Towards a systems approach for chronic diseases, based on health state modeling

Modernization of diagnosis, and personalization of therapy

In discussions on the “valley of death” challenge in health innovation (Butler D, 2008; Wehling, 2009), which concerns the problem of translating scientific and technical advances into impact at the level of patient outcomes, beyond time-limited clinical studies, oncology is often mentioned as an area of medicine in which there has been relatively good progress in terms of such translation into regular practice. In this medical specialty, many advances in our increasing scientific understanding of the molecular basis of disease, and patient heterogeneity, have been translated into solutions that benefit patients with specific tumor profiles. Looking across different areas in Oncology, the most successful paradigms that evolved effectively couple the modernization of diagnostics (i.e. the ability to determine tumor subtype based on its biological profile) with the use of targeted therapies (which have been designed for a specific tumor type, or a set of tumor types, with a characteristic biological profile). This personalized health paradigm emphasizes understanding of patient heterogeneity at the level of biological profiles, because it was possible to link diagnostic capability at the level of tumor-derived DNA with its interpretation in terms of the biology that drives the growth and survival of that type of tumor, resulting in impressive improvements of patient outcomes in many tumors (where both tools converge). However, this paradigm has also raised economic sustainability concerns, as tension increases between stakeholders who a) invest in the development of solutions based on this paradigm, and b) those who need to pay for healthcare of tumor patient populations, which are increasingly segmented, with many segments associated with relatively high costs (see above, and Koelsch et al., 2013).

In this renowned area of medicine, many innovations based on this paradigm have advanced quite far in the innovation translation pipeline, leading to practical solutions for global deployment, reimbursement in different healthcare systems, education of healthcare providers, and integration into regular care processes. Considering the effort that is required for such a level of system-wide change in the real world, those successes are indeed quite impressive, keeping in mind, however, that there are many areas of medical need that remain a considerable challenge in oncology, including the phenomenon of tumor recurrence despite short or mid-term effects of targeted therapies.

If we consider to extend this paradigm to other areas of medicine, we need to take into account that tumors have many characteristics that are fundamentally different from many common chronic diseases, complicating the application of exact copies of the approach in non-Oncology areas, apart from some exceptions, such as diseases with a strong genetic component (which tend to be rare). Therefore, we need to learn how to consider the particular characteristics of a disease, at a diagnostic and therapeutic level, as we extend the oncology paradigm of personalized health to other indications. This challenge, so far, has been hard to crack, triggering a ‘lessons learned’ discourse that can be quite healthy in the context of a possible adoption of the proposed platform. Understanding the very slow progression of many common chronic diseases, from different angles, is part of the scientific challenge, as outlined below.

Another area of medicine that has witnessed much progress in terms of developing a modern approach to the convergence of innovation in diagnosis and therapy is AIDS. Being an infectious disease makes it a case that is quite different from oncology and most chronic diseases, although aspects of the oncology paradigm of targeted therapy and personalization have been re-used here as well. Once the AIDS epidemic was recognized as a major health challenge, relatively fast progress on understanding the key characteristics of viral populations, and the biology of their interactions with host (defense) biology, has enabled the development of highly personalized combination therapy approaches, depending on the DNA level composition of the viral population in that specific patient, at a particular point in time (Lengauer & Sing, 2006; Lengauer et al., 2014). As in oncology, much of the progress in this area of medicine was catalyzed by technical progress. For example, easier access to relevant omics technology (see below), enables faster, easier and better diagnosis of the state of virus populations, as a basis of therapy personalization. Campaigns for collaborative multi-stakeholder, interdisciplinary solutions for battling this infectious disease have also played an important role in contributing to the relatively fast impact on patient outcomes, although challenges remain, e.g. related to the high costs of many years of combination therapy close to the ‘cutting edge’ of molecular medicine. Interestingly, note that both areas (oncology and AIDS) have increasingly moved away from the use of single therapies, to a more sophisticated, cutting-edge combination therapy approach that involves the early recognition of disease recurrence. As a consequence, I propose a connected set of successful paradigms from oncology and ADIS as pillars of the platform described below, recognizing that much needs to be learned on how to apply those paradigms to chronic diseases with a more limited genetic contribution. In that context, a key question will be to find out where the most meaningful, interpretable and actionable diagnostic signals are, to guide our choice of interventions, based on the patient profile, at a particular stage in disease progression.

