Dynamic Medicine
If the ongoing development of medical “sensors” — tests, monitoring, and imaging — is combined with an understanding of dynamic response, medical care would be dramatically improved. Instead of relying primarily on the outcomes of statistical studies of interventions and outcomes, individual medical care can be better guided by the real-time viewing of the effects of interventions. Moreover, dynamic response is a sensitive probe of the healthy state and susceptibility to systemic failure. The testing of dynamic response expands dramatically what can be known about a system and is an essential step toward true health care: Maintaining individuals in dynamic states that have lower susceptibility to disease and disability.
Picking up a jar is different from picking up a cat. In the former we might calibrate the amount of force needed and apply the same force to many jars of the same type. In the latter, we have to adopt a different approach, one that recognizes that the cat might do any of many things and we have to be ready to respond to what happens on the way.
Medical science cannot become too focused on the medical equivalent of picking up jars: Determining the intervention, particularly medications, that are able to treat a particular condition in its typical form to the typical person. The gold standard of medical studies, double blind experiments, are well designed to determine whether treatment X works well on average. This is not a good approach to variations across individual conditions. Physicians then have to adapt their treatments to individuals, not just different conditions, but how individual patients respond to particular treatments. A way to advance this attention and concern for individual differences involves experiments that characterize the response of individuals to interventions. There are variants of double blind experiments and a wide range of other ways that medical information is gathered that go beyond the usual double blind experimental approach. However, a shift to an adaptive dynamic strategy comparable to how we pick up cats would be different.
There are two enabling technologies, the rapid development of medical sensors, and the theoretical understanding of nonlinear dynamics and response. The expectations of dramatic changes in medicine through genome based research also are focused on individual differences and dovetail with the need for an understanding of dynamics in medicine.
A dynamic approach to medicine is not a new direction. It is an update of the oldest strategy—learning from what happens when you try something. The key concept is that treatment-induced changes/outcomes should be monitored dynamically, and treatment should be corrected based upon the observed changes. Experiments for evidence based medicine have to be adapted to this approach using dynamic models and protocols that go beyond testing whether treatment X works or dosen’t work.
Moreover, in many cases treatment should not wait for illness to be apparent. The state of health should be monitored using measures of the dynamic response of physiology to perturbations. Dynamic response is used in everything from nervous system reflex measurements to allergy tests. The reflex in response to touching the sole of a foot can reveal spinal cord disease. Putting small amounts of a variety of substances in skin pricks, including bee venom and penuts, reveal something that is not present now, but might happen: Deadly allergic responses. Identifying such susceptibility to medical conditions enables health related interventions. By expanding the use of such approaches, adjusting the time frame of observation, and increasing our understanding of how this should be done, dramatic improvements in the effectiveness of medicine are possible.
To explain dynamic medicine, it is easiest to describe a few examples. Figure 1 illustrates several ways dynamics can be part of medicine. Panel A shows the most common idea that a condition is evaluated and a treatment prescribed in order to achieve an outcome. Panel B shows the case where the condition itself has a dynamic associated with it. For example, a rising temperature. Panel C shows the case of dynamic response, as happens when a physician induces a reflex or performs an allergy test. Panel D shows the case where treatments themselves trigger dynamics and that dynamics is observed and influences the course of treatment.
Let’s now consider these ideas by considering examples in more detail.
Dynamic response
A first example is the now classic research of Ho and Perelson in the mid 1990s on the dynamic response of HIV to protease inhibitors [1,2], Prior to this research, in the late 1980s and early 1990s, HIV infection was known to cause AIDS. However, the process of infection involves a long “latent” period which can extend for many years. This latent period has little or no symptoms. There was an ongoing debate about whether treatment of the infection should begin during the latent period, or only when symptoms of immune deficiency began.
During the latent period, the level of HIV in the blood of an infected person was found to be nearly constant over long periods of time. It seemed as if not much was happening.
