012 and Accenture forecasts that the market for health AI will increase from $600m in 2014 to $6.6bn in 2021. Given the massive cost of healthcare, $3.2tn in the US, the interest in AI solutions is unsurprising, especially with a looming shortage of health workers. Research has forecasted a projected shortfall of 41k-105k physicians within the US alone by 2030, with the global shortfall of health workers projected to reach 12.9m by 2035.
CB Insights has identified some of the areas of focus for venture backed healthcare AI companies including: Medical Imaging and Diagnostics, Patient Data & Risk Analytics and Drug Discovery. However, one trend I’m very interested in is the potential of AI to extend the consumerisation of healthcare by making healthcare information and insight available to end users. The current model of healthcare delivery, with GPs acting as gatekeepers to services that are generally delivered in hospitals, was a good fit for a world in which acute disease is the leading cause of death but is out of date. Today, more than half of americans have at least one chronic condition and chronic disease accounts for 7 in 10 deaths and 86% of US healthcare spend.
Source: Health Sciences Institute
Moving to a more proactive model of healthcare, which focuses on prevention rather than treatment, will be key to improving outcomes and reducing the cost of chronic disease. To get to this point presents a challenge for the existing healthcare system, particularly given the shortage of health workers. The strains on the healthcare system could be eased by consumers taking more ownership of their health but, currently, few are capable of managing their health without insight & support from medical professionals and the cost that incurs. By bringing that insight and support to users’ smartphones, AI has the potential to both significantly improve and streamline the management of chronic disease.
Chronic disease is a challenge for healthcare systems because conditions are complex, long term and linked to multiple risk factors. Treatments often involve lifestyle changes rather than interventions such as prescription drugs or surgery. Chronic disease must be managed in the course of a patient’s everyday life, not just the short time they are in front of a doctor. However, while a doctor cannot be with a patient every hour of the day, a smartphone can. The smartphone yields the potential of 24/7 monitoring of and engagement with patients. That’s is important not only to the patient, the data generated and what we can learn from it paves the way to the future of healthcare. The range of diseases that could benefit from AI powered disease management is huge, including heart disease & COPD, diabetes, auto-immune disease, allergies and obesity.
Although these diseases vary widely in terms of symptoms, there are consistent challenges for sufferers such as tracking symptoms, identifying trends & patterns and understanding how to improve their condition. To address these pain points, management apps are likely to cover a core set of capabilities including user data capture, visualisation, connectivity (to devices and third party data), AI powered insight and personalised recommendations. I think of the capabilities as a hierarchy, with each building on the level below (see figure).
The disease management hierarchy of features
Level 1 - Data capture
One of the challenges of managing chronic diseases is the scarcity of data. Doctors see patients infrequently, limiting their ability to properly track and monitor their patient’s health or understand the results of their interventions. Patients are often left to manage their symptoms as best they can and rely on rigid prescriptions (don’t eat dairy, take one tablet per day) or use their own best judgement based on how they are feeling.
WebMD Allergy data capture
A wide range of apps have been released enabling users to track their own health covering conditions such as obesity/wellness (MyFitnessPal, Lose It), diabetes (mySugr, Diabetes:M), allergies (WebMD Allergy, AlliApp), digestive health (Cara, Bowelle) and asthma (AsthmaMD, AsthmaSENSE). By capturing data over time, these apps give users a better view of how their symptoms have developed and allow users to begin to see correlations between inputs such as diet or sleep and symptoms. While capture of patient data in digital form isn’t on its own a significant benefit for patients, digitised data enables the provision of higher value services. As such, a well designed UI which makes patient data input as frictionless as possible is important to ensure the capture of clean, useable data at scale. Many chronic disease sufferers must track multiple events per day (meals, exercise, medication, symptoms etc) and any unnecessary friction in the process risks users moving back to pen and paper based tracking or not tracking at all.
Level 2 - Visualisation
One Drop data visualisation
Source: Google Play
The obvious next step up from capturing user health information is to visualise it. Users can view changes in their symptoms over time and how those changes have correlated with changes in their behaviour. Well designed visualisations can encourage data capture by assisting users to understand their health and identify factors that improve or worsen their symptoms while adding an aspect of self-discovery and satisfying curiousity. However, while visualisation can help to spot patterns in their experiences, users must still have an idea of what they are looking for, which can be difficult for complex conditions that have a range of risk factors.
