Explanatory models are needed to integrate RCT and observational data with the patient's unique biology

Abstract

In this review, we make the case for evidence-based medicine (EBM) to include models of disease underscored by evidence in order to integrate evidence, as it is currently defined, with the patient's unique biology. This would allow clinicians to use a pathophysiologic rationale, but underscoring the pathophysiological model with evidence would create an objective evidence base for extrapolating randomized controlled trial evidence. EBM encourages practitioners not to be passive receivers of information, but to question the information. By the same token, practitioners should not be passive executors of the process by which information is generated, appraised and applied, but should question the process. We use the historical examples of the evolution of EBM to show that its subordination of a pathophysiological perspective was unintentional, and of essential hypertension to illustrate the importance of disease models and the fact that evidence supporting them comes from many sources. We follow this with an illustration of the benefits a pathophysiological perspective can bring and a suggested model of how inclusion of pathophysiological models in the EBM approach would work. From a practical perspective, information cannot be integrated with the patient's unique biology without knowledge of that biology; this is why EBM is currently so silent on how to carry out its fourth stage. It is also clear that, regardless of whether a philosophical or practical definition of evidence is used, pathophysiology is evidence and should be regarded as such.

Introduction

‘Grant me the strength, time and opportunity always to correct what I have acquired, always to extend its domain; for knowledge is immense and the spirit of man can extend indefinitely to enrich itself daily with new requirements.’ The Oath of Maimonides¹

Knowledge provides a basis for a belief or action. In philosophy, it is usually defined as a group of true, justifiable and actionable beliefs. Evidence in this context is anything which provides justification for a belief. Evidence, from a practical perspective, is defined by its purpose; it is information that reduces uncertainty in the context of decision-making. However, this functional definition is constrained by the epistemological framework from which it is derived. Evidence-based medicine (EBM) uses a hierarchy which narrows the definition of evidence to information arising from systematic reviews and randomized controlled trials (RCTs). EBM encourages practitioners not to be passive receivers of information, but to question the information. By the same token, practitioners should not be passive executors of the process by which information is generated, appraised and applied, but should question the process. The price for not doing so is that one's thoughts and actions become shaped by, as MacIntyre put it, ‘unrecognised theoretical ghosts’.² EBM assumes a number of philosophical positions, and it is important for practitioners of EBM to be aware of them and to question them.

If one defines knowledge as a set of true, justifiable and actionable beliefs, evidence can be loosely defined as anything which provides the justification. No single type of evidence can claim to be absolutely foolproof, to always establish ‘truth’. Clinical knowledge draws its justification from a range of sources, which include clinical experience and scientific research, and the weighing of each is a nuanced process which defies any attempt to be simplistically captured. EBM has, however, espoused a method of critical appraisal which is simplistic in its use of an evidence hierarchy which ranks quality based on the presence and quality of randomization and control. This has created an artificial separation between ‘evidence’, which is effectively defined as information from RCTs, systematic reviews and observational studies, and pathophysiology, which is explanatory information about mechanisms of disease. The origins of this separation lie in the philosophical assumptions of EBM methodology which view randomization, control and repeatability as sufficient conditions to establish causality, and the use of a frequentist interpretation of the probability calculus (Table 1). It has persisted partly because the paradigms of EBM have become firmly established during the rapid rise of the EBM movement, and partly because a perceived benefit of EBM has its challenge to expert opinion which has, historically, been inextricably linked with pathophysiological rationale.

Table 1.

Interpretations of the probability calculus

Theory	Interpretation of the probability calculus	Implications of the interpretation
Logical theory	Probability is a degree of belief in a hypothesis or prediction that is held equally by all human beings.	Relates evidence to theory. Implies a hierarchy which places the best explanatory theory as the gold standard.
Subjective theory	Probability is a measure of the degree of belief of a particular individual in a hypothesis or prediction.	Relates evidence to theory, and implies a hierarchy which places the best explanatory theory as the gold standard.
Frequency theory	Probability is the limiting frequency with which an outcome occurs in a long series of similar events.	Relates events to each other. Implies a hierarchy which places RCT evidence as the gold standard.
Propensity theory	Probability is the propensity toward a particular outcome that is inherent in a set of repeatable conditions.	Relates events to each other. Implies a hierarchy in which the gold standard is evidence derived from repeated interventions in an individual or across a group in whom the conditions for each trial are consistent.

