Digital Medicine – The New Frontier for AI in Healthcare

Abstract

Traditionally, the field of medicine has largely been practiced using population-based approaches.1 This is clearly illustrated in the context of drug development for many indications that span oncology to infectious diseases and pain management, among others. For example, patient therapy is conventionally given using one-size-fits-all protocols, which virtually precludes realizing optimal objective response rates and overall survival outcomes. Emerging strategies are seeking to harness novel analytics platforms to interrogate population-based big data sets to parse out improved treatments that will still largely be used2 to treat individuals. It is important to note, however, that while it is common knowledge that substantial inter-patient variability in treatment outcomes exists, the degree of intra-patient variability at the N-of-1/single patient level, can potentially be so substantial that truly personalized healthcare will require dynamic modulation of the intervention and/or the intensity of care in order to sustain the optimization of efficacy and safety. While population-driven strategies will ultimately enhance the ability to identify effective interventions, the ability to use a single patient's data to guide the patient's own care will also be needed to effectively implement these interventions with unprecedented precision. The ability to individualize care will broadly impact multiple facets of medicine, ranging from drug-based therapy to emerging digital therapeutics platforms that are harnessing software as treatment. By harnessing digital medicine to realize N-of-1 interventions, it is possible to render patient responses that are more uniformly efficacious compared to traditional approaches. Effectively integrating these platforms into healthcare delivery workflows is expected to improve healthcare outcomes and cost effectiveness over traditional approaches. Among the many areas that can benefit from AI/digital medicine, N-of-1 medicine and digital therapeutics have tremendous potential to improve outcomes and reduce healthcare cost inflation. These advances have been bridged with substantial growth in the private sector to scale AI and digital medicine deployment into practice. According to Galen Growth Asia and Rock Health, two digital health analytics platforms, US$5.0B was invested in Asia and US$7.4B was invested in the US in 2019 demonstrating clear investor enthusiasm for investing in the future of healthcare.3 This special issue discusses key areas where digital medicine and AI are poised to markedly individualize intervention and enhance treatment outcomes, citing examples from the field, and provide examples of recent advances that will serve as a gateway towards next-generation, AI-enhanced digital medicine. The promise of personalized medicine will ultimately reduce the inter-patient and intra-patient variability in treatment outcomes that are observed, which are the key drivers of sub-optimal efficacy and safety that is observed across the stages of early drug development, clinical trials, and the postapproval point of care.4 Unfortunately, implementing the entire workflow of personalizing care is often incomplete, as there are many considerations that must be considered in this domain, ranging from how interventional trials are designed, to the strategies employed to maximize the number of patients that initially and continue to respond positively to treatment. One major challenge in developing novel drug treatments is the ability to sufficiently interrogate the space created by the range of possible drugs for a given indication. This is a challenge that confronts all aspects of treatment design, regardless of indication. Therefore, it impacts everything from oncology to metabolic diseases and metabolic diseases and beyond. In a recent study, an orthogonal array composite design (OACD)-driven set of drug combinations was subsequently used to optimize the identification of multi-drug regimens. This study confirmed the minimum number of validation experiments needed to pinpoint the drugs that resulted in maximum efficacy and safety, which will play a key role in accelerating the selection of drug candidates for N-of-1 therapy (1900122). The impact of these and other approaches was discussed in detail in a recent report that described the use of AI to design powerful combinations for the treatment of tuberculosis (TB). Importantly, non-obvious combinations were shown to outperform the standard-of-care combinations. Furthermore, AI was successfully used to re-optimize the drug dosing required to eliminate TB infection in preclinical models. Of note, these studies markedly reduced the number of preclinical experiments needed to realize these substantially improved outcomes (1900086). To demonstrate the role of AI in boosting outcomes at the point-of-care, AI-based analysis in rats with colorectal cancer showed that drug interactions in a 4-drug combination were dynamic, varying between synergy and antagonism over time. Even though the rats were genetically identical, their responses to drug treatment were highly varied from one another, and N-of-1 dosage guidance was needed in order to streamline the reduction of tumor burden across all subjects. Most importantly, Nof- 1 dosage guidance, which varied over time, was needed in order to mediate any apparent efficacy for most of the subjects. This means that efficacy can be achieved with drugs that do not appear to be efficacious when arbitrarily dosed. Furthermore, this study showed that the incidence of treatment response could be improved across all study subjects (1900127). Among the many disorders that will continue to increase in global incidence while presenting substantial challenges to the medical community, liver cancer, liver transplant, and fatty liver disease represent indications that can be addressed by AI (1900167). A recent report detailed the