In the early 1990s new drugs, the protease inhibitors, were discovered to inhibit the production of HIV, without affecting it after production. Perelson, a theoretical immunologist, analyzed the dynamic response measured by AIDS researcher Ho. This involved measuring the HIV concentration at frequent intervals after treatment with the protease inhibitor began. Originally measuring at daily intervals, they found that the level of HIV was dramatically reduced after only a single day. Using a simple predator-prey model for HIV production and elimination by the body, Perelson showed that in the latent period the virus (the prey) had to be replicating at a very fast rate to balance rapid elimination by the immune system cells (the predators). Later experiments measured the HIV level every hour, and improved models confirmed the earlier results.
This meant that the constant level of virus in the body was actually a balance of rapid production with rapid elimination: A ferocious battle was going on that was not seen in the constant values. The active nature of the infection in the latent period led to the understanding that AIDS is the final outcome of active HIV infection and to the early treatment of HIV by multidrug therapies. This is one of the main advances in HIV treatment till today.
The work of Perelson and Ho shows several elements of dynamic medicine: an imposed change, frequent monitoring, and a model to understand the observations. Studying the dynamic response is central to both experimental and theoretical understanding of the behavior of all complex systems.
How much time does it take?
The Perelson and Ho work illustrates perhaps the most important question about dynamics: What is time scale of change? Originally, the researchers didn’t know the time scale at which to observe the dynamics because the dynamical information was not known. It is hard to underestimate how important it is to know the time scale of change and compare it with those of other processes. The times scale sets the frequency with which observations are needed, how fast of an action is needed in order to intervene, and therefore what intervention process is needed. Responding rapidly enough to stop progression of a disease, or having enough patience to observe when there is a slow progression, are flip sides of the time scale question.
Once the time scale is known, there is still a challenge of understanding what type(s) of dynamic responses are possible. Converging to a bad or good outcome is a standard expectation for dynamics, but change can be much more complicated: divergence, worse before better (or better before worse), oscillation between different states, and chaotic behavior are all possible types of dynamics. Knowing what the type of nonlinear dynamical behavior is for a particular type of system (or set of types) is required if effective actions are to be made.
A confusing and yet common type of response happens when a system’s self-regulation counteracts the effect of an intervention. When the intervention stops, the system may be worse off than before. These are described as “rebound” effects [3]. Examples that have been discussed extensively include increased stomach acid production after stopping use of an acid production inhibitor, and increased blood clotting when use of an anti-clotting agent is stopped. For some cases (but not all!), a surprising way to handle such a circumstance may be to push the system in the opposite direction. The system’s regulatory response may then shift the system in the desired direction. Consider a person who leans to the right. We might push them to the left to correct their leaning. But it may actually work better to push them further toward the right. When their system resists the push, letting go may cause them to correct their position. The way the push is delivered matters. A short push might be effective, but a long push (or push and hold) might work exactly the opposite way. How much force also matters.
Understanding the general ideas of dynamic response is far from enough for progress. We have to identify what to measure, how to perturb the system, and characterize the way or ways the system responds to that perturbation, including what the differences in response mean about individual differences, health and disease. Having many different ways to perturb the system is important for being able to identify the behavior of such a complex system.
What is dynamic medicine?
Ultimately the HIV study, and many other studies of dynamic response, resulted in a treatment based upon a fixed protocol. Each individual doesn’t necessarily need the same kind of study of their response to medications in order to specify their treatment.
This is different from incorporating dynamic response into treatment itself—checking the patient’s response to a treatment and adjusting it. Perhaps ironically, the process that doctors intuitively use when following the progress of a patient and changing the treatment accordingly is difficult to study using modern double blind experiments. In statistical studies treatment options are fixed in advance. So adjusting the medication dose, or which medication is used, based upon observations over time doesn’t fit. Moreover, the very blindness that is used to guarantee objectivity contradicts adjusting the treatment. Thus, using such experiments, it is hard to scientifically improve the process of physician observation and changes in treatment.
