Why and How Should We Integrate Biomarkers into Complex Trials? A Discussion on Paradigms and Clinical Research Strategies

This paper is a consensus paper as one of the results of the workshop “Designing research aligned to the whole systems model of health, disease, and healing” as part of NAFKAM’s 2012 Northern Lights Workshop series held in Sommarøy, Tromsø, Norway. The overall goal of the 2012 workshop was to define issues that need to be included in whole systems research and point to relevant methodological issues. The workshop started with the whole group of participants applying a World Cafe style brainstorming on topics and issues for whole systems model research designs. The potential role of biomarkers and whether and how they should/could be integrated in complex trials was identified as one of several relevant topics, and the group of co-authors was assigned as “expert” panel for this topic.

Within the expert panel, the participants began with a short brainstorming session on why biomarkers can be a useful outcome in complex trials. Taken from there, the expert panel developed a principal design template and discussed its operationalization with several, specific examples of increasing complexity. Since the group was small and rather homogenic in their opinions, the discussion was free and not formalized. The paper represents the consensus statement of the expert panel.

Research in complementary and alternative medicine (CAM) encounters a variety of challenges, such as potentially synergistic, multimodal, and complex interventions, aiming at symptom amelioration, and psychosocial and behavioral changes in the patient with long-term effects. Furthermore, these interventions are often dependent on the relationship between practitioner and patient, on specific medical settings, and on patients’ preferences, expectations, beliefs, and motivations. Moreover, patients seeking CAM care often suffer from chronic diseases and multiple pathologies [1]. The challenges these complex and multifactorial conditions impose on clinical research design are particularly relevant for whole medical systems. Typical examples for whole medical systems are traditional medical systems such as Chinese medicine, Ayurveda, Kampo, Tibetan medicine, Unani-Tibb, Native American medicine, as well as homeopathy, anthroposophy, and the rapidly emerging field of integrative medicine. These whole systems not only include multiple treatment modalities, but also alternative diagnoses and patient-practitioner interactions, as well as techniques for changing the patients’ (or in case of primary prevention, clients’) behavior, all of which are frequently implemented in highly individualized fashions and in system-specific settings [2‒5]. Nonetheless, rigorous investigation of clinical effects in controlled clinical trials is possible [6‒8], and even randomized clinical trials have been suggested [9].

Moreover, research in CAM frequently faces the challenges that treatments are not derived from conventional biological hypotheses and the therapies are implemented regardless of whether the “conventional” biological mechanisms, comparative effectiveness, component efficacy, or even safety aspects are documented in the research literature [10]. On the other hand, many of these interventions, in particular in whole medical systems [2], have been practiced for centuries or even millennia, so that at least an implicit empirical knowledge on clinical effectiveness and safety within certain cultural settings is available [11, 12].

All these preconditions, the complexity of the interventions and the treatment setting, the fact that many of the patients suffer from several chronic conditions or diseases, and the fact that there is generally no clear hypothesis on the biological mechanism of action, make the applicability of the gold standard in clinical research, the randomized controlled trial, particularly challenging [1, 2, 10, 13] although, as we see it, not impossible. In the field of CAM, there has been a growing awareness that the development of appropriate methodological and statistical frameworks for the investigation of complex interventions is one of the key answers to the quest [1, 2, 8, 10], including a partial re-evaluation of the “outcomes” concept [13].

Considering all the previously discussed challenges, why should we even attempt to utilize biomarkers as outcomes in complex trials? Treatment effects, measured by biomarkers, are usually considered a sign of specific efficacy. However, could and should we not be content if we are able to show the clinical effectiveness of CAM interventions to the benefit of the patient?

We believe that there are at least three good arguments for research strategies to also include biomarkers:

1 Pragmatism: CAM interventions in conventional medical care settings are often delivered in addition to conventional care or as being part of integrative medical care. Enhancing the understanding of complex CAM interventions with regard to the conventional biomedical model is likely to facilitate the integration of the various approaches and prevent possible undesirable interactions among different treatment modalities. The synergistic effect of a common basis of understanding would improve care, increase patient safety, and improve practitioner and patient satisfaction.

2 Epistemology/hermeneutics: many systems of traditional healing are based on theoretical models about health and disease, or even broader, the organization of life in general. Broadening the spectrum of outcomes in complex CAM trials from emotional/motivational, functional/behavioral to biological/physiological mechanisms may allow for a “cross-talk” between the traditional and “western” natural science models of health and disease with the overarching goal of a common understanding and a better approximation of “medical realities.”

3 Strategic: even though we recognize that the currently acknowledged “biopsychosocial model” reaches substantially beyond the boundaries of the “biomedical model” of health and disease, we observe that much of the current medical research, also within scientific CAM contexts, is still being conducted along the tenets of the biomedical model. If the aim is to establish complex CAM interventions in a model of good patient care, demonstrated effects on biomarkers will increase the potential for integration of CAM interventions in the predominant models of contemporary medicine.

If we agree that biomarkers as outcomes are important for the understanding of CAM interventions, a hypothesis- and strategy-driven process for the selection of the most appropriate biomarkers is needed. Most importantly, the chosen biomarkers should be valid to the indication/condition/syndrome studied. This is best achieved, if they are oriented towards the core subjective health complaint or symptom of the patient. Moreover, as Paterson et al. [13] consequently emphasize, outcomes which are appropriate for complex health interventions ideally are able to reflect changes and dynamics.In complex health interventions we often see that a treatment, even under strictly controlled conditions, shows its effects in the way a subject adapts to a challenge. As Paterson et al. [13] point out, a process is something that enables the individual to adapt to varying experiences. Therefore, while trying to design a trial and searching for the right outcome, we may be trying to quantify something like a “flexibility-stability” continuum. With regard to including biomarkers among the outcomes, this means that we have to be aware of whether we expect a treatment to show its effect in the magnitude of an outcome or by altering its variability.

Two biomarkers exemplifying this principle are body temperature and heart rate variability. A stable body temperature (thermoregulation) is a core requirement in endothermic animals (like mammals) and a mechanism that provides a stable environment for the chemical reactions in the body cells. Body temperature undergoes slight variations due to circadian rhythms and activity; it is a classic example for homeostatic regulation [14]. Consequently, deviances from the normothermic optimum, such as fever, are generally a sign of disease or, at least, a prodromal symptom. Heart rate, on the other hand, is a classic example for a physiological variable made for flexibility and adjustment. Heart rate varies with regard to physical (e.g., sport) or psychological (e.g., stress) challenges and supports behavioral actions needed in response to the external (e.g., climbing stairs) or internal (e.g., thoughts and emotions) environment. The loss of heart rate variability is generally a pathological sign of cardiovascular disease. Consequently, heart rate variability would be a classic example for a dynamic variable, where a desirable treatment outcome could be an increase in heart rate variability [15].

These two examples illustrate that changes in processes (e.g., heart rate variability) as well as levels (e.g., homeostatic body temperature) may be indicators of improved health. It is highly relevant to consider the nature of the underlying mechanism when choosing a biomarker. First, the biomarker should relate directly to the question (e.g., whether the mean or median of a measure or an index of variability is appropriate) and will to some degree impact the study design. However, the question of how sensitive the biomarker is to change will to a large extent drive the study sample size needed to detect an effect.

These conceptual issues may be further challenged when trying to integrate biomarkers into complex interventional trials, since biomarkers are usually identified and selected on the basis of a clear, biological hypothesis investigated in a design which allows for the application of Wilhelm Wundt’s (1832–1920) fundamental “principle of isolated variation” in experimentation (e.g., cited in Krauth [16]). The principle of isolated variation requires that the treatment groups are varied according to only one particular variable, which can be expected to exert an effect on the dependent or outcome variable. All other sources of variation must remain constant. The classical pharmacological randomized controlled trial is a typical example in that it only varies the content of the active drug against a placebo. The application of the “principle of isolated variation” is mandatory when biological mechanisms or the efficacy of a specific component of the therapy (component efficacy) of a treatment are to be studied (see also Fønnebø et al. [10]).

If the aim is to investigate the conventional biological underpinnings of CAM interventions, biomarkers are mandatory. With regard to biomarkers and the “principle of isolated variation,” complex trials in the CAM field struggle with two major challenges:

(a) The mechanisms of action of the intervention are commonly unclear, and the explanatory models are rarely grounded in conventional physiology/pathophysiology. Therefore, it is often difficult to decide which variation of the experimental or treatment conditions is being isolated, and often several variables unavoidably change simultaneously. Consequently, biomarkers may also vary in uncontrolled ways in several of the treatment conditions. This can make the study results difficult to interpret.

(b) The CAM interventions are complex and often address several aspects of the patient and his or her symptoms, which from an understanding of the (e.g., organ-) specific conventional pathophysiology may be seen as unrelated to the disease or at least only relevant at the second or third level. However, these (cumulated) second- or third-level effects may actually have an impact on the specific pathophysiology, but the impact may show itself on a different, likely a longer, time line.

In the light of the complexity of the problem, can we integrate biomarkers into these multidimensional trials, and is there a need to do so? The answer to both questions is clearly “yes” [2, 13], and there has been an explicit call for “objective change indicators” which should preferably be biomarkers [13].

Even though this group of authors favors or is at least supportive to the concept of the use of biomarkers, we are well aware that a wrongly selected biomarker will not add to our understanding of mechanisms of effect of an intervention, quite the opposite. Since biomarkers are often seen as making study outcomes more credible and they are therefore “fashionable,” it is also fundamental that the biomarkers used are in fact valid for (i) the symptom pattern and (ii) the intervention.

It is tempting to choose biomarkers which are convenient or readily available. A good example is the biomarker “salivary cortisol” which has often been utilized when stress reduction was suspected to play a role. However, salivary cortisol as a hormonal measure is a very dynamic stress marker, vulnerable to many different factors that play specific roles in patients, such as medications, comorbidities, and everyday stressors [17‒19]. Therefore, salivary cortisol is not particularly suitable to detect long-term effects. It can, however, be used validly in standardized subexperiments, where a dynamic response is systematically evoked and where salivary cortisol is a well-established marker for the stress response [20, 21] (see example 4).

It is exactly for these reasons that the biomarker expert panel was established at the Northern Lights workshop “Designing research aligned to the whole systems model of health, disease, and healing.” The research strategy developed and suggested here intends to guide researchers in their selection of biomarkers, in order to avoid designs, where the chosen biomarker is not valid for the study. Such a situation is, at best, a waste of resources but can also, depending on the nature of the biomarker, set the study participants at risk, without adding useful information to the understanding of the mechanisms of effect and should be avoided.

The central question is: how can biomarkers be selected strategically so that the likelihood that they are valid for the intervention and the design is maximized? The challenge can be addressed through the implementation of two general principles which refer (a) to the process of selectingthe appropriate biomarkers and (b) the application of small “subexperiments” embedded within complex trials.

(a) The selection of biomarkers should be driven by the physiological system where the treatment-induced change can be expected, such as the CNS, the autonomic nervous system, the immune system, etc. Moreover, the level where the change is to be expected needs to be taken into account (e.g., for the CNS: at the level of the brain or below?) [22, 23]. The most rational guidance is provided by the relevant symptomsof the patient. If this is, for example, pain, then a pain-related biomarker would be appropriate.

(b) “Subexperiments” or challenges to the system are needed when complex interventions cannot be expected to alter biomarkers at a resting level/state but may alter the physiological response to psychophysiological challenge (e.g., mental stress, physical activity). In these subexperiments, specific challenges to the physiological system in focus, e.g. a stress test, are applied, and the change in response to the challenge, meaning an evoked response (e.g., change in heart rate variability), is measured as outcome (see also Fig. 4 below).

In the following section, we suggest a generalized trial design which provides a rational guideline for the selection of biomarkers, taking into account the challenges and frameworks identified above. Furthermore, we provide three different virtual trial examples to serve as examples for the application of the template, with different indications (low back pain, esophageal reflux disease, breast cancer survivors) and different complex interventions (traditional Chinese medicine, TCM, Ayurveda, and integrative care).

Usually in a trial, the study participants are selected according to a health condition or disease by a diagnosis that is operationalized by inclusion and exclusion criteria. From here, either all or a selected subgroup of patients will receive a CAM treatment. The CAM treatment usually also has (i) its own definition of health and disease and (ii) its own theory of pathodynamics. Consequently, a second set of selection (inclusion and exclusion) criteria according to the premises of the whole medical system or other complex intervention may or may not be implemented. Figure 1 describes this process.

In the first step, the key symptom/variable which is expected to be addressed by the intervention needs to be identified, and an appropriate outcome measure as operationalized for this symptom/variable needs to be selected. This may often be a self-report outcome. Self-reports are of immense importance even though this paper focuses on biomarkers, because self-reports provide a measure for the most relevant aspect of any trial: they represent a measure of suffering, the outcome with the highest relevance for the patients and thus for the whole trial! A common example would be perceived pain as the key symptom and the visual analogue scale as its operationalization for measurement (see example TCM, Fig. 2). It is assumed that any type of successful intervention should be able to induce symptom relief.

The next step is to identify a biomarker, which is as closely related to the selected main symptom and/or self-report as possible. This biomarker will be the main or primary biomarker for the trial. For pain, this could for example be a neurophysiological measure taken from the quantitative sensory testing system (QST). QST includes a series of tests applying challenges in the form of controlled pain stimuli to the pain-processing system [22, 24‒26] (see example TCM, Fig. 2). If the trial resulted in a change of the subjective measure and the primary biomarker, then this would document a health condition-/disease-related specific effect of the treatment.

However, every intervention that somehow addresses the main symptom is likely to also influence the related biomarker. For many of the purposes that biomarkers are integrated into complex trials, this would be sufficient. Nonetheless, in order to identify specific intervention-related mechanisms of action, it may make sense to select other, additional biomarkers, which are less related to the main or core symptom, but more closely related to other aspects of the intervention. In order to be as specific to the intervention as possible, other symptoms, which are of relevance for the particular intervention, need to be identified. This is of major relevance in all whole medical systems trials, as the whole system interventions as a general rule address symptom complexes. One or more of these additional symptoms need to be identified. In our TCM example (Fig. 2), this is sleep disturbance, and consequently, the most appropriate biomarker for this symptom needs to be identified, which in this case could be a sleep EEG. We advocate that this additional symptom and biomarker should also be associated with a validated self-report measure.

As mentioned above, a theoretical example from a TCM trial on low back pain illustrates this strategy well (Fig. 2). Patients with low back pain are identified, and a selected subgroup is subjected to a complex TCM intervention. The visual analogue scale is chosen as subjective measure for the most relevant symptom, pain. Since the QST provides a biomarker directly targeting the pain-processing pathways and networks, it is a commonly utilized biomarker for pain-related interventions [22, 24‒26]. In TCM theory, a TCM diagnosis for the pain syndrome could be associated with sleep disturbances, therefore sleep disturbances are a second TCM-related symptom. In addition to a validated self-report, such as the Pittsburgh Sleep Quality Index [27], sleep quality could be assessed by EEG in a sleep laboratory. Consequently, if the TCM intervention reduces pain and shows a related pattern in the QST measures, then this outcome supports a specific effect on the pain-processing networks through the TCM intervention. If, in addition, the sleep EEG and Pittsburgh Sleep Quality Index are also indicative of a better sleep quality, then this result shows a pattern, which can be directly related to TCM theory. If the design was such that the TCM intervention was compared with other active conditions, which likewise reduced pain and related QST measures, but only the TCM interventions showed an effect on sleep disturbance or demonstrated the largest effect independent of pain score change, then this could be interpreted as a specific effect of the TCM intervention.

Our next hypothetical example (Fig. 3) illustrates the principle of a trial on gastroesophageal reflux disease (GERD), treated with a complex Ayurvedic intervention. After selecting subjects according to conventional criteria for GERD, a subgroup of patients suffering from Amla-Pitta according to Ayurvedic diagnosis could be further selected [26]. The main symptom in both the conventional and the Ayurvedic disease entity is heartburn (due to acid reflux), which can be measured by esophageal pH monitoring. Therefore, the related biomarker could be the determination of esophageal pH with an esophageal sensor monitoring pH using a pH catheter. The esophageal pH monitoring is also a good example for the earlier discussed principle that disease is often characterized by the loss of flexibility and that a healing process may be shown in an increased flexibility after treatment. The subjective impact on the patients’ life and their suffering can be measured with a variety of instruments [28‒30]. The example here includes the reduction in quality of life specifically related to gastrointestinal reflux disease, measured with the QOLRAD [31]. In addition, the Ayurvedic Amla-Pitta syndrome may also include a disposition for irritable bowel syndrome with diarrhea (IBS-D), and a tendency for aggressive behavior. Consequently, gastrointestinal transit and salivary testosterone were selected as additional biomarkers [32, 33], specifically chosen for the Ayurvedic theory and intervention. A bowel habit patient diary can be used as self-report measure for gastrointestinal transit and the aggression questionnaire as subjective measure for aggression [34, 35].

As in our TCM example, a relief of heartburn together with an increase in esophageal pH would indicate a treatment effect on GERD. If this effect were associated with a reduced tendency for IBS-D and normalized gastrointestinal transit, as well as a reduced tendency for aggressive behavior and normalized salivary testosterone, and this pattern of effects would not show in other active control conditions, then this result would clearly relate back to the particularities of the Ayurvedic treatment and theory.

Our last example constructs a trial of a multimodal complex treatment program for breast cancer survivors (Fig. 4). The main symptom, which should be addressed by this treatment program, is fatigue. Therefore, the spontaneous EEG could be selected as the appropriate biomarker in order to indicate the CNS correlates of fatigue, namely increased slow wave activity [36‒40]. The FACIT fatigue scale provides a reliable, functional measure of fatigue [41, 42]. In addition, a major component of the multimodal treatment program focuses on stress and stress management under the assumption that the diagnosis of a potentially life-threatening disease and the subsequent therapy, often including surgery and chemotherapy, are major stressors. Of course, to measure stress, many self-report instruments are available and should be carefully selected and applied. In our example, the Perceived Stress Scale, one of the many well-established and valid instruments, is chosen [43, 44]. Moreover, here stress is seen as a variable that reflects a certain level (stress level) and would thus be a static condition, but which can also be seen as a dynamicevent, the stress response to a challenge. Natural killer cells could be selected as a suitable biomarker for the measurement of stress level [45, 46]. As for the measurement of the stress response, a stress interview, the Trier Social Stress Test [47], could be administered and skin conductance, salivary cortisol response, and other cardiovascular measures (e.g., heart rate response) measured [15, 20, 21]. Thus, the stress interview is a good example of a “subexperiment” embedded in a trial [47].

We hope that the suggested strategies will assist other researchers in selecting biomarkers that are meaningful and valid for the interventions studied. As complex as the challenge is, these authors are convinced that a hypothesis- and symptom-driven framework is the most promising with regard to the validity of the selected outcomes. Nonetheless, integrating biomarkers in complex CAM trials is logistically challenging, tedious, time consuming, and costly.

Why should we do it? Research on CAM struggles with the fact that complex interventions have often been applied over centuries or even millennia, and the clinical effectiveness of some is known and is sometimes even striking. However, there is rarely a good understanding (at least from the perspectives of conventional hermeneutics) as to why they work. Why is that so important?

It is widely accepted in the CAM field that a profound knowledge of the clinical effectiveness and the relevant context factors of these complex interventions is required in order to design patient-centered care models that are accepted by the global medical community and health care stakeholders. However, if we want to improve CAM interventions and tailor them around the complexity of the individual patient’s needs, knowledge of the physiology involved in these interventions is mandatory. Moreover, this approach may be a valuable research tool in the attempt to analyze complex and whole medical systems as they currently exist.

We are well aware that others have previously made useful and explicit statements in relation to the integration of meaningful biomarkers into clinical trials [2, 10, 13]. We wished to continue and extend this discussion in order to provide strategies on how this can be done most appropriately, guided both by the symptom pattern and the particularities of the CAM intervention chosen.

The Northern Lights Workshop in Sommarøy, 2012, was supported by a grant from the faculty of health sciences, UiT, The Arctic University of Norway, Tromsø, Norway. We are thankful to Åsa Sohlen for indispensable technical support.

The authors declare that there are no competing interests.

This paper is the result of a workshop which was part of NAFKAM’s 2012 Northern Lights Workshop in Sommarøy, Tromsø, Norway. Therefore, all authors were involved in the development of the methodology and the concepts presented, and in the drafting of the overall paper outline. F.M. and S.M. developed the outline of the workshop. C.K. provided the conceptualization of the graphs. S.M., C.K., and S.W. provided specifics for the examples. All authors revised the manuscript carefully and approved the final version of the paper.