molecular alterations that can be used for regimen selection for disorders such as cholangiocarcinoma, the development of novel biomarkers for immunosuppression in transplantation, as well as AI-based approaches (e.g., neural networks, big/small data analytics) that can be used to prolong the optimal delivery of the selected therapies. On a global scale, non-alcoholic fatty liver disease (NAFLD) is rapidly increasing in incidence, and will be a major cause of liver cancer. Leveraging AI for drug selection and simultaneous consideration of the factors in the gut microbiota that can influence drug efficacy and safety will play a key role in realizing actionable and optimized treatment strategies for NAFLD. In addition to identifying the right drugs and doses that should be used for combination therapy, a prevalent challenge in clinical care includes the sequence of how these drugs are administered. In a recent study, an important step forward to address this barrier was demonstrated using an approach called Innovative Design and Active Learning (IDEAL). IDEAL is based on the use of permutation orthogonal arrays, which decreases the number of experiments needed to sufficiently assess 3-,4-, and 5-drug combinations (1900135). Of note, following the identification of the drug combinations that resulted in optimal cytotoxicity towards Raji lymphoma cells, it was shown that sequential drug treatment outperformed simultaneous drug treatment. However, simultaneous treatment resulted in prolonged efficacy compared to sequential delivery. This work represents an important advance to further improve the efficacy and clinical intervention guidelines of AI-identified drug regimens. With regards to the clinical/in-human validation of interventional AI, a recent study demonstrated the use of AI to markedly reduce the dosage of TDF, a drug that has been associated with substantial side effects including kidney failure in patients being treated for of human immunodeficiency virus (HIV). Specifically, patients were calibrated using a parabolic response surface (PRS)-based AI platform by being given varying doses of TDF while measuring corresponding viral load. Based on this approach, patient-specific PRS profiles successfully identified an average TDF dose reduction of ≈30%, which resulted in undetectable disease (1900114). Digital medicine can also be leveraged to bridge diagnostics and therapy to address gut health (1900125). In the domain of diagnostics, a recent report detailed how robotics in the gut could be harnessed for applications that include capsule-based endoscopy paired with neural network-based software for improved image recognition, capsule endoscopes for pH sensing to monitor reflux disease, and wireless capsules for gastrointestinal bleeding detection, among others. Emerging ingestible technologies are also being explored for sustained drug delivery, which can then be potentially further leveraged for improved adherence. In addition to therapy development, biomarker identification will play a key role in monitoring treatment outcomes. Recent work is exploring how deep learning-based AI platforms can be paired with approaches such as mass spectrometry and collectively leveraged to interrogate large parameter spaces to enhance the actionability of biomarker signatures. This and other studies will empower the entire spectrum of drug development and personalized medicine to ultimately improve outcomes for patients compared to traditional approaches (1900163). A key downstream milestone for these studies will include the demonstration that they can be validated towards eventual drug approvals with clear demonstrations in improved outcomes in the clinic, development pipeline, and healthcare systems compared to current models. Successfully bridging digital medicine with pharmaceutical innovation has been challenging and has faced limitations. The perception of the areas in which digital platforms can reduce the cost of drug innovation has thus far been limited. In addition to mediating improved therapeutic outcomes, these studies represent examples of how digital medicine should be leveraged to impact the economics of drug development and may eventually go on to impact the cost of everything from drug pricing through general healthcare, especially since these platforms can be implemented across the entire continuum of intervention. Modulating treatment using indices comprised of efficacy and safety, in some oncologic indications, already represents a practice-changing shift. Markedly accelerated drug combination validation, increased numbers of responders to treatment, reduced long-term complications, among other endpoints that have been observed through the aforementioned studies signal the potential for strong alignment between digital medicine and unprecedented capabilities in cost-effective optimized and personalized healthcare. Digital therapeutics (DTx) are evidence-based interventions delivered as software, aimed at preventing, managing or treating a condition, sometimes in combination with traditional drug-based approaches. This rapidly growing domain sits at the intersection of healthcare and technology. As an example, a recent study demonstrated the use of CURATE.AI, an indication-agnostic platform for digital therapeutics applications. In this study, dynamic modulation of training intensity on a multi-tasking software platform revealed major differences between study participants in areas such as performance improvement under high and low intensity training. These N-of-1 profiles may serve as a way to optimize the efficacy of digital therapy for a broad array of indications (Figure 1).5 The promise of digital therapy is already being explored in commercial settings and deployment in established healthcare systems. Evidation Health recently showed that data collected from commercially available wearable devices can be utilized to create indices of sleep quality, heart rate, and mobility that