There are many contexts in which physicians adjust medications in order to achieve some target result or measure. For example they adjust interventions in order to bring a blood test within a range of accepted variation. In dynamic control theory language, the physician is acting as a kind of thermostat for regulating physiology. This is a basic form of dynamic response of the combined system including patient and physician. Just as a thermostat measures the temperature in the room and intervenes when it is outside a range, so the physicians uses physiological measures to adjust medications. The dosage might be informed by standard clinical trials of fixed dosages. Some efforts have been made relatively recently to incorporate dosage adjustment in trials as part of their design [4]. Still, these studies do not study the process of adjustment itself, their innovation is to use fixed target measures rather than fixed dosages as the treatment (independent variable) of the study. Physician practice must be guided by combining a variety of sources of information and adapted to individual needs, without direct testing [5]
We see that there are two strategies doctors intuitively use. In the first, the doctor identifies a condition and the treatment that should be followed, prescribes it, and doesn’t expect to follow the outcome because it is likely to be successful. In the second, follow up monitoring or scheduled appointments for checking on how the treatment worked are part of the treatment process. The former condition and action model fits within the current evidence based scientific medical approach very well; the latter doesn’t, at least not easily. The reason has to do with the limitations of double blind experiments in incorporating the extended time process of multiple observations and adjusted interventions.
Science of dynamic medicine
While double blind statistical studies don’t generally study what physicans do as they adapt their treatments to individual conditions and responses, it is possible to do so. At least in some cases. The first step is to study the dynamics of the system itself, including the dynamic response that occurs when particular interventions are made. This dynamic behavior is then represented as a mathematical model with parameters, some of which are common accross individuals, and some that must be fit to each individual. The second step is to identify a protocol for physician intervention, including a set of specific observations and interventions. The protocol incorporates the effect of fitting the individual dynamic model, and the set of interventions that are to be performed as inferred from that fit. The protocol can then be tested to evaluate the effectiveness, not of a specific intervention but as a whole.
The hard part, of course, is realizing this processs: doing the actual work of putting together the measures that need to be done, the dynamic response that is to be expected, and the treatments that should be done. All of which is preliminary to the experiment itself to validate a treatment protocol.
Adaptive treatment
Despite the rarity of relevant double blind experiments, adaptive treatment based upon observation continues to be an essential aspect of medicine. Hospitalized patients are monitored to see how they are doing and care and interventions are adjusted in response.
As just one example, hospital based treatments of infection monitor white blood cell counts, vital signs, and other measures to determine the effectiveness of antibiotic treatment. The underlying problem is that antibiotics are not always effective. This is a growing problem due to the increasing appearance of antibiotic resistant strains of pathogens. Antibiotics can also have adverse effects, including allergic reactions.
The potential of dynamic medicine in the future is to use observations after giving a regular, or even a test dose, of an antibiotic to see whether it will be effective against a particular strain of bacteria, and whether there is an allergic reaction. The observations might measure the bacteria themselves, or other markers of the effectiveness of the antibiotic or its potential side effects. With such information, it should be possible not just to determine if the antibiotic is effective, but also to adjust the dosage. Instead of giving a standard dosage that is almost surely able to eliminate the bacteria in everyone, the dose given might be reduced based upon the response in a particular individual to a particular strain of a bacteria. Confidence in use of lower doses, that can still achieve the desired outcome, would minimize adverse effects. In addition to allergic reactions, these can include reducing or eliminating important parts of the microbiome and weakening the immune system’s ability to combat other infections. Today, sometimes, this is already done.
Today there are tests of antibiotic effectiveness that use bacterial cultures, and skin tests of allergic reactions. However, we can much more widely incorporate tests of effectiveness and of adverse outcomes by incorporating dynamic response into the treatment protocols themselves. Such treatment protocols should be validated using improved statistical tests.
Using dynamic response for infections requires that we understand how the elimination of the last bacteria from the system occurs, so that the infection can’t come back. This may be challenging but is surely worthwhile when the arsenal of antibiotics is being challenged by antibiotic resistant strains. The benefits of advancing the study and use of dynamic response are much more general, however. Any treatment triggers a dynamical response that underlies its effect. That process should be be understood on an individual basis to enable the best care for each person, which requires observing it rather than assuming a generic response. Just as lifting a cat requires observing its behavior and not just applying a prescribed force.
Anticipating disease
While treatment of disease is the primary focus of medicine today, the anticipation and prevention of disease should become more prevalent.