Capturing the state of biological systems

Our ability to study, measure and understand complex biological systems has increased with many new tools and methods - although that doesn’t mean that it is easy to put the many different pieces of the puzzle together, in our mind, or in computational models. It is certainly more complex than ‘fixing a radio’ (Lazebnik, 2002), although the author’s thoughtful points about unresolved issues in the biomedical research approach, including the lack of a formal language that helps communities to connect the pieces in such systems, were indeed very helpful, and influential. Enabling technologies in that area includes a maturation of our ability to capture states of biological systems at a more comprehensive level, using genome-wide technologies (or simply ‘omics’). Such omics technologies now exist for many different levels of biological systems, including DNA, variants for RNA, protein and metabolite-level system dynamics (i.e. genomics, transcriptomics, proteomics, and metabolomics; Spivey, 2004). Depending on the sample we take and how we process it, omics technologies can generate very rich datasets about the ‘expression state’ of thousands of molecules in those systems (that are represented by the samples that were taken). However, there are many complex problems in data generation and interpretation as well. Inferring overall ‘health states’ (see below) from such measurements is possible, but non-trivial, and, at present, still resource intensive (Chen et al., 2012).

Around the year 2000, at about the same time as the hype on the sequencing of the human genome and its ability to revolutionize medicine, there was also much excitement on the promise of such omics technologies (Spivey, 2004), leading to thousands of publications with datasets based on human and non-human samples (e.g. from species that are commonly used as preclinical models of human disease). However, most of those datasets represent ‘snapshots’ in time, with unclear positioning in terms of disease progression states, exact cellular composition, and other ‘metadata’ that would help with interpretation and comparison. Now that the first wave of excitement has given way to a second wave that aims to build on lessons learned from the first omics wave, there is increasing awareness about the importance of understanding disease progression, beyond ‘snapshots’ with limited ‘annotation’. This trend is likely to be enabling for the proposed approach, as it helps to connect ‘health states’ in time, with biology, at a comprehensive level. Looking back at how we handled omics waves could also be tremendously helpful in designing guiding principles for handling technology hypes in general.

Engineering of patient-centric connected health solutions

Technical advances in a variety of areas, from mobile technology and the widespread use of smartphones, to health-related sensors, machine learning and digitalization of healthcare, are increasingly producing ‘real world’ impact based on convergence between different technology fields, beyond exciting prototypes, in chronic diseases (Dobkin & Dorsch, 2011; Kvedar et al., 2016). While the more difficult-to-change and highly regulated healthcare and health innovation sectors are expected to develop more slowly compared to less-regulated industries, e.g. those that can improve products quickly based on consumer-centric feedback loops, there are emerging paradigms with reusability potential. Patient-facing solutions with interfaces for other stakeholders, including healthcare providers, are one of the fastest-moving areas here.

For example, in respiratory diseases, such solutions have connected improved therapy (e.g. new COPD and asthma drugs) with ‘real world’ data on patient outcomes collected using mobile technology around ‘smart inhaler’ devices for those drugs, alongside with patient-centric views on smartphones, and the involvement of healthcare providers or clinical trial teams (Bender et al., 2017; Clift, 2016; Perez, 2015). This smart inhaler paradigm for designing “beyond the pill” solutions appears to provide value to multiple stakeholders, as a) the patients get better feedback on how they are doing with the inhaler-based therapy, including the aim to prevent stressful exacerbations, b) healthcare providers have more data to optimize care pathways, c) the developers of relevant drugs get more information on ‘real life’ settings and problems, enabling faster learning, while d) device developers get better feedback on how to optimize their devices in terms of usability, functionality and other health impacts, and how to connect the engineered systems with other components. Note that the ability to generate such value close to patients’ homes, outside classic healthcare settings (e.g. hospitals), is a factor driving excitement in the digital health sector, which has identified the management of chronic diseases as a key challenge and opportunity (for a more comprehensive overview, see Kvedar et al., 2016).

In the context of the proposed platform (below), this and similar patient-centric paradigms fill an important void in the current healthcare and health innovation landscape, as they a) add low cost solutions closer to patients, in their natural environments, minimizing travel to clinics, b) have the potential to contribute diagnostic signals, and c) improve the ability to connect system components, across stakeholders, enabling a more data-driven approach to system-level learning.

Measuring morbidity and disease burden, more globally

Improvements related to the direct and indirect effects of chronic disease morbidity (and improvements in terms of healthy aging) at population level need to be monitored in a reliable manner, to enable learning based on the impact achieved by different types of candidate solutions. The better we can measure impact, the more efficient our learning process. This needs to be based on a trusted methodology that works in a variety of settings, in different countries, to allow fair comparisons. Recent advances in this area include the “global burden of disease” (GBD) methodology developed by IHME (Institute for Health Metrics and Evaluation), and the framework developed by ICHOM, which establishes a first version of a system for capturing relevant impact. In terms of biomedical innovation, and the key role of clinical studies in validating specific hypotheses in human populations, we can now capture a diversity of patient outcomes, including QoL (Guyatt et al., 1993; Norman et al., 2003) and health-related functions such as mobility (an emerging area enabled by sensors that record different kinds of movement patterns, e.g. accelerometry). At the economic sustainability level, measures such as QALY (QoL-adjusted life years) have added the highly debated ability to differentiate among years of life extension with high and low QoL when judging value provided by innovation (Shiroiwa et al., 2010). As a consequence, we now have a basic arsenal of tools to monitor the short and long-term results of solutions we develop, at different levels, in clinical trials, in regular clinical care as well as outside clinical settings. With this, we do have an improved ability to evaluate the impact of system-wide solutions that better connect the fragments, e.g. across the successful paradigms described in this article. However, this does not mean that the current toolbox for measuring outcomes across different settings is perfect and needs no further optimization. It is a very complex topic that will certainly require more innovation and adjustments down the road. At the same time, we can start to pragmatically use what we already have. In that context, multi-morbidity, in a landscape of increasing chronic disease burden, is one of the areas that may benefit from increased attention, with regards to the capture of impact, as well as the combination of several paradigms in relevant solutions.