are significantly different in cohorts with chronic conditions such as multiple sclerosis and type 2 diabetes,6 as compared to age/gender matched controls. In other studies, Evidation used data from wearable devices to quantify the burden on activities of daily living related to acute conditions that are often invisible to the healthcare system such as viral infections,7 and demonstrated significant associations between adherence to medications for chronic conditions and wearable device usage patterns.8 Evidation recently collaborated with Apple and Eli Lilly to assess the feasibility of using consumer-grade smart devices (e.g., tablets, smartphones, smartwatches and sleep monitors) to remotely and unobtrusively monitor cognitive impairment symptoms.9 Throughout the 12-week study period, Evidation collected a total of more than 16 TB of data. Machine learning models revealed that patterns of phone usage, keyboard interactions, and daily routinely behavior were most relevant in differentiating between elderly adults with (n = 31) and without (n = 82) a diagnosis of cognitive impairment, demonstrating the promise of using data collected from consumer-grade devices to unobtrusively measure cognitive impairment in everyday life. Digital platforms have also been deployed effectively in the field of mental health. According to the latest GBD study, depressive and anxiety disorders were the leading causes of disability amongst all mental disorders in the US. Both were amongst the top 10 causes of disability overall, with depression 5th and anxiety 7th in the GBD study. Despite these high numbers, less than half of the over 46 million adults with a mental health disorder were seen by a mental health service in 2017. Mental health conditions have been recently addressed using digital platforms for cognitive behavioral therapy, delivered via the internet (iCBT). A clinically-validated platform developed by SilverCloud Health is currently deployed across the UK National Health Service, Europe, USA and Canada and has successfully treated over 300000 patients.10 Unlike traditional modes of face-to-face therapy, these programs are digital and mobile-first, asynchronously delivered and have enabled 24/7 access beyond the allotted duration of face-to-face CBT therapy, which is usually only 8 weeks in duration. They also increase patient access by embedding seamlessly into existing health care system workflows and electronic medical record systems. A recent SilverCloud white paper reported a marked reduction in depression and anxiety symptoms across the range of acuity (Figure 2).11 This white paper also reported reduced patient visits to the emergency department for mental health issues by 5%.11 The enormous potential of this field has inspired a number of efforts and associated regulatory innovation to improve mental health and associated indications with software-driven solutions. Solutions developed by SilverCloud and others that deliver clinically validated outcomes at scale with technology (versus clinical labor), can potentially bring down the cost of care and impact healthcare economics in general. This will have “knock on” positive externalities on managing other conditions and comorbidities, thereby leading to better outcomes and lower overall costs of population health management. Other digital therapeutic applications include Pear Therapeutics (Substance Use Disorder), Akili Interactive (Pediatric Attention Deficit and Hyperactivity Disorder, Adult Major Depressive Disorder,12 and Acceset (Digital Para-counseling for Mental Health). Digital delivery will effectively reduce the burden on health care systems and translate to changes at the individual, patient group, and societal level. That said, there are also current limitations to the universal deployment of DTx. For example, healthcare delivery systems are set up to receive reimbursement via traditional methods, even if costlier. If DTx is not well integrated into existing workflows and payment flows, the risk is that it sits out in the cold and may fail to get adoption. Digital medicine that is driven by big and small data AI platforms, many of which are highlighted in this editorial, is making the prospect of next generation N-of-1 medicine and truly personalized healthcare a reality for both drug and software-based intervention. These advances are forming a foundation for the potential prospect of digital-first and digital-only therapeutics complementing digitaldeveloped drugs as technology-optimized, first lines of treatment. The fruition of this vision will broadly impact patient compliance, treatment outcomes, and healthcare economics and policy to ultimately change way that technology is harnessed to develop, deliver, and pay for/cover optimized healthcare. D.H. is an inventor of patents pertaining to AI-based therapy and CURATE.AI. G.T. holds equity in SilverCloud and Evidation Health via B Capital's prior investments. Dr. Dean Ho is Provost's Chair Professor in the Departments of Biomedical Engineering and Pharmacology, Director of the N.1 Institute for Health (N.1), Director of the Institute for Digital Medicine (WisDM), and Head of the Department of Biomedical Engineering at the National University of Singapore. Dr. Ho's team is spearheading multiple prospective/interventional clinical trials at the interface of artificial intelligence/digital medicine and disease indications including oncology, infectious diseases, digital therapeutics, and beyond. Image credit: National University of Singapore Gavin Teo is Co-founder and General Partner of Straits Venture Capital, an early stage VC fund investing in Southeast Asia technology and healthcare companies. Gavin is also a guest Lecturer at National University of Singapore and an Advisor to B Capital Group. Prior to co-founding Straits VC, Gavin was a General Partner at B Capital Group, where he led the firm's investment committee and global healthcare practice. Gavin earned his MBA from Wharton.