There already are common examples that illustrate how dynamic response can be used to anticipate disease. The stress test [6] for susceptibility to heart disease, is probably the most common example. As with HIV, the response of the body to an imposed change (stress) is measured. The test uses a controlled amount of exercise (walking on a treadmill) as the imposed change and monitors the response of the heart through measuring changes in the heart rate, blood pressure, and electrocardiogram. These observations are compared with standards of healthy and unhealthy responses (e.g. a simple model) to determine the level and type of medical intervention that is needed to avoid a heart attack.
Similarly, the Glucose tolerance test (GTT)[7] is often administered to diagnose Diabetes Mellitus in pregnancy. It makes use of the ingestion of a controlled amount of sugar to determine whether the body is likely to develop gestational diabetes — i.e. diabetes that occurs during pregnancy because of the higher demands on the homeostatic control that are required.
There are other examples of dynamic response used in medicine today. However, there are many more that are possible. While the focus in recent years has been on genetic factors leading to susceptibility, a more direct and meaningful test is to monitor the dynamic response of a system to various triggers. The early identification of disease can, in many cases, be done effectively by measuring dynamic response.
Anticipating disease extends the idea of early detection. It also gives rise to additional decisions about anticipatory treatments. When someone is predisposed to a disease but doesn’t have it yet, how should treatment proceed? For example, prophylactic mastectomy for women who are genetically predisposed to breast cancer is one of the areas of debate and decision today. In general, anticipatory treatment of medical conditions is in its infancy. It should be recognized that such treatments are treatments of susceptibility rather than disease. Since, in many cases, the dynamic response is a measure of susceptibility, it is also a mechanism for evaluating the effectiveness of treatments.
Regulation and repair, aging and homeodynamics
Another area for which dynamic response is essential is the “treatment” of aging. Currently most of the attention on aging is for people over 65. But aging is a process that involves the failure of progressively larger subsystems of the body starting from very small ones. Aging begins quite early, as is apparent from the retirement of athletes who are pushing the capabilities of physiology to the limit. They are finding that by the later thirties there are basic changes that prevent athletic competition at the highest level of the sport.
Homeostasis is the most basic idea in medicine. Homeostatic mechanisms act to protect the cells, tissues and organs from conditions that cause them harm by regulating temperature, salt balance, oxygenation, and many other properties of the system. Thus, impaired homeostatic mechanisms can be blamed for the deterioration that is characteristic of aging. Despite the name, homeostasis is a dynamic property not a static one. It is the ability of the body to maintain certain internal conditions in the face of a changing external environment. This requires dynamic response not just stasis (A static system does not maintain homeostasis!).
At this point in time we have very limited understanding of how to treat the loss of regulatory and repair mechanisms, with the exception of diabetes and hyper or hypo thyroid conditions which affect metabolic control. The main reason we do not have an understanding of how to treat these conditions is that we have not measured the dynamic response of the system and how it changes over the life cycle.
Studying the healthy dynamic response is not easy because of the inherent complexity of multiple interacting physiological control systems. This complexity, however, can be dealt with by modern modeling techniques and the increasing ability to obtain large amounts of biological information. This would result in an understanding of the dynamic state of health and the recognition of the processes of aging that reduce the ability of natural homeostatic mechanisms to protect and repair the body. By bolstering these natural regulatory and repair mechanisms, the later deterioration and failures associated with aging may be dramatically postponed.
Connecting to genomic medicine
Dynamic medicine should also play an important role in the plans that are being developed for genome based medical treatments. The idea of genome based medicine is to use a person’s individual genetic makeup to help us personalize his or her treatment, avoiding problems that are much more subtle than major allergic reactions and providing a treatment that will be more effective.
The plans for this type of medical treatment have a fundamental limitation which can be alleviated by using dynamic medicine. Traditional statistical evidence based research can only provide a very small amount of information relevant to personalized medicine. A statistical study requires a large number of people, randomly divided into treatment and control groups, to compare outcomes.
Individualized treatment doesn’t lend itself to statistical studies because there are too few individuals of a particular genetic type with the condition being treated. This is a problem that is not exclusive to genetic fingerprint treatments: It is already a problem in studying multidrug therapies because there are too many possible combinations of drugs to test all of them. As an aside, this is a general problem in complex systems, including testing computer chips, because it is impossible to do enough tests to find all possible errors [8].