Open science culture

An intensive, open debate on the best approach for a particular problem, and open access to data that can help to select among alternative approaches, are features associated with good science. Since the successful paradigm of open source software in informatics has infected an increasing number of areas related to health innovation, including bioinformatics, the screening of chemical libraries, as well as the generation of tools for research on new kinds of drug targets, an intense debate has developed on the role for an extension of those fragmented experiences in particular areas of science into a more comprehensive, interdisciplinary ‘open science’ approach that includes innovative models for enabling a faster translation of research results to patients (Laverty et al., 2015; Low et al., 2016; Munos, 2010). Indications in which innovative approaches at a similar level of complexity were tested ‘end-to-end’ have added further fuel to the debate, with the malaria field taking a prominent position among those translational pioneers (Wells et al., 2016). Relevant aspects include a) the openness of raw data, code and algorithms (avoiding ‘black box’ solutions), which applies to computational as well as experimental protocols, b) ‘reproducible research’ for enhanced transparency and reproducibility between research groups, c) sharing of data, insights, knowledge, and tools, based on initiatives that provide some structure to data sharing (e.g. Dataverse ). Bernard Munos (e.g. Munos, 2010) has proposed that the increasing adoption of such approaches is linked with sustainability in health innovation, considering biomedical complexity. A recent conference in Oxford (“Drug discovery: creating a new ecosytem”, 2–3 June 2016) has been able to gather a variety of pioneers in that area. However, while there are many experiences in this growing global community that can help to select a particular open science model for a given medical problem, we are still in the earlier stages of that learning process, in terms of tackling the rather difficult ‘valley of death’ problem in translation towards patients.

Dealing with real world evidence

The digitalization of healthcare, as well as technology convergence in non-healthcare areas, are resulting in an increasingly diverse and fragmented landscape of data related to different aspects of health, from hospital records, to claims for reimbursement, to fitness device data, and data produced by patient-centric solutions for chronic diseases as outlined above (Strategy & report, 2015). Such ‘real world evidence’ is often contrasted with data generated in controlled clinical studies, such as randomized clinical trials that test specific medical hypotheses. Important differences exist, for example, in our ability to make conclusions from either data, based on solid statistics, methodology and theory. In that context, it can be helpful to develop improved capabilities for dealing with real world evidence in a way that is consistent with ‘good science’ principles, in collaboration with disciplines that have a history in that area, e.g. epidemiology and HEOR (health economy and outcomes research). The example of the “Global Burden of Disease” study of IHME (Lozano et al., 2012), which succeeded in integrating thousands of different real world evidence data sources into an over-arching model that enables a number of analyses, could be helpful in that context.

Business model innovation

Considering the important role of economy in the health care/innovation sustainability discussion (with an assumption of limited resources), we can observe that the classic economic model of reimbursement for healthcare actions based on a “fee-for-service” concept, is gradually being replaced by a potentially more sustainable “value-based care” model (EFPIA, 2016). This model is based on the following principles: (i) coordinating around patients all the elements of the care continuum; (ii) shared commitment of all healthcare system players to the outcomes that matter to patients; (iii) generating and tracking data on those outcomes; (iv) benchmarking performance transparently for informing management decisions; and (v) paying for outcomes rather than for inputs and processes. However, this is a complex structural change and therefore unlikely to be an easy transition, so it may take decades before this transformation reaches all aspects of health-related systems globally (with progress being tracked by programs such as the Pharmaceutical Outcomes Research & Policy Program at University of Washington). In the meantime, pioneering institutions around the world are moving from pilots to organizational change, to become a leader in this important transition, increasing their fitness for the future at an earlier time point, when changes are still easier to manage and resource. One of the exciting opportunities in this area could be an improved ability to better align incentives across stakeholders, based on the above “value-based care” principles, as a basis for more collaborative solution development.