RCT = randomized controlled trial

The separation between evidence and pathophysiology has no logical basis. Pathophysiology can provide justification for a belief and reduce uncertainty in the context of decision-making; it therefore fits the definition of evidence, and is evidence. From a practical perspective, an understanding of mechanism (i.e. of the latest models) is important at two stages in the current EBM approach (Table 2). First, when assessing the validity of a finding, it is important to ask whether it is biologically feasible. This was illustrated in a trial of remote intercessory prayer in bloodstream infections which demonstrated effectiveness of the intervention even though it was delivered 4–10 years after the patient's illness!³ It was published as a cautionary example that a good quality trial can yield erroneous results because a single randomization may not balance out all confounders. The argument raised is that the establishment of causality requires an explanatory mechanism. Second, evidence needs to be integrated with the patient's unique biology. However, EBM is silent about how to identify the best models of biology with which to integrate the evidence (as EBM currently defines it); this task has, ironically, been directed back to expert opinion and has been fraught with difficulties. We argue that the adoption of the best disease model, continuously underscored by evidence to inform the reassessment and development of the model, could provide an evidence base for extrapolating quantitative information about benefits and harms.

Table 2.

The current EBM methodology

Step	Action
1	Converting the need for information into an answerable question.
2	Finding the best evidence to answer the question.
3	Critically appraising the evidence for validity, impact and applicability
4	Integrating the appraisal with clinical expertise and the unique biology, values and circumstances of the patient.
5	Evaluating how effectively and efficiently steps 1–4 are performed and seeking ways to improve.

A further unintended consequence of the division between evidence, as it is currently defined, and pathophysiology, is that the former has been prioritized in continuing medical education, evidence databases and clinical decision support tools. In addition, many undergraduate curriculums no longer include a pathology course. A potential consequence of this is that clinicians may be kept up to date about what the latest treatments are, but not about current understanding of their mechanisms or the underlying disease process. Unless doctors seek out and remain up to date about this information on their own (only feasible for a small subset of conditions in which the clinician has a particular interest), this could have two practical implications. First, doctors would not have the latest information to anticipate benefits or harms when deciding whether a treatment recommendation is applicable to their patient. Second, doctors would be unable to provide their patients with the very latest understanding of their disease and treatment. These problems are compounded by the exponential growth of the medical and scientific literature.

The evolution of EBM

It is worth tracing the history of how RCTs and the evidence-based movement came into being, as this history illustrates that the subordination of the pathophysiological perspective was unintentional. The history is summarized in Figure 1.^4–8

Figure 1.

Historical milestones in the evolution of EBM

RCTs emerged by applying the ideas of Ronald A Fisher, who postulated that randomization, control and repeatability are sufficient conditions to establish causality. These conditions for causality have since been superseded by the more robust Bradford-Hill criteria (Table 3), but EBM methodology continues to rely on Fisher's postulates. The originators of RCTs needed to choose the appropriate statistics, and since these all relied on probability, the best interpretation of the probability calculus had to be chosen. Four interpretations of the probability calculus are possible (Table 1), but RCTs from their inception adopted the frequency interpretation. This was mostly a practical decision; the interpretation allows events to be related to each other and is therefore well-suited to the study of outcomes and the quantification of the harms and benefits of different clinical management approaches. The frequency interpretation implies a hierarchy of evidence which places RCTs as the gold standard (Table 4). The logical and subjective theories, by contrast, relate theory to evidence, and, if used, would have implied a hierarchy which placed the best explanatory theory as the gold standard.⁹ The frequency interpretation had the limitation that there was no robust way to extrapolate RCT findings to another population (because there is no model with which to do so) or to an individual (because this is impossible in the frequency interpretation); clinicians could only inform their patients about probabilities of outcomes. The adoption of Fisher's postulates and the frequency interpretation made the methodology of EBM the only one in the scientific world to give the lowest rather than the highest priority to the development of a good explanatory theory. The extent to which this has been a strength or a weakness has been much debated. The appeal of the double-blind RCT was that it promised to put clinical decision-making on an objective, more scientific basis by replacing clinical judgement with statistics and standardized methodology. The primary concern motivating this change was the need to establish standards of practice. Its methodology offered the promise of an accepted standard to which treatments could be held, and the need for this became increasingly apparent as more and more treatments started to appear. Its ability to address problems of bias was also very appealing.

Table 3.