Dynamic medicine can help solve this problem. Studying dynamic response can use many different measures and their dynamics to expose the system behavior. This is a lot of information, and the ability to make use of a lot of information is essential to personalization.
More specifically, the key is to identify the short time effects of treatment that can let us know whether the treatment will be harmful or effective in the long term. This must still be done using studies of large numbers of people. Still, models can be validated by the response of the system so that it is possible to properly interpret the short time effects. This would enable the individual response of a person in “real time” to be used to guide their treatment.
Monitoring the dynamic response of physiology to treatment is not a complete answer to the problem, since we might do harm on a faster time scale than we can observe. However, the ability to link treatment to a large amount of information about the health of an individual and the impact of an intervention increases the likelihood of improved outcomes and avoiding harmful side effects. Moreover, many more possible treatments can be tried on a single patient, and the treatment can be fine-tuned over time based on the observations.
This is also important for the problem of FDA approval of treatments, because the FDA could approve processes of intervention that involve monitoring and adjustment of treatment. This allows a larger latitude to what can be tried but requires specific notions of what constitutes an effective treatment process. Many medications that are not allowed today because of small percentages of adverse side effects could be allowed if monitoring gives early warning about which individuals will have the side effects and which will not. Examples exist [9].
Dynamic disease
It is helpful to compare dynamic medicine with the concept of dynamic disease that has been discussed in recent years [10]. A dynamic disease is a sudden change in the qualitative dynamics of intrinsic physiological processes. For example, the tremors arising in Parkinson‘s and changes in the rhythmical properties of heartbeats. Dynamic disease is about a change that occurs in the dynamics of physiology, even in absence of treatement. In this case, the dynamics is an aspect of disease. The idea of dynamic medicine is distinct in that it focuses on the dynamic response to treatment. Both recognize the importance of dynamics of the system in considering health and sickness.
Dynamic medicine recognizes that all medical interventions are dynamic processes: In the simplest picture, they change the state of the system from a state of sickness to a state of health. This dynamic process should be the subject of increased understanding and utilization in the treatment process itself. Moreover, the idea of dynamic disease focuses on failures of specific dynamic processes. The idea of dynamic medicine recognizes that even homeostasis is the result of successful dynamic response.
Complex systems and health
The general study of complex systems suggests that health care can be improved by increasing the recognition of dynamic qualities of health and treatment. Medical interventions are dynamic processes whose impact on the state of the system can be monitored to better recognize the effectiveness of the treatment. Health is a dynamic condition, and understanding the dynamic response of the healthy state should give rise to substantial new possibilities for early intervention, improved support of natural protective and healing mechanisms, and better long term health.
Acknowledgement: Originally written in 2001, revised and published in 2017 based upon helpful comments of Luci Leykum, David Aron and Taeer Bar-Yam.
References
- D. D. Ho, et al., Nature 373, 123 (1995). http://www.nature.com/nature/journal/v373/n6510/abs/373123a0.html
2. A. S. Perelson, et al., Science 271, 1582 (1996) http://science.sciencemag.org/content/271/5255/1582
3. A. Sharma, WF. Ebling, WJ Jusko, Journal of pharmaceutical sciences 87, 12, 1577–1584 (1998)
4. MC Riddle, J Rosenstock, J Gerich. Diabetes care 26, 11, 3080–3086 (2003).
5. T Hornick, and DC Aron, Cleveland Clinic journal of medicine 75, 1, 70 (2008).
7. http://www.fpnotebook.com/Endo/Lab/GlcsTlrncTst3Hr.htm
8. Y. Bar-Yam, The Limits of Phenomenology: From Behaviorism to Drug Testing and Engineering Design, Complexity (2015). doi: 10.1002/cplx.21730 http://www.necsi.edu/projects/yaneer/phenomenology/
9. J Tajiri, et al. Archives of internal medicine 150, 3, 621–624 (1990).
10. Chaos Vol. 5, Issue 1 (1995) http://aip.scitation.org/doi/abs/10.1063/1.166069