Alzheimer’s disease

Innovation related to this neurological disease is an interesting indication in the context of the proposed systems approach, and platform design, for several reasons:

An interesting side effect of the history of disappointing or failed clinical trials in this indication is that it forces the biomedical research community to reconsider their approach and collaborative paradigms for the sake of patients, causing a healthy discussion on advancing the biomedical research community culture, with possible effects in other indications as well, where reusable learnings are generated. For example, this field has suffered from a strong bias for a small set of hypotheses and paradigms, based on particular types of evidence, as a basis for designing and developing therapeutic solutions. The excitement around those hypotheses has resulted in lackluster discussion of things that didn’t fit this ‘group think’, including alternative hypotheses and solutions. Compared to the discourse in this indication in the 1990s, we can now notice an increasing readiness to learn from this experience.

This neurological disease is a typical chronic disease in the pathobiology sense I discuss below, i.e. there is slowly accumulating tissue damage outpacing regenerative mechanisms, which results in a progressive decline of tissue functions, which then show up as increasingly severe clinical symptoms and patient outcomes are impacted over time. The need to better recognize early stages of disease (Braak & Del Tredici, 2015), together with innovation in terms of interventions that target the pathobiology of exactly those stages, is now widely seen as the most promising approach for enabling translational progress in the field. This is likely to lead to sophisticated methods for combining diagnostic signals across many system levels, including the cognitive, molecular, imaging and other aspects described above, requiring a platform for improving public reference versions of relevant computational models.

At the same time, there has been good progress in areas related to digital health, e.g. in the early detection of a possible cognitive decline using speech patterns (Morris & Lundell, 2003), which may add a cheap and easy-to-deploy screening method to the ‘early stage’ diagnostics innovation. The Alzheimer’s exemplar may also help in exploring connections between the various aspects of the proposed systems approach and platform. Below, I will propose a road map for developing such a platform, including starting points derived from Alzheimer’s disease.

Increasing readiness to design innovative clinical studies is also visible in this indication, related to particular subpopulations and health states, e.g. the APOE4 subpopulation, which carries a higher risk of fast progression towards more severe health states, compared to sporadic cases without such genetic risk factors (Mahley et al., 2009). This may help to close important gaps in the data landscape that prevent progress.

Regenerative medicine, and chronic diseases

Tissue and organ regeneration principles. We know that, based on knowledge accumulated in scientific fields related to regenerative biology and medicine, a) many animals have an amazing ability to regenerate tissues, organs and limbs (after injury or other damage), and therefore recover in terms of function (i.e. health); and b) that there is considerable evolutionary conservation at the level of the involved biology between humans and non-human vertebrate animals with high regenerative capacity. Discoveries in the past decades in that area have nurtured the hope that a deeper understanding of biomolecular systems involved in tissue, organ and limb regeneration will lead to the development of improved therapeutic and diagnostic solutions for areas of medical need, in which improved regeneration could contribute to better outcomes. In many common chronic diseases we can, during the progression of those diseases in the patient over time, observe a slowly progressing imbalance between accumulating tissue damage, and regenerative mechanisms activated in that tissue as a response to this accumulating damage. I will illustrate this principle using several examples, below, and, in the spirit of the proposed ‘systems approach’ to chronic diseases, highlight connected aspects from biology, medicine and economy that may enable the development of better solutions.

Chronic liver disease. The liver is a very important organ, contributing to overall health with its many functions related to homeostasis (the ability to keep us within a healthy range, regarding physiological parameters). Depending on our lifestyle, and other factors, such as our genetic profile, liver tissue can be increasingly damaged by different factors, including high consumption levels of alcohol, and an unbalanced (Western) diet (leading to steatohepatitis). This then leads to reduced liver function, with an impact on our body’s ability to maintain homeostasis, and therefore health. On the other hand, the liver is also known for its regenerative capacity, a notion that was further reinforced by more recent observations of liver regeneration after the application of antiviral therapies (Zois et al., 2008). In the early stages of such slowly (and silently) accumulating organ damage, the liver may still be able to deal rather well with the repeated insults, and maintain most of the important functions, and therefore overall health. Over time, however, the accumulating damage outweighs the ability to regenerate and maintain organ function, shifting the system towards an unhealthier balance between damage and regeneration. As liver damage increases, and liver function decreases, first clinical symptoms may appear that are often ignored, for a variety of reasons. With a diagnosis of ‘late stage liver disease with advanced liver cirrhosis and portal hypertension’ by a relevant medical specialist (i.e. a hepatologist), a comprehensive reaction of the healthcare system (i.e. a care pathway) with diagnostic and therapeutic aspects is triggered, based on medical guidelines and current understanding of disease progression-related risks). While some patients’ lives can be extended through liver transplantation, those who die from liver disease on the transplantation waiting list, waiting for such relief, is unfortunately rising. This adds fuel to a discussion on the need for developing new solutions for this growing medical problem. What we know so far about the different stages of chronic liver disease in such patients can be summarized in the following simplified disease progression model, which includes a heterogeneous mix of causal factors involved in creating the liver tissue damage:

Here, each health state is a stage in disease progression that is characterized by a combination of features that can affect diagnosis, assigning a particular patient to an earlier or later stage in the disease progression (with many consequences related to the clinical management or further diagnostic monitoring of the patient). Based on current knowledge, it seems that, despite the heterogeneity of causal factors involved in creating the relevant tissue damage, there is a clear response pattern of this organ, with limited variation. In other words, once the balance between regeneration and damage has passed a certain level, a rather fixed pattern of progression towards later stages is observed. Some variation is noticed, however, between patients in the time spent between such stages, i.e. we can distinguish ‘fast’ or ‘slow’ progression relative to average times, between stages. As such chronic liver disease progression stages have multiple links with the overall health of the patient, in the context of the proposed health state modeling framework they can help to define health states, considering that the same patient may also display other co-morbidities, including other chronic diseases with their own progression stage.

In a similar manner, other chronic diseases result in slowly accumulating tissue damage over the years, until homeostasis and tissue function is affected to a degree that it becomes very difficult to get the patient back to a healthy state with a high QoL. Therefore, the liver disease example may help us define relevant paradigms for understanding health states that consider co-morbidity. Unfortunately, such accumulating tissue damage is currently not easy to detect in earlier stages, in real world settings that require a high standard of patient comfort, low cost, ease of use and risk mitigation. For example, repeated invasive sampling of biopsies for assessing the condition of liver tissue in the progression is usually not practiced due to clinical risks associated with biopsy generation. Therefore, we need to develop new solutions that will increase our ability to recognize not only health states and progression stages that have a clear clinical profile, based on currently available tools in the healthcare system, but also help us understand and recognize the health states that are in between very healthy states and the easier-to-recognize advanced stages of chronic diseases. This is a general principle and a grand challenge that no single discipline or stakeholder can fully address on their own.

With this context, what are some of the more interesting interdisciplinary connections involving biology, medicine and economics in developing such new solutions?

Skin ulcers. The above discussion on chronic liver disease implies a fundamental challenge shared by other chronic diseases, i.e. that we are dealing with slow dynamics in the recognizable transitions between distinct health states. This means that our iterative learning cycle between designing a study, implementing it, sharing the results and designing the next study, combined with the time needed to observe sufficient change between health states, leads us to rather long studies that stretch many years. Therefore, fast learning based on short iterative cycles is difficult, apart from problems that can be addressed in shorter timescales. Together with an overall tendency towards short-term approaches including funding, this means that progress on understanding the above problems, including health state refinement, will be hard to accelerate.

With this in mind, let us explore the possibility of finding complementary medical problems related to a) similar interdisciplinary complexity of chronic diseases and b) learning related to health states, c) which would allow us to develop a fast-learning, collaborative network on top of short iterative study cycles and d) a systems approach that facilitates such interdisciplinary exchange. Our example for such a medical problem is again related to the principle of the fateful balance between slowly accumulating tissue damage, and scientific questions in the field of regenerative medicine. When our skin tissues are healthy and have a regenerative capacity within the normal range, we have all experienced how superficially visible wounds usually heal within a few weeks or even days. Once we look a bit closer at this area of tissue regeneration, we can notice a variation in terms of the speed and quality of healing, depending on a variety of factors, such as wound size, shape, depth, use of dressings to promote healing and prevent infection, infection management and so on. In addition, we may have heard about bad outcomes related to wound infection that led to amputations. Based on that common experience, most of us are not used to think of skin wounds as a major medical challenge in the context of the chronic disease challenge. However, if we look even closer, we find that there are many patients with one or more chronic diseases who have considerable problems due to disturbed healing, with surprisingly harsh outcomes linked with how wounds were managed (Hunt et al., 2011; Park et al., 2013). But what is the link between the chronic disease challenge, and this medical problem? And how does it relate to our discussion on systems approaches?

Diabetes complications: Patients with diabetes, in later stages of disease progression, when slowly accumulating tissue damage has reached an advanced stage, may have to deal with a variety of clinical complications, affecting organs such as the eyes, kidney and foot. Complications of the foot typically present themselves clinically as ‘diabetic foot ulcer’, a type of non-healing, chronic skin wound, to a well-trained expert, such as a specialized wound nurse or physician (Driver et al., 2010; Lim et al., 2017). Unfortunately, many patients carry such dangerous wounds for too long, and therefore have to face unfavorable outcomes, when it’s too late to manage the problem with currently available tools. Considering the progress we have made in terms of care coordination and clinical innovation in diabetes, this is one of the remaining problems related to diabetic complications.