The Bradford-Hill criteria for the establishment of causality, using insulin as an example

Criterion	Example
Strength of association	Hyperglycaemia in type 1 diabetes is strongly associated with deficient insulin production. Insulin administration is strongly associated with a decrease in glucose levels.
Consistency	The association is consistently found in many studies.
Specificity (the cause leads to a single effect not multiple effects)	Insulin consistently lowers glucose levels
Temporality (the cause precedes effect with a plausible time gap)	Administration of insulin precedes a fall in glucose levels with a time gap that is consistent with insulin's mechanism of action.
Biological gradient (dose–response relationship)	Higher doses of insulin produce greater decreases in glucose levels.
Biological plausibility (agreement with existing knowledge)	Insulin is known to lower glucose levels by promoting glucose uptake from the bloodstream.
Biological coherence (non-contradiction of existing knowledge)	Glucose-lowering effects of insulin do not contradict existing knowledge about metabolism or endocrinology.
Experimental evidence	The effects and mechanism of action of insulin have been established by an extensive body of basic research.
Analogy (similar causes known to produce similar effects)	A range of hormones exist which enhance insulin action or produce similar glucose-lowering effects.

Table 4.

An example of an evidence hierarchy. Many hierarchies have been defined, and none are universally accepted, but all rank evidence in the same general order

Level	Type of evidence
Ia	Evidence obtained from meta-analysis of randomized controlled trials.
Ib	Evidence obtained from at least one randomized controlled trial.
IIa	Evidence obtained from at least one well-designed controlled study without randomization.
IIb	Evidence obtained from at least one well-designed quasi-experimental study.
III	Evidence obtained from comparative, correlation or case studies.
IV	Evidence obtained from expert committee reports, opinion statements and/ or clinical experience of a respected authority.

As the system for developing drugs evolved, regulatory agencies came to require RCT evidence of a drug's effectiveness. At the same time, the pharmaceutical industry recognized the persuasive power of positive RCT evidence in their marketing, and the threat, as they perceived it, of negative evidence. This development meant that many RCTs were designed and conducted to bring a drug to market; while some of these also answered clinically important questions, some did not, and opportunities to study clinically important questions were lost. More worryingly, the need to accurately and comprehensively report all trial data, positive or negative, came into conflict with the financial interests of the industry, posing a threat to the balance and reliability of the knowledge base. Finally, the use of RCT data for advertising contributed to the trapping of many new therapies into a cycle of hype and disillusionment which precluded a rational consideration of the evidence. RCTs are not exclusively run by industry, but the contribution of industry to RCT evidence has been substantial, as running good-quality trials is resource-intensive. Other biases which emerged included technological (methodological), gender, cultural and publication biases.^10–13 The dream of an unbiased evidence base, envisaged by its early pioneers, remains to be fully realized.

The adoption of the methodology of randomization and control brought the consequence that the pathophysiological perspective was subordinated. There are some celebrated and often-cited justifications for this. A cautionary tale of the effect of confounding factors is provided by the use of HRT to prevent coronary artery disease. This intervention had a solid pathophysiological rationale, and showed benefit in observational studies. However, RCT evidence established that HRT has no impact on cardiovascular risk.¹⁴ Pharmacologic suppression of arrhythmias after myocardial infarction was found to increase the risk of death.¹⁵ In acute lung injury, improvements in oxygenation with inhaled nitric oxide did not improve survival.¹⁶ These examples show that pathophysiological rationale, when acting alone, does not always get it right, and RCT evidence can establish what the actual benefits and harms are. However, these examples provide a case for testing pathophysiological rationale (the basis of all good science), not abandoning it.

The pharmaceutical industry first adopted and then abandoned the approaches of experimental pharmacology and pathology, instead deciding to create high-throughput pipelines to develop compounds which bind to receptors of interest. As a result, disease models were abandoned both at the conception of the treatment (by the pharmaceutical industry) and at its final test (by the EBM movement), being consigned to the phases in between. This approach has yielded few new treatments, and we have learned the hard way that translational medicine requires the use of models if it is to succeed. This fact has now been recognized and there is a concerted effort underway to ground translational medicine in pathophysiological and pharmacological approaches. However, the same has not happened in the EBM movement, even though it faces an analogous challenge: to extrapolate and integrate the information it generates about benefits and harms.