AsthmaMD data visualisation
Level 3 - Connectivity
While smartphone apps can capture data from users, that data is often more useful when combined with other data sources such as data from connected devices, genomic data and environmental data. Apps which connect to devices such as fitness trackers or glucose monitors can capture data passively, without user input, reducing friction and increasing the volume of data collected. In addition, connecting with other data sets, such as genomic or environmental data allows users to analyse a broader range of inputs and spot previously hidden patterns. This connectivity is likely to be increasingly important with the use of ML based approaches which can analyse multiple data data sets to identify complex correlations and as the ecosystem of solutions providing relevant data grows.
As the ecosystem of connected devices and potential suppliers of data has increased, platforms such as Validic have emerged to simplify the integration of data from multiple sources (see figure), significantly reducing the complexity of using data from third parties or connected devices
The Validic Ecosystem
Source: The Doctor Weighs In
Level 4- Insight
The next level beyond helping users to collect and visualise their health data is analysing that data to deliver insight into their health. With sufficient data, ML can be used to identify the complex range of factors that impact users’ symptoms, identifying previously unnoticed trends and patterns. Given the lack of information currently available to sufferers of chronic disease, the prospect of an app that can give them insight into their personal triggers and risk factors has huge value to users and the broader health system. Apps at level 4 have the potential to personalise insight to every user, moving healthcare away from generalised population level insights to understand each individual patient.
In order to reach level 4 and deliver on the promise of ML, apps need large volumes of user data to build personalised models. That means that the businesses most likely to succeed are those that can encourage frequent interaction to build the most comprehensive view of users. This can raise something of a ‘chicken and egg’ problem of how to provide enough value to users to incentivise them to enter data without having the data needed to provide insight. Most likely, solutions will initially provide logging and visualisation to capture enough data to build out ML powered insight, although for some conditions, simple well understood analyses can be offered immediately to provide value while capturing the data needed for more sophisticated insight. Another alternative is to start with manual analysis of data, potentially through a coaching based premium business model, which has the added advantage of generating further data to train ML.
Level 5 - Actions
The final capability of an AI powered disease management app is moving beyond delivering personalised insight to provide personalised coaching based on their individual circumstances. Depending on the condition covered, recommendations could take the form of foods to eat or avoid, when and how much medication to take or particular exercises to complete. Moving from helping a user understand their condition to making recommendations comes with additional risk and regulatory oversight (and associated cost) but creates a 24/7 health coach specific to each user.
By learning from interactions with users apps at this level will tailor messaging to each individual user to best drive health behaviours through personalised recommendations. Recent research has shown that combining self tracking with e-coaching (and particularly personalised e-coaching) is effective in promoting a healthier lifestyle and the implications of each chronic disease sufferer receiving real-time personalised healthcare recommendations are huge, both for health outcomes and for the cost of care as automation allows the delivery of proactive preventative healthcare at scale.
Monetisation of consumer focused digital health products can be challenging as the prevalence of free apps makes users unwilling to pay. In addition, the importance of capturing data rewards businesses which can drive user engagement at scale, which is best accomplished through free apps. As such, it is likely most companies targeting the space will adopt freemium business models, with free basic features such as data capture & visualisation monetised through premium features (e.g. personalised insight or recommendations), sale of third party products or services (e.g. blood testing, connected healthcare devices), coaching services or enterprise partnerships.
What Forward Partners is looking for
Looking across the digital health landscape, I see numerous examples of products offering level 1-3 of the digital health hierarchy but not yet products offering levels 4 and 5. It may be that existing players are best placed to add these capabilities to their products but I believe that there is scope to build disruptive solutions that leverage the capabilities of ML to deliver greater value to users and achieve massive scale. To do that, products must first tackle levels 1-3 with world class UI, product design and distribution to build the user engagement and training data needed to tackle levels 4 and 5. Given the immense cost of chronic disease, there is huge potential for businesses which can leverage AI to help consumers to better manage their health and there are huge benefits to bringing patients into the centre of healthcare.
If you are building a product in this space, or plan to, we‘d love to hear from you. We’d be particularly keen to understand your plans for building a data set and how you plan to offer value to users while you are doing it.