Other chronic diseases that affect skin regeneration: To further complicate matters, other chronic diseases also have an effect on such tissue regeneration in the skin after wounding, including venous disease (i.e. leading to ‘venous leg ulcers’) (Margolis et al., 2002). Proper regeneration of the skin with a full restoration of tissue function (i.e. avoiding a scar with reduced function) requires many cells to do the right thing at the right time in the right context. Once the damage has occurred, there is a wave of signals going through the tissue that triggers that complex and dynamic regenerative response by many cells, including resident cells that go through all kinds of changes, as well as invading cells from the immune systems that arrive on the scene. As a result, a detailed molecular understanding of such skin regeneration is rather difficult, complicating efforts to develop new solutions based on that knowledge.

Towards systems approaches: I mentioned that many chronic diseases, such as liver disease, are difficult to study in terms of disease progression, because of the long timespans involved, which slow down the data-driven, iterative learning cycle. With regards to skin regeneration problems in the context of diabetes and other chronic diseases, the situation is a bit different, because a) changes related to outcomes can be measured in weeks and months, rather than years, b) the fluid produced by open wounds enables omics-type profiling close to the biology of tissue regeneration vs. damage, and c) the most affected tissue is relatively accessible, or easy to monitor, compared to tissues located further inside the body. The combination of those aspects could allow a fast-learning systems approach that combines the paradigms described above. This could then allow connections to be made between:

Building a capability for fast learning based on short iterative cycles that enable data-driven approaches, including machine learning and expert-based learning, in the context of an interdisciplinary collaborative network, therefore seems an attractive opportunity in this area of medical need.

Design principles guiding the development of the platform

I have mentioned the need to a) modernize diagnostics, extending the paradigms developed in leading areas of medical innovation, such as oncology and AIDS; b) connect better across system components that cross disciplines, e.g. medicine, biology and economics of health, e.g. in the context of more patient-centric connected health solutions; c) measure the impact of new PM solutions at that level, in a way that reflects the most relevant outcomes, enabling feedback loops that facilitate faster learning at systems level. But how can we best develop an enabling platform for such systems approaches based on those paradigms, which enables community-based learning, using open science principles, as well as feedback loops that involve real world evidence? Let us start with the center of the proposed platform, the health states, and their computational modeling across medicine, biology and economics.

Based on the knowledge we have accumulated in terms of disease progression in Alzheimer’s disease and chronic liver disease, I propose to aggregate the medical, biological and economic knowledge across these two indications in a way that allows the extraction of computational and theoretical platform components, as described below (with an eye on later reusability). In a next step, we would test the extension of such a two-indication platform to additional indications (including wound states reflecting skin regeneration), before we would explore the development of an even more comprehensive platform that captures all frequent chronic diseases, their progression stages and complexity at the level of multi-morbidity, for a particular patient. While such a comprehensive platform could have a variety of applications, its use in the design of sequential combinations of interventions, where timing depends on the recognition of a particular health state, is emphasized.

Summary of enabling paradigms

Stage 1: Building an initial platform, across two diseases

If we want to build an initial platform that captures current knowledge in both Alzheimer’s disease and chronic liver diseases, how could we get started, based on what exists already? In both indications, we have a relatively good understanding of disease progression states, from a medical, biological and economic point of view. This includes earlier stages of disease (when symptoms tend to be mild, with minimal impact on QoL and healthcare resource usage) to more advanced stages (when symptoms are more severe, with a more dramatic effect on QoL and healthcare resources). At the disease progression level, we can use the following starting points: a) Alzheimer’s disease progression theory of Braak & Del Tredici (2015); b) the review by Pellicoro et al. (2014), summarizing chronic liver disease progression. At the computational level, we can use the following starting points:

Modeling health states. In Alzheimer’s disease, there is already a rich history of using computational modeling of distinct health states in the context of disease progression and disease severity, as reviewed by Green (2007). In the words of the author, “Markov models may be particularly useful when a decision problem involves clinical changes, across discrete health states, that are ongoing over time”, with such models “representing the course of a disease in terms of mutually exclusive ‘health states’ and the transitions among them”. For example, a 6-month cycle has been used to calculate transition probabilities between states, using clinical trial and epidemiological data. Such computational models are used to assess the value provided by medical interventions, e.g. those that result in a delay of progression towards severe health states. More recent updates on usage of such models, including additional applications, is provided by Green & Zhang (2016). Based on the small numbers of health states modeled so far, integration of additional data could lead to a population of alternative models with different numbers of health states, initially for Alzheimer’s disease only. It may also be necessary to extend the simplistic Markov modeling approach, e.g. considering progress made in projects like the 100K cohort (Hood & Price, 2014) or the Google Baseline study (Piller, 2015). Once an initial version of the computational platform is developed, it could be extended to chronic liver diseases.

Semantic framework. IMI, a renowned multi-stakeholder platform for the development of pre-competitive assets with long-term effects related to medicine, has developed starting points in that area. This includes the Aetionomy project, which enables the representation of complex networks of information, including cause-and-effect relationships, based on the BEL language and the Semantic Web data format RDF. Braak and Del Tredici’s theory on connections between pathology and clinical symptoms in Alzheimer’s (Braak & Del Tredici, 2015) can be generalized within the platform, in areas that are linked to the above health state models, to capture a slowly progressive tissue damage pathobiology, with regenerative biology context, affecting the dynamics of progression.