There is no doubt that the EBM movement has had notable successes. The death rate from cardiovascular disease in the US almost halved between 1980 and 2000; half of this reduction was attributed to evidence-based therapies.¹⁷ Guidelines based on EBM have helped to achieve a large and sustained decrease in the morbidity and mortality produced by anaesthetics.¹⁸ Conversely, groups who are under-represented in clinical research experience worse outcomes. EBM has also increased the uptake of new therapies. It is worth commenting, as an aside, that EBM is curiously unsuited to examination by its own methods, because it would be hard, if not impossible, to do an RCT with EBM as the intervention. The success of EBM is, ironically, an argument against the rigidity of its evidence hierarchy. All of the evidence in favour of it would be rated in the hierarchy as ‘poor-quality’, and yet proponents of EBM would not believe that the significance of their achievements is diminished by the manner in which those achievements have been measured. We do not deny the successes of EBM and do not propose that the methods of EBM need to be abandoned. Rather, we propose that they need to evolve to be more inclusive and nuanced. To illustrate why, it is worth considering some lessons from history about how knowledge is generated.

Knowledge generation: vignettes from history

The practice of medicine has traditionally relied upon the construction of a model that explains how the human body functions, how that function is disturbed and how it can be restored. The model is articulated and advanced by a combination of inspired insights, experiment and the slow accumulation of experience. Ever since the Zhou dynasty in China and the teachings of Plato and Aristotle in Ancient Greece, questions of how to assess and affirm medical and scientific aims have been addressed to expert opinion, and so it is that expert opinion has had a profound influence on the development of the knowledge base. Throughout the history of medicine its influence has been mixed, capable of producing great advances but also of resisting advances and creating setbacks. We have chosen the historical example of essential hypertension as an illustration that ‘best evidence’ can come from a myriad of sources which we ignore at our peril.¹⁹ Figures 1, 2 and 3 summarize the key historical milestones.^20–38

Figure 2.

Historical milestones in the early evolution of concepts relating to essential hypertension, covering the period from the early concepts developed in antiquity to the Enlightenment

Figure 3.

Historical milestones in the development of a treatment strategy for essential hypertension in the modern and postmodern periods. Parallel advances in epidemiology, pathophysiology and pharmacology are laid out

The Philosopher of Science Thomas Kuhn first coined the terms ‘paradigm’ and ‘paradigm shift’ to describe how scientific advances occur,³⁹ and the history of science shows that it is easier to articulate or advance a paradigm than to challenge it. As Max Planck once put it, ‘a new scientific truth does not triumph by convincing its opponents and making them see the light, but rather because its opponents eventually die, and a new generation grows up that is familiar with it’. This dynamic can be clearly seen in the paradigm shift that occurred from ‘hypertension the essential adaptation’ to ‘hypertension the risk factor’ which took almost 100 years to complete. The challenge to the old paradigm came from many different sources: expert opinion, non-expert opinion (Korotkov was a trainee when he published his landmark paper, and he never published again), science that relied on the development of models, RCTs that abandoned models and, every now and then, serendipity. Patient preference played its part in creating the demand for new treatments; commercial interest played its part in financing them. No one source could claim to be the supreme source of evidence. No single group could claim to be the sole mover of progress. The most important lesson about this history is that the development of the correct explanatory mechanism was crucial not only to develop effective treatments for hypertension, but to make the case for treating hypertension in the first place.

Explanatory models underscored by evidence: an evidence base for extrapolation

A major disadvantage of RCTs is that, by abandoning the use of models, extrapolation of their findings to other populations and situations becomes challenging, as does the anticipation of unexpected or unintended effects. This is the problem of ‘external validity’, or of establishing the capacity of the treatment to have an effect in a dose, formulation and context other than those tested by the trial.⁴⁰ Extrapolation of findings may relate to differences in dose, timing, duration or a wide range of patient factors including age, co-morbidities and polypharmacy.

Thompson has proposed that evidence should be used to continuously inform and develop the best explanatory model.⁹ Mathematically this can be done by employing Bayesian statistics. This is a statistical approach which relates theory to evidence. We propose that this approach can be used to provide an objective basis for integrating information from RCTs and observational studies with the patient's unique biology. In this approach, the gold standard would be the best explanatory model of the disease and the mechanism of action of the treatment. Evidence underscoring this model, drawn from the full spectrum of basic, translational and clinical research, would be used to continuously update, refine and challenge the model. The model would be available for practitioners of EBM to use in deciding when and how to extrapolate evidence to other populations. The existing evidence base of systematic reviews, RCTs and observational studies would exist alongside explanatory models. Nuanced decisions would need to be made about when to rely on the model to extrapolate information derived from another population, and when to repeat studies in the target population. The development of the best explanatory model would be applicable to any condition and would not be limited by how common the condition is. It could therefore act as a source of evidence where none currently exists. The development of the best explanatory model is the focus of biological research; the systematic use of Bayesian methods would provide a method of testing the best explanatory theory which is independent of expert opinion, thereby preserving this perceived benefit of EBM while broadening its conception of evidence. The potential benefit of this approach in a range of different situations is illustrated in Tables 5–7.^41–52

Table 5.

Benefits of the use of disease models in three clinical settings

Situation	Example	Details	Benefit of a disease model
Treatment decisions in clinical practice	RCT evidence is either unavailable or is not the best type of evidence to use	Clinicians frequently have to rely on pathophysiological rationale and clinical experience to deal with situations poorly represented by the existing evidence base. Examples include the elderly, patients with comorbidities and drug interactions. Idiosyncratic benefits of certain treatments may also be encountered; e.g. in the ICU setting, individual patients may be found to have a better response to one vasopressor or another, even though RCT evidence shows no difference between them at a population level.	A well-developed continuously updated model would provide an objective evidence base for these kinds of management decisions.
Drug dosing	Anticipation of the dose–response relationship	RCT and meta-analyses may sacrifice applicability of their findings in their quest for robustness. For example, a meta-analysis may establish in general terms that a treatment is effective, but not which dose is effective.	An underlying model may be able to anticipate the dose-dependence of benefits and harms. Methods currently employed to select first-in-human doses for clinical trials draw on experimental and clinical data as well as pharmacokinetic and pharmacodynamic modelling. Similar approaches are also being used to inform dose extrapolation between different age groups in humans. The same approaches could easily be adapted to anticipate the dose-dependence of benefits and harms for which only one or a limited number of doses are assessed by RCTs.
Diagnostics	Multidisciplinary team meetings in the UK	Multidisciplinary team meetings (MDTs) in the UK were set up to reduce regional variations in cancer outcomes. The multidisciplinary team consists of an oncologist, radiologist, surgeon, pathologist and the treating physician. At the meetings, the clinical presentation and imaging are discussed, the pathologist presents the results of pathology findings and provides a perspective on the mechanism of disease before the treatment plan is decided.	Discussions of this kind show that there is a real clinical need for a pathophysiological perspective. The clinical practice of the pathologist is informed by evidence which plays a role in developing minimum reporting standards for pathology specimens so that the best possible information is available to inform diagnosis, treatment and prognosis. These standards require knowledge of the disease mechanism, and pathologists provide a perspective firmly based in pathophysiology.

Table 7.

Benefits of the use of disease models for patient education and new knowledge generation

Situation	Example	Details	Benefit of a disease model
Improved patient education	Obesity	The mechanism of the link between obesity, diabetes and cardiovascular disease is informative and worth highlighting to patients. Adipose tissue is not merely a storage site for fat, but an endocrine tissue which secretes a range of hormones and both pro and anti-inflammatory cytokines. When the storage and metabolic capacity of the tissues for lipids is exceeded, fatty acid intermediates accumulate in the cytoplasm (lipotoxicity) causing tissue damage. In obesity, the balance of secretion of pro- and anti-inflammatory cytokines is shifted towards a pro-inflammatory state which becomes chronic. The consequences of these effects are wide-ranging and complex.	These mechanisms (which are still being worked out) help to disentangle negative perceptions of body image, which fuel a desire to lose weight for the sake of appearances, from medical understanding of obesity as a risk factor, which aims to prevent disease. They also explain why some patients who are not overtly overweight can still develop the metabolic syndrome, diabetes and cardiovascular disease. This example highlights the importance of clinicians remaining up to date about mechanisms so that they can educate their patients. Obesity has been the focus of intense research and is a rapidly advancing field.
Knowledge generation in tandem with RCT evidence	Beta-blockers	RCT evidence can produce paradigm shifts, even though this is not its primary aim. Based on an understanding of their acute pharmacology (a reduction in the heart rate and force of contraction), beta-blockers were contra-indicated in heart failure as they would be expected to worsen it. The first report challenging this view was a case series in 1975. The observation of excessive beta-adrenoceptor stimulation and reduced beta-adrenoceptor density in cardiac failure patients convinced some researchers that beta-blockers may be beneficial. A series of RCTs established beta-blockers as cornerstones of heart failure management.	Further study revealed that beta-adrenoceptors were not just involved in short-term inotropic and chronotropic responses of the heart, but that the downstream signalling could switch to regulate responses such as cardiomyocyte fibrosis and apoptosis in the long term. Therein lay the mechanism of the benefit. This example shows that RCT evidence itself can inform the development of a model. Adopting an approach in which the current EBM methodology exists beside the continuous development of disease models provides fertile ground for both approaches to inform and challenge each other's perspectives.
Knowledge generation superseding RCT evidence	Systems biology	New frontiers are now being opened which challenge RCT evidence on a fundamental level. RCTs are reductionist, and a challenge to the conventional reductionist approach has been posed by systems biology. This approach allows the experimenter to look globally at an entire system.	Many physiological systems have been found to exhibit behaviours which are ‘emergent’ and cannot be anticipated based on a collective understanding of their parts. Accounting for these effects not only requires the development of a model, but requires that evidence informing the model include evidence from a systems biology perspective.

Table 6.

Benefits of the use of disease models in the practice of EBM

Situation	Example	Details	Benefit of a disease model
Anticipation of harms	Drug-eluting stents	Drug-eluting stents were designed to reduce restenosis, the major cause of stent failure occurring as a result of an excessive healing response. Drug-eluting stents include cytostatic or antimitotic agents. There are now concerns that these stents increase the risk of thrombosis, and one proposed mechanism is that by inhibiting cell division, drug-eluting stents prolong the pro-thrombotic stimulus of an exposed endothelium at the stent site.	If extrapolation of the evidence could be informed by an explanatory model of wound healing and thrombosis, the risk of thrombosis could be anticipated, and the conditions under which the harms caused by prolonging exposure of the endothelium outweigh the benefits of controlling the healing response could be worked out.
Anticipation of harms	Rosiglitazone	Rosiglitazone was withdrawn because of the identification of adverse cardiovascular effects, believed to be at least partially mediated by its effects on remodelling. The adverse effects of rosiglitazone came as a surprise; they need not have if a complete model of PPAR-γ action had been used to anticipate them.	PPAR-γ is involved not only in the regulation of insulin signalling, but also in remodelling. This immediately raises the possibility that PPAR-γ agonists could have remodelling effects which might be harmful. The drug and dose chosen would need to selectively stimulate the insulin-sensitizing effects while minimizing the other effects of PPAR-γ stimulation.
Anticipation of benefits	Antioxidants	A wealth of experimental work suggested that antioxidants would be of great benefit in reducing the risk of cardiovascular disease, but RCT outcomes ranged from equivocal to negative. These results raised many questions about species differences, dose extrapolation and the timing of treatment. The trials could be criticised for using inappropriate doses and examining patients too late in the disease process. The basic science could be criticized for failing to account for species differences.	This is a situation in which detailed attention to the development of a model can help inform both the optimal timing of the best dose (for clinical studies) and the selection of an appropriate therapeutic target (for basic and translational studies). Few researchers doubt the importance of oxidative stress as a disease mechanism, but if it acts as an initial trigger, giving antioxidants may be ineffective once the downstream events have been initiated and perpetuated, and the pathways activated by oxidative stress may yield better therapeutic targets.

Conclusion

Evidence is, and should be, viewed as anything which provides justification for a belief (its philosophical definition), or which reduces uncertainty in the context of decision-making (its practical definition). The examples discussed in this review show that the artificial separation between pathophysiology and evidence has no logical basis. In addition, pathophysiology has a practical benefit to offer by informing the integration of evidence, as it is currently defined, with the patient's unique biology. The adoption of Bayesian approaches to inform pathophysiological rationale would help to place it on an objective basis, thereby preserving the independence from expert opinion of the EBM approach. We do not expect that this approach will be a panacea for the limitations of EBM, and the ultimate test will be whether the approach improves patient outcomes. There is probably no single answer to the question of whether the benefits of EBM seen so far would have been greater or lesser if EBM had used an alternative approach or defined its hierarchy differently. Nevertheless, from a practical perspective, information cannot be integrated with the patient's unique biology without knowledge of that biology; this is why EBM is currently so silent on how to carry out its fourth stage. It is also clear that, regardless of whether a philosophical or practical definition of evidence is used, pathophysiology is evidence and should be regarded as such.

DECLARATIONS

Competing interests

None declared

Funding

None

Ethical approval

Not applicable

Guarantor

RM

Contributorship

Both authors contributed equally

Acknowledgements

None