Stage 2: Extend to skin regeneration

Therefore, the stage 1 platform would capture disease progression in two diseases, using Markov models from Alzheimer’s as a starting point, with knowledge from liver diseases, a very different indication with different characteristics, enabling the reusability-centric design of health state modeling and semantic framework. Although it would be desirable to then add new knowledge from both indications to the platform, their slowly progressing nature will limit the speed of such iterative optimization. To add faster-learning extensions that benefit the platform development effort, additional areas of medicine will therefore provide a natural focus of stage 2.

As discussed, skin regeneration as observed in non-healing wounds (skin ulcers) is a promising candidate for stage 2 medical focus areas. Based on existing knowledge on proteins found in wound fluid, a fluid similar to but also distinct from blood, panels of candidate protein-level biomarkers could be designed that capture ‘wound states’. Wound outcomes linked with such wound states and various ‘standard-of-care’ interventions could be tracked using nurse-centric Digital health tools that describe progress towards wound closure and healing (i.e. wound outcomes). Such tools could also capture features important for the early detection of potential complications, and facilitate the involvement of relevant experts based on wound states. Smart dressings for wounds that can benefit from negative pressure therapy are under development, within the Horizon 2020 program in Europe, potentially complementing the protein-based monitoring of wound states. In addition, new technologies are available for non-invasive assessments of skin architecture, based on confocal microscopy (Lange-Asschenfeldt et al., 2012). When connected into a comprehensive solution for tracking wounds in wound nurse type settings, such a connected health solution could not only generate valuable real world evidence on a diversity of ulcers, but also advance the ability to recognize and model health states as discussed above. For example, would the platform enable comparisons between innovation efforts in different countries and healthcare settings in a way that is more difficult at present? Testing their ability to model patient journeys in different populations, countries and healthcare settings could help to refine such models in a way that generates a second generation of reference models with an improved ability to capture ‘real world’ variation, where it matters. The more data become accessible for comparison and optimization, including new data types that extend our knowledge on the biology of different states, as well as diagnostic advances that change the health state recognition part of the models themselves, the more useful such health state models will become, in each iteration of improvement through data-based learning. Connecting them with tools early on that facilitate such iterative optimization will be crucial, including algorithms for the integration of multimodal diagnostic data related to such health states. Special attention should be given to make sure that different patient journey clusters are represented, if differences between those clusters can affect the definition or interpretation of health states, and transitions between health states.

Additional indications. Beyond skin regeneration, additional areas of medicine that are currently not yet identified may become a focus in stage 2, if they present an opportunity to add faster-learning cycles to the health state learning process, within that platform. Candidate datasets include longitudinal observational human studies that generate data aimed at learning health states and state transitions, including those in early stages of disease, such as the 100K project (Hood & Price, 2014) or the Google Baseline study (Piller, 2015). Other reusable aspects of the platform may also require comparisons beyond those 3 initial indications, e.g. to design a reusable semantic framework around the health state modeling effort.

Stage 3: A platform across many diseases

Once stage 2 is mature enough, the possibility of extension towards multi-morbidity in chronic disease progression space may present itself. Patterns of multi-morbidity frequently observed in epidemiological data could provide starting points for interdisciplinary focus, in terms of ‘real life’ health states composed of several chronic disease aspects. Limited resources in elderly care settings could provide an economic aspect, linked to an existing system of cross-indication progression monitoring, including tools for QoL and risk indicators in those settings.

At that stage, we may be able to approach a more generic and more patient-centric theory of chronic disease progression and health states, across indications. Convergence among the developments discussed above would facilitate its maturation. In particular, diagnostic innovation in slowly progressing chronic diseases would improve our ability to accurately diagnose different stages and variants of disease, including improved understanding of earlier stages of disease that are characterized by a clinically ‘silent’ progression of tissue damage that increasingly outpaces regenerative, damage control or repair mechanisms. In reference models that capture average disease progression in defined populations, transitions between the states in those models would be calculated using a variety of data, from different diseases, and different types of diagnostic evidence, to capture the characteristics of that population.

Once we have enough information about the most common health states and their reliable recognition, this new capability can be connected with machine learning capabilities that help to refine such models in various settings, based on an initial reference model and some starting conditions that, over time, increasingly fit the features of that particular setting (e.g. the educational and expectation profile of the involved participants, including healthcare providers and patients, as well as established processes, habits and culture). Optimization would be achieved based on feedback loops based on measured outcomes, including patient outcomes, as well as healthcare utilization and other economic aspects, at population level. Reference clusters of patient journeys could be refined, e.g. by adjusting the weights given to particular features in the clustering, filtering for the most predictive features, and including new features not covered in the reference model. Such adaptations should then find their way back to the next generation of reference models as well, so they become easier to adapt to different settings in the next round of optimization.

Design of combined interventions

When single interventions are not sufficient to achieve the desired outcomes, combinations of interventions either delivered at the same time, or in a particular sequence over time (that considers increasingly refined health state recognition capability), will be an interesting option to consider in stage 3. This could mean that we will be increasingly able to iteratively approach a near-optimal personalized solution for a particular patient, at a particular time in their disease progression, extending the paradigms learned in oncology and AIDS to more indications, and into longitudinal data space. In principle, such value may not need to be limited to therapeutic interventions linked with health states, it could accommodate preventative interventions and even monitoring actions. For example, intervention ‘IN1’ designed for health state ‘HS1’ would lead to a subsequent health state HS2, which triggers intervention ‘IN2’, and so on. After ‘HS1’ there may be a branching point that, in some patients, leads to another state, ‘HS3’, which does not match well with intervention ‘IN2’, but requires monitoring that, once state ‘HS4’ is reached, triggers intervention ‘IN3’. Ideally, the diagnostic recognition of those states HS1-4 would be achieved with a single, reusable diagnostic procedure that is able to differentiate those health states based on non-invasive, low-risk approach (Figure 2).

Figure 2. Design of combined interventions, as an application of health state modeling.

Health states (HS1-4), which match state definitions in probabilistic Markov models, are connected with interventions (IN1-3), defining the time aspect in the PM vision (“the right intervention for the right patient, at the right time”). Each health state would have annotation in terms of pathobiology, health economics and clinical picture.

Patient-facing solutions

Based on those reference models, user-friendly, patient-facing solutions could be developed, which compare the individual’s profile, at a particular time, with the most relevant reference model. If the individual’s profile is indicating faster-than-average progression towards one or more disease progression paths, a number of options could be explored, with tools that help monitoring the effect on progression at the level of health states. For example, the beneficial effects of lifestyle changes or therapeutic interventions on that profile could encourage continuation and compliance to relevant guidance. Gamification-related approaches could be useful in exploring aspects related to user engagement, considering expertise from user-centric design.

Clinical studies

Similar usage may apply to clinical studies, if they span follow-up periods that contain health state transitions. In addition, increased ability to recognize distinct health states could generate hypotheses for new study tools linked with well-established measures and outcomes.

Care coordination

Coordination of care, including the actions of different healthcare providers, as well as social/elderly care, is a very complex topic. The proposed platform could facilitate such efforts by simplifying the recognition of health states that require actions by specific components in the system, at a particular time, to then follow the effects of those actions on health state transitions more closely. For example, it could facilitate early recognition of worsening condition, complications, and other signals that require attention, and their collaborative interdisciplinary management.

Care pathway redesign

While (health)care processes are often quite stable over time, once the team and process landscape is up and running, there are periods where such process landscapes are under discussion, to optimize particular outcomes, as well as economic constraints, e.g. for certain patient clusters that show high costs but below-average outcomes. Teams with a history of ‘care pathway redesign’ could start to engage in the proposed platform, as early as late phases of stage 2. If it is possible to improve the recognition of clinically and economically relevant health states in such settings, care pathway redesign projects would likely be able to extract value from such insights, as they try to determine the best time and mode to intervene in specific types of patient journeys. It could also help them with the exploration of a large range of options for care pathway changes, e.g. in a visual form that supports time-efficient discussion and consensus formation in complex, interdisciplinary groups, based on connected health state diagrams. On the other hand, such collaboration would enable early influence on the design of the proposed platform, to facilitate collaboration at such interfaces, which could become increasingly impactful on both sides over time as health state recognition (via diagnostic tools) and modeling add value to care pathway redesign projects. Such collaborations could therefore benefit both sides, the developers of the proposed platform, as well as the teams involved in care pathway redesign. This includes educational aspects required for such development, with the particular side effect that the platform community gets anchored into ‘real life’ settings as early as possible. Those focused on disease biology aspects of health states would benefit from such collaborations by improved anchoring of their efforts in ‘real life’ care settings as well.

Device development

Institutions involved in device development could benefit from the platform in areas related to chronic disease progression, e.g. in projects aimed at developing smarter, more connected devices that contribute to multi-stakeholder value (similar to ‘smart inhalers’, see above). For devices linked with therapy application, the platform could help to manage projects, considering challenges related to the different timelines and cultures in therapy and device development, by allowing collaboration without excessive dependency between projects.

Disease biology

Applications of the platform, related to disease biology, include:

Semantic web

By closely linking the effort on the development and optimization of health state models with initiatives focused on the representation of semantic aspects of relevant data, the following applications and value can be envisioned:

Recent progress made in relevant multi-stakeholder communities, such as FORCE11, towards consensus on guiding principles in related areas includes: