Methodological Pluralism in Surgical Evidence: from Concept to Practice Authors: Frédéric Dumont, Charlotte Bourgin, Timothée Vignaud Published: 2026 ABSTRACT AbstractEvidence-based surgery has traditionally been framed through drug-development paradigms in which randomized controlled trials (RCTs) occupy a central position. Yet surgical innovation raises specific methodological challenges, including operator dependence, evolving techniques, learning curves, centre effects, and the centrality of patient-centred outcomes. This perspective article examines how scientific evidence in surgery has been constructed historically at the intersection of mechanistic reasoning and empirical observation, and argues that contemporary practice requires a pluralistic approach to methodology. Drawing on the history and philosophy of medicine, we trace the emergence of modern evidence-based medicine and the limitations of transposing its conventional hierarchies uncritically to surgical care. We then discuss complementary frameworks for evidence generation in surgery, including the IDEAL model for staged evaluation of innovation, prospective observational studies and registries, Bayesian methods, artificial intelligence and big-data approaches, qualitative and mixed-methods research, and innovative trial designs adapted to surgical realities. Rather than opposing these approaches to randomized trials, we propose a coherent model in which different methods address distinct dimensions of “surgical truth”, from physiopathological plausibility and causal inference to implementation, patient experience, and real-world effectiveness. In this perspective, scientific validity in surgery is best approached through the integration of multiple forms of evidence within adaptive methodological pathways, particularly in oncologic surgery where interventions are complex, innovation is rapid, and the balance between survival, function, and quality of life is central.Keywords: Evidence-based surgery Randomized controlled trials IDEAL framework Bayesian statistics Big data and artificial intelligence Surgical innovation and methodology Introduction Establishing the reliability of observed facts in the life sciences and in surgery is inherently complex. Scientific truth is progressively built at the intersection of mechanistic understanding, empirical testing, and clinical usefulness, and must therefore be interpreted as a robust and revisable form of knowledge rather than as an absolute or context-free certainty. Randomized controlled trials (RCT) remain a cornerstone of therapeutic evaluation, particularly when the objective is to estimate the average causal effect of an intervention under controlled conditions. However, surgery poses specific challenges, including operator dependence, procedural evolution, learning curves, contextual adaptation, and the importance of patient-centred outcomes, which often require additional forms of evidence. For this reason, methodological pluralism in surgery should not be understood as a rejection of randomized evidence, but as an organized complementarity between trial methodology and other approaches. In practice, different questions call for different designs. Early surgical innovation may be better assessed through staged IDEAL evaluation, technical reporting, and prospective observational studies; comparative effectiveness may require randomized trials; implementation and adoption may be clarified by registry data and qualitative research; and complex or low-signal clinical questions may benefit from Bayesian analyses and carefully validated artificial intelligence tools. Such an approach is particularly relevant in oncologic surgery, where innovation is rapid, interventions are complex, and the balance between survival, function, and quality of life is central. The future of evidence-based surgery therefore depends less on defending a single hierarchy of proof than on building coherent methodological pathways adapted to the question being asked. This requires academic structures, funding strategies, and regulatory frameworks that support the full evaluation of surgical innovation, from mechanistic rationale to real-world implementation, while preserving both scientific rigor and clinical relevance. Ideologies, social media, or newsworthy innovations give value to subjective opinions and easy, yet poorly constructed knowledge, to the detriment of carefully built-up scientific facts. Surgery must justify its role and value via rigorous, adaptive scientific approaches1 This manuscript is a perspective review aiming to understand the construction of scientific evidence in surgery by exploring the philosophy of science, the history of medicine, and emerging scientific and methodological approaches. The perspective review uses a multidisciplinary approach. The literature search was selective and inherently subjective, based on the major schools of thought that have shaped the scientific approach to surgery, predominantly within the Western tradition. Philosophical concepts Two major Western philosophical currents have exerted considerable influence on conceptions of truth: realism and empiricism. Over the history of medicine, these concepts have coexisted, interacted, opposed each other. Realism Realism sees the outside world as real, independently of its observer. Proof of the existence of an observed fact is not reasonable as it is independent of the outside world. Phenomena follow cause and effect principles, leading in medicine to the animal-machine2 mechanistic model supported by Descartes (1596-1650). Understanding the pathological mechanisms of diseases makes it possible to deduce treatments rationally. This theoretical realist reasoning made it possible to develop a theoretical understanding of anatomy and physiology, the key principles in surgery, and ones that have progressed considerably. It is important to distinguish this philosophical realism from the contemporary notion of “real-world evidence”, which refers to empirical data collected in routine care rather than to a metaphysical position about reality Empiricism Classical empiricism, as articulated by David Hume (1711–1771), questioned the assumptions of realism and considered that it is necessary to define that which can be known before considering that which exists. The mechanisms of the reality perceived by our senses are unknown3. Empiricism literally means testing something and is thus based on experience. It is necessary to analyze from practical observations4. For early empiricists, it was necessary to accept that we live in a world that remains partly unknown and elusive to human understanding unknown5. Medical empiricism observes a series of cases that are subjected to reasoning that takes the unknown world into account. Using a mathematical probability model thus becomes the best suited tool, making it possible to make reasoned deductions within a great many physiopathological uncertainties. As they have come into contact, these two concepts have evolved and influenced each other. Modern evidence-based medecine did not emerge through the replacement of mechanistic reasoning by empirical observation, but rather through repeated attempts to integrate the two. Biological plausibility remains essential for generating hypotheses, understanding causal pathways, and refining interventions, whereas systematic clinical observation and formal comparative studies are required to test whether these hypotheses translate into meaningful patient benefit. In this sense, contemporary evidence-based medicine should be understood less as the triumph of one epistemological tradition over another than as a pragmatic framework that seeks to articulate theoretical coherence with empirical verification. The History of Scientific Proof in Medicine and Surgery Antiquity and the Middle Ages The foundations of scientific reasoning were laid down in Antiquity. Some ancient physicians promoted systematic clinical observation and the subsequent confrontation of these findings with established medical writings.6 The Middle Ages were a period of strong reliance on textual authorities and philosophical argumentation, in which empirical testing of medical claims played a limited role 7. This period laid the foundations for a literate, structured, and holistic medicine, including the creation of a university curriculum grounded in antique medicine8. Medical thinking had its own dogmatic rationale and truth went largely uncontested. This uncertainty was in opposition with the definition of a scholar at the time, “one who knows” for his knowledge was believed to be ordained by God.9 The Renaissance and the Enlightenment (16th-18th centuries) From the late Middle Ages to the Renaissance, empirical reality and direct observation of phenomena progressively took precedence over purely textual authority. Pioneers such as Andre Vesale (1515–1564) considerably improved anatomical knowledge and used a scientific approach to correct past errors, while William Harvey (1578–1657) exemplified this epistemological shift by demonstrating the circulation of blood through systematic observation and experimental dissection, explicitly moving away from learning “from books” and “from philosophers’ opinions” towards learning “from the fabric of nature”. This new empirical knowledge was often confronted, sometimes violently, with established authorities holding radically different views10, yet it proved crucial for the progress of surgery. The cultural movement of the Renaissance truly gathered momentum in the 17th century, partly thanks to the development of printing. The British scientist and philosopher Francis Bacon (1561–1626) foreshadowed empiricism by theorizing ordered observation and suggesting a methodology that sought to reduce subjectivity11, although the uncertainty inherent to empiricism remained difficult to accept at a time when the efficacy of treatments was still modest. In this context, surgeons such as Ambroise Paré (1510–1590), a barber-surgeon, represent early figures of surgical empiricism. By systematically describing his wound treatments, abandoning cauterization with boiling oil and allowing accumulated clinical experience to override established dogma12, Paré showed how careful observation of outcomes could drive therapeutic change even in the absence of a fully articulated theoretical framework. Barber-surgeons were among the first to justify their results with figures, yet they remained socially despised by scholars, who routinely condemned surgical charlatans13 performing at fairs and marketplaces14 and portrayed them as uneducated quacks relying on techniques devoid of recognized theoretical foundations. Despite this, barber-surgeons gradually gained public trust, to the point that one of them was appointed to incise Louis XIV’s anal fistula at the royal court in 1686. Such episodes illustrate how tangible surgical successes, grounded in simple outcome observation, could gain legitimacy and disseminate despite strong doctrinal resistance. The 18th century deepened this movement by introducing explicit skepticism toward past knowledge and dogmatic principles derived from divine intervention. The mysterious nature of life was recognized as partly beyond human understanding, and ignorance of treatments was assumed15, which paradoxically opened space for empirical clinical studies built on rudimentary but systematically applied techniques. The study carried out by James Lind in 1747, in which twelve sailors suffering from a common disease were divided into six groups of two and exposed to different treatments, with only those receiving lemon juice recovering quickly, is often cited as an early controlled experiment. Yet Lind himself did not fully endorse the idea that the disease was caused by a lack of fruit16. The degree of certainty in medicine remained under debate17. Probabilistic deductions seemed reasonable, but this tool was in its infancy and much too recent to be seen as a scientific reference18. The discovery of vitamins by Frederik Hopkins in 1912 brought the scientific debate to a close. The scientific community had needed to understand before it could accept realistically that these sailors had been suffering from scurvy caused by a lack of vitamin C. The Modern Period (19th-early 20tth Centuries) The 19th century and the first half of the 20th century were times of great innovation, both medical and surgical. Many of these advances mainly driven by technical progress rather than by a philosophical or methodological paradigm shift in the way reality was approached. The broad technological progress such as anesthesia, antisepsis, asepsis, improved hemostasis, and better instruments expanded what surgeons could safely do in the operating room19. The microbiology of Louis Pasteur and Robert Koch, or Rudolf Virchow’s cellular pathology transformed medical knowledge and surgical practice. The causes of the great ills of the time, such as tuberculosis, were identified. The first major treatment successes that resulted from these discoveries, such as vaccination for example, came into being without complex mathematical tools. Rationalism took precedence over empiricism. This period was also marked by numerous initiatives to structure observed data in order to evaluate practices, such as counting deaths after amputation during the Napoleonic wars.18 Debates emerged between practitioners who championed numerical measurement and those who favored a more qualitative approach, with the latter criticizing the former for turning medicine into a matter of figures alone18. The clinician Armand Trousseau affirmed in 1861 that the statistical approach was the death of intelligence. Surgery moved from being on the surface to surgery of the deep organs, a real revolution in the field, leading Theodor Kocher (1841-1917) to say in 1909 at his Nobel Prize acceptance speech that “It has become possible within less than half a century to expose all the organs of the body, brain and heart not excepted, without danger, and carry out the necessary surgical measures on them”20. Validation of this progress was limited to descriptive statistical analyses. The experience of Ignaz Semmelweis (1818–1865) is emblematic of this tension. He documented that hand‑washing reduced maternal mortality from around 9.9% to 3.4% in obstetric wards, a difference that was mathematically compelling, yet his conclusions were rejected by contemporaries who adhered to alternative theories of “childbed fever”. This episode illustrates that strong quantitative evidence alone is insufficient to change practice when it conflicts with prevailing theoretical and institutional frameworks. On the other hand, analytic biostatistics still had little success, and their results were often contradictory and unconvincing. The Contemporary Period (20th-21st Centuries) In surgery, the great successes of the late 19th and early 20th centuries sometimes evolved into a form of dogmatism, with “schools of surgery” whose authorities defended the knowledge they had. In breast cancer surgery, the opposition between the speculative affirmations of the school of William Halsted (1852–1922), defending the radical mastectomy21,22 on the basis of an anatomical model of orderly lymphatic spread, and his detractors who favored more conservative surgery, was settled by an RCT published in 198523. This debate illustrated that therapeutic decisions were driven more by fear, hope, and the cultural context than by robust evidence, and medical truth was subject to negotiation.22 Fifty years was a long time for solving a scientific debate and showed that a reference tool was needed to determine uncontested realities. This was the RCT. In the post-Second World War period, the philosopher of science Karl Popper (1902–1994) had a strong influence on the scientific community and favored the empirical statistic methodology. He formalized a falsificationist5 view of scientific reasoning, which resonated with the emerging empirical and statistical methodology. For Popper, the defining characteristic of a scientific statement was its falsifiability, that is, the possibility that it could be refuted by empirical observations. Scientific statements were never definitively proven; rather, they gained credibility when repeated attempts at refutation failed. This logic parallels the use of null-hypothesis testing in biostatistics, where evidence accumulates against a hypothesis without ever establishing absolute truth. The Popper’s work can be seen as providing a philosophical articulation of principles that were already implicit in the growing use of clinical trials and statistical testing. Thomas Kuhn (1922-1996) also profoundly reshaped the philosophy of science by introducing the idea of scientific progress not as a purely cumulative and rational process, but as a succession of paradigms separated by periods of crisis and scientific revolutions. He argued that observations are theory-laden, that the sharp distinction between discovery and justification breaks down, and that science is cumulative only during periods of normal science. His approach offers a more satisfactory description of actual scientific progress, at the cost of diminishing the role of formal rationality and leaving issues such as hypothesis confirmation, scientific explanation, and causation largely unaddressed24. At the methodological level, in the 1950s, Ronald A Fisher (1890-1962), and the epidemiologists Austin Bradford Hill (1897-1991) and Archie Cochrane (1909-1988) established the legitimacy of scientific rigor in medecine. With the enormous success of his 1972 book Effectiveness and Efficiency, Archie Cochrane popularized the idea that therapeutic decisions should be based on controlled trials and systematic reviews rather than on authority or habit.25 Bradford Hill was one of the first contemporary physicians to use randomized controlled trials 26. Together, these contributors revitalized biostatistical culture and promoted RCT as central tools in modern medical methodology27. RCT became predominant in clinical research, to the detriment of rationalistic approaches based on physiopathology. Nevertheless, many of the major advances made in fundamental sciences, such as immunology or oncogenetics, were also key elements of this period. The richness of these discoveries highlighted the complexity and multiplicity of physiopathological mechanisms. The results of a treatment targeting a single pathological mechanism could no longer be predicted and needed to be tested via clinical trials. Hyperthermic intraperitoneal chemotherapy (HIPEC) gained dominance following cytoreduction surgery in the 1990s, based on experimental research showing that tumoral cytotoxicity was improved with hyperthermia and local application of chemotherapy. The treatment of carcinosis progressed significantly during this period, without RCT, and with survival rates rising from 12 months in the 1980s28 to more than 40 months29,30 in the 2010s. Additional fundamental research showed the efficacy of HIPEC was dependent on multiple factors (type, concentration, dose level and dose fractioning of the drug, contact time, dilution of the solute, level of hyperthermia, intraperitoneal pressure, type of tumoral cell line).31 The multiplicity of these parameters made the choice of treatment confusing and mysterious. RCT clarified the value of HIPEC, with benefits in certain conditions (HIPEC with cisplatin in ovarian cancers)30 and not in others (HIPEC with oxaliplatin in colorectal cancers)29. The RCTs in HIPEC have remained rare because funding has not been supported by the pharmaceutical industry and has relied almost exclusively on government sources. The issue of HIPEC clearly illustrates not only the need for RCTs in surgery, but also the difficulties involved in implementing them in this field. Between 2014 and 2017, only 2% of RCTs in oncology were surgical.32 In the 1990s, an American work group33 claiming to act in the name of Thomas Kuhn (1922-1996) defended evidence based medicine as a major change in the medical paradigm. For this group, physiopathological explanations were not enough to make a clinical decision while solid data from clinical research were essential. This prefigured the pre-eminence of biostatistics in the concept of scientific proof and major medical journals made evidence-based medicine a reference. RCTs have become the gold standard for establishing truth in medicine, a status reinforced by political, regulatory, and institutional decisions that govern reimbursement of new treatments and guide therapeutic choices27. While they remain essential for evaluating many therapeutic interventions, particularly drugs, they are often not the most suitable design for etiological, prognostic, or numerous diagnostic questions, where observational studies may provide more appropriate answers. The physiopathological belief that cancers reduce life expectancy or quality of life and that their excision has beneficial effects has proved to have many exceptions. The thyroid and prostate cancer, known to have indolent histological forms, went through a period of overdiagnosis34,35 and surgical overtreatment34,35 based on this theoretical concept. Epidemiological studies have shown a marked discrepancy between the increase in incidence and the stability of mortality from these cancers. Many of these detected tumours would never have led to death but were nevertheless treated. The “truth” in surgery Surgical truth is the correspondence between clinical thought and the reality of things. The history and philosophy of science help us understand the complexity of truth in surgery, whose main dimensions can be summarized as follows. A first dimension is mechanistic or physiopathological truth, which relates to the correctness of the biological and anatomical explanations used to justify a surgical intervention. When a procedure is developed on the basis of embryological anatomy, tumour biology, or haemodynamic models, its plausibility depends on how accurately these models describe the phenomena under consideration. This form of truth is primarily grounded in experimental and theoretical work, including laboratory studies, imaging, and detailed anatomical observations. A second dimension is probabilistic or statistical truth, which concerns the average effects of an intervention in a defined population. Randomized controlled trials, observational cohorts, registries, and meta-analyses aim to estimate how often a given outcome (e.g., survival, complications, recurrence) occurs under different strategies, together with measures of uncertainty. In this sense, truth is expressed through effect sizes, confidence or credible intervals, and probability distributions rather than through absolute categorical statements. A third dimension is clinical or practical truth, which relates to the usefulness and relevance of an intervention for individual patients in specific contexts. An operation may be mechanistically plausible and statistically beneficial on average, yet offer limited value to a frail patient with different priorities or to a health system with constrained resources. Clinical truth therefore incorporates patient preferences, functional outcomes, quality of life, and feasibility in routine practice and is often informed by mixed methods, including qualitative research and outcome-focused studies. A fourth dimension is institutional or normative truth, which emerges when regulatory agencies, guideline panels, and professional societies translate available evidence into recommendations and reimbursement decisions. At this level, truth is operationalised through thresholds of evidence, risk–benefit assessments, and societal values, and may lag behind or diverge from emerging mechanistic or clinical insights. In surgery, these forms of truth do not always coincide and may even conflict. Mechanistic models can be elegant but misleading (e.g., numerous preclinical and animal studies have suggested the efficacy of oxaliplatin during hyperthermic intraperitoneal chemotherapy for peritoneal carcinomatosis, without robust statistical evidence in randomized study)29; statistically significant findings may have limited clinical importance (e.g.; drainage after rectal resection has been shown to be ineffective in preventing complications in a randomized trial, with no subsequent change in clinical practice given the very low deleterious impact of its use36. For this reason, methodological pluralism is not a call to relativism but an attempt to align and reconcile these different dimensions of truth through appropriate combinations of methods. Mechanistic and experimental studies clarify why an intervention might work; randomized and observational designs estimate how often and for whom it works; qualitative and patient-centred approaches illuminate how it is experienced and implemented; and institutional processes determine how it is formalised in practice. Approaching truth in surgery as a robust, revisable, and multi-layered construct rather than as a single hierarchical standard helps explain why no single design, including the randomized trial, can fully capture the complexity of surgical care. It also justifies the need for adaptive methodological pathways that combine several sources of evidence when evaluating surgical innovation, oncologic strategies, and long-term outcomes Scientific integrity The most obvious violation of scientific truth is a lack of integrity and scientific fraud. Integrity and honesty determine the credibility of science and its societal value. In 2020, 2,300 scientific publications worldwide were retracted for fraud37. This retraction has increased since 198038 and has become a major concern. The main scientific journals (The Lancet and The New England Journal of Medicine) have recently been caught out by wide scale and highly mediatized fraud in the creation of big data39,40. The failure of these journals to detect the falsified data can be interpreted as a consequence of structural limitations in conventional peer review, which neither provides reviewers with access to raw data nor permits verification of the logistical plausibility of the reported data flows. In the exceptional time pressures of the COVID‑19 pandemic, combined with an excessive level of trust in the big‑data provider and in scientific integrity41. Scientific fraud is defined by the U.S. Office of Research Integrity (ORI) as fabrication (invention of data), falsification (distortion of data or results), or plagiarism (copying data or words from other sources). The main cause of article retraction in surgery is duplicate publication, which is equivalent to plagiarism. Fraud is by definition intentional and jeopardizes the author’s honor and career, leading to dismissal and even imprisonment. The identified causes of this misconduct include ambition and academic and institutional pressures42. The outcomes movement and patient-centred evidence From the 1980s onwards, the “outcomes movement” reshaped how surgeons evaluated their treatments43. This movement highlights the limitations of relying on pathophysiological considerations as the primary drivers of surgical strategy and reorients outcome assessment toward what truly matters to patients. In oncologic surgery, this shift meant that overall survival could no longer be the sole criterion of success; the quality of that survival and the preservation of function became equally important, making less radical procedures acceptable when they offered equivalent survival with superior patient-reported outcomes. This shift reflects a predominantly empirical approach, rather than one grounded chiefly in the theoretical and physiological aspects of treatment and disease, and has substantially transformed both surgical practice and medical research44. The example of overdiagnosis of thyroid and prostatic cancers illustrates this paradigm shift. A synergy between theoretical knowledge and the empirical evidence provided by statistics has gradually emerged as the necessary combination for approaching scientific truth in surgery. One of the greatest advances in oncologic surgery over the past 40 years, total mesorectal excision (TME), aptly illustrates this evolution. Insights from embryological development and patterns of lymphatic spread in rectal malignancies laid the foundations for adopting TME45. This strategy was validated by multiple empirical studies, except RCTs and became a gold standard in rectal surgery. A New Method for Surgical Research: the IDEAL method In the 1990s, a controversial editorial (“Surgical research or comic opera”) criticized the methodology of surgical studies46 and their lack of RCT. In response, British surgeons and methodologists stressed the limitations of current trials for surgery. Unlike drugs, a surgical technique evolves because of the learning curve and the lack of standardization of techniques. The quality of an operating technique is subject to a wide range of variables, including surgical experience, operating volumes per center, technical progress or the means available. For the same anatomical definition of hepatectomy, the surgical mortality rate has dropped by 80% in 10 years and is now less than 2%47. The truth about hepatectomies is thus different today compared to what it was 10 years ago. Unlike medical research, research in surgery requires better piloting of the innovation and development phases48, adopting a less linear evolutive process49, integrating the process behind the complexity of the procedure and learning the techniques into evaluations50, as well as recognizing that complex procedures can work better if they are adapted to local contexts instead of being completely normalized51. The British Medical Research Council (MRC) thus designed the IDEAL method (Table 1) in the 2000s52 to respond to the need for scientific and ethical rigor, as well as preservation of the discipline’s innovation, all while integrating surgical specificities. Phases I and II of the IDEAL framework formalize the innovation and development of surgical techniques, the poor relations of rigorous methodological evaluation in surgery. These phases must remove the obstacles to phase III, that of the RCT, maintaining its key role as scientific proof. Table 1Development phase for innovations in surgery based on the IDEAL research framework.Table 1 dummy alt text1 Idea2a Development2b Exploration3 Evaluation4 Long-term studyAimProof of conceptDevelopmentLearningEvaluationSurveillanceNumber and type of patientsA single number Rigorously selectedFew, selectedManyExpansion of the indicationManyExpansion of the well-defined indicationAll eligibleNumber and type of surgeonsVery fewinnovatorsFew, innovators and sometimes surgeons accept the technique quicklySeveral, innovators and surgeons accept the technique quicklyMany, a majority of surgeons having accepted the techniqueAll eligibleScientific productionDescriptionDescriptionMeasurement, comparisonComparison, complete information for non-participants in a randomized controlled trialIndividual, description, audit, registry, risk adjustmentProcedural techniqueEvolutive, first casesEvolutive, Prospective developmentEvolutive, refinement, learning communityStableStableMethodStructuredProspective development studyDatabase or feasibility RCTRCT with or without additional modificationsRegistry, national databases, risk adjustmentResults evaluatedProof of concept, technical realizationSafety, technique and procedural successSafety, clinical objectives (specific and graded), short-term results, feasibilityClinical objectives (specific and graded), medium- and short-term results, cost-efficacy analysisRare events; long-term results, quality assuranceEthical agreementSometimesYesYesYesNo Development phase for innovations in surgery based on the IDEAL research framework. Since the IDEAL framework was introduced in 2009, there has been a steady increase in use of the IDEAL framework, particularly in the UK. This growing adoption is especially notable in stages 1, 2a, and 2b, which are specific to surgical innovation and represent a novel methodological paradigm in research.53 Despite its conceptual strength, the implementation of the IDEAL framework remains uneven in practice. Many surgical innovations continue to be introduced into routine care without having undergone all the recommended stages of formal evaluation, due in part to enthusiasm for innovation, funding constraints, challenges in standardization, pressure for rapid adoption of new technologies, or insufficient recognition by public authorities52. The advent of robotic innovative surgery is emblematic of this evolution. It was initially implemented outside the early IDEAL phases (1 and 2a), which contributed to media scandals following severe complications54. The first RCTs failed to demonstrate clear benefits of robotic surgery55,56, yet this did not slow its expansion despite the substantial additional costs. Over time, the long maturation of knowledge and the learning curves required for a new surgical technique have now been achieved for robotic surgery, with recent studies demonstrating its safety57 and/or benefits58. These limitations do not undermine the relevance of the framework, but rather highlight the need for structured implementation at the national level, including the establishment of appropriate regulatory frameworks, the support of mandatory prospective registries, and the development of dedicated funding mechanisms for the staged evaluation of surgical innovations. Such an organization is essential to ensure that the adoption of innovations is grounded in robust evidence, while maintaining patient safety and consistency of practices across the healthcare system. Innovative randomized trial designs in surgery Several randomized trial designs have been specifically adapted to address the structural limitations of conventional RCTs in surgery. Expertise-based or expertise-standardized trials randomize patients to surgeons or teams with established expertise in the procedures being compared, rather than to individual operators at different stages of the learning curve. By standardizing expertise across trial arms, these designs aim to reduce performance bias and make the comparison more reflective of mature practice59. Sham-controlled trials, in which patients are randomized to undergo a procedure versus a sham intervention, can in principle provide strong evidence for the efficacy of technically simple interventions that are associated with high placebo responses. In surgery, however, they raise significant ethical and practical concerns, particularly when the sham procedure entails non-trivial risk without potential direct benefit. For this reason, sham-controlled designs remain limited to highly selected contexts and must be justified by a compelling scientific rationale and robust safeguards for patient safety60. Pragmatic trials seek to evaluate interventions under real-world conditions, including heterogeneity of centres, surgeons, and patient populations, and focus on outcomes that are directly relevant to clinical practice and health systems. By relaxing some of the strict inclusion criteria and procedural standardization requirements of explanatory trials, pragmatic designs can capture the effectiveness of surgical strategies in routine care and support decisions about implementation and resource allocation61,62. Taken together, these innovative designs illustrate that randomized evaluation in surgery does not need to conform to a single rigid template. When combined with IDEAL staging, prospective registries, Bayesian analyses, and qualitative studies, they contribute to a pluralistic methodological ecosystem in which the strengths of randomization are preserved while its implementation is adapted to the realities of surgical innovation and practice. Emerging Approaches to Evidence and Truth In 2007, Rothman63 specified that “our challenges concentrate on methodology because we focus all our attention on solving unsolvable biases or explaining contradictory results”. Rothman was a pioneering methodologist in epidemiology. He forcefully rejected the routine use of significance testing and the conventional p<0.05 threshold, arguing that it promotes a simplistic dichotomization of results into “significant” versus “non-significant” and thereby obscures the more informative reporting of effect estimates and their uncertainty. He also criticized the prevailing tendency to prioritize statistical precision while treating exposure measurement as an afterthought, encapsulated in the idea that “we trade statistical precision for measurement.” Finally, he advanced a deterministic, multicausal view of causation, exemplified by his “causal pies” model, which emphasizes interactions among component causes and highlights the inherent limitations of any single model in capturing the etiologic complexity of chronic diseases. Recent surgical research provides concrete illustrations of this methodological pluralism. Contemporary studies on physiological reserve, frailty, and functional recovery after surgery show how multidimensional preoperative assessment can refine risk stratification beyond age and traditional comorbidity indices, and how prospective, multicenter cohort designs can generate clinically actionable evidence outside the strict framework of randomized trials55 Evidence can be generated through the accumulation of knowledge from retrospective studies, which are then pooled and analyzed in meta-analyses57. These examples highlight that modern surgical evidence is an integrated ecosystem that combines prospective cohorts, and advanced syntheses of heterogeneous data—an evolution that naturally converges with the emerging domains of big data and artificial intelligence. This methodological pluralism remains difficult for public authorities to grasp, as their perception of scientific value is still too heavily focused on randomized trials. For etiologic questions, randomized controlled trials remain the gold standard for assessing the scientific value of intervention questions, but they have several limitations and many alternative methods have been developed. Several methodological adaptations have been proposed to make randomized evaluation more suitable for surgery. Big Data and Artificial Intelligence The term “big data” was first used in 1987 in a non-academic field, and in 1998 in an academic field64. By definition, big data in healthcare denotes massive, heterogeneous collections of digital health information whose scale, complexity, and speed of generation exceed the capacity of standard information systems to store, process, and analyze them effectively.65 Big data in surgery draws on multiple sources, including electronic health records, clinical studies, imaging datasets, and surgical or disease registries.66 Big data is developing exponentially67 through a combination of the growing increase in computer storage capacity, computer calculations, data processing techniques, the digitalization of hospitals and healthcare systems. Analyzing data from a large number of patients has profoundly modified our approach to surgical science, bringing us closer to real life and moving us progressively further away from the biased samples used in studies. This big data can cover a significant proportion of the population of a country, such as the Surveillance Epidemiology and End Results database registry which covers about 45% of the American population68 or the digitalized data of the country’s national healthcare system69. This big data facilitates the discovery of rare phenomena69 , These big data resources facilitate the discovery of rare phenomena and allow computational mechanistic representations of biological systems70, but they are affected by multifactorial sources of inaccuracy issues71. In surgery, big data is mostly used to evaluate the healthcare system72 but still only rarely modifies individual healthcare practices. The veracity of the facts analyzed by big data still remains limited because of the issues surrounding how data is structured or, worse, falsifications. Big data collects large volumes of disorganized, transformed data from multiple different sources73. This data must be processed correctly and cleaned effectively as otherwise there can be reliability issues73. Obtaining coherent, traceable and reliable data is one of the challenges, and the main objective, of modern biomedical research74 The technology for processing big data to extract new knowledge is recent but is now well mastered and is called artificial intelligence (AI), with its most popular field being deep learning through a convolutional neural network. The scientific value provided by AI raises questions within the scientific community. Just like when statistics were in their early stages, the statements produced by AI lack explicability. Humans do not understand, or understand poorly, how AI makes its predictions or its decisions. The computer system cannot explain its reasoning, thus limiting its interpretability.75 The historical examples of Lind and Semmelweis illustrate this epistemological issue perfectly: these two physicians had identified effective treatments, but because their results were not understood, they were not accepted. The “grand scheme of things” of rationalist thought is the blind spot of AI’s analyses. Beyond interpretability, several challenges limit the epistemic value of AI and big-data approaches in surgery. [76], [77], [78] External validity and transportability remain major concerns, as models trained on high-volume academic datasets may not generalize to lower-volume centres, different patient populations, or evolving surgical techniques. Reproducibility is similarly fragile, since complex analytical pipelines depend on specific data preprocessing, software versions, and random seeds that are seldom fully reported79,80 Finally, regulatory oversight of AI-based medical devices is still maturing; wile some models aim to improve understanding of AI reasoning 77,80 , the absence of standardized post-market surveillance raises concerns about performance drift over time. Opaque, non-reproducible, or insufficiently validated models cannot claim the same epistemic status as transparent and verifiable evidence, even when their apparent predictive performance is high. Bayesian Statistics Scientific truth in statistical methodology is still the subject of sometimes intense debate81. Frequentist statistical theories dominate in the health field and are opposed to Bayesian theories which are dominant in cognitive sciences. During the 1980s, Bayesian statistics made only tentative inroads into medical research, with published applications remaining relatively uncommon. Over the same period, Bayesian methods were only marginally incorporated into university curricula in applied medical statistics82. In medicine, Bayesian approaches began to develop in the 1990s, supported by advances in statistical computing, and experienced renewed growth during the COVID-19 pandemic period83,84 helped accelerate this process85. Bayesian methods are particularly well suited to settings characterized by uncertain and rapidly evolving data, in which it is not possible to specify stable, well‑founded prior hypotheses, such as during epidemics85. In Bayesian inference, the parameters being evaluated (for example means, rates, or survival parameters) are not treated as fixed constants determined by hypotheses, but as random quantities whose probability distributions express our uncertainty. This uncertainty is represented either before observing the study data (initial or prior probabilities, which do not depend on the observed data), after observing the study data (revised or posterior probabilities, which are conditional on the observed data), or in relation to future data (predictive probabilities)86. The Bayesian methodology has now been recognized by the FDA87. Frequentist theory states that a statement is either true or false, and frequentist probability is the frequency of a phenomenon that is observable through the repetition of trials in the context of identical systems. The conceptual approach of truth in medicine is different between Bayesian and frequentists. Bayesian statistics assumes an underlying clinical truth about the effect of a treatment, but it treats our knowledge of that truth as uncertain and continually updated in light of new data and prior evidence. Bayesian theory treats probability as a degree of belief about a single underlying reality represented by the available data, explicitly combining prior knowledge with the observed results to update confidence in the truth of a proposition. This leads to inferences that align with how clinicians and patients naturally reason about uncertainty, and allows incorporation of previous evidence on the expected therapeutic effect; Bayesians thus argue that their approach better matches real clinical questions and can detect modest true effects. Frequentist statistics assumes that there is a single fixed true effect of a treatment. In this framework, the truth is binary: the treatment effect either exists or does not exist, and statistical procedures are designed so that, under hypothetical repeated sampling, they will eventually reveal this truth with controlled error rates. Frequentist probabilities do not quantify a degree of belief in the truth but instead describe the long‑run frequencies with which certain data would be observed if that hypothesis were true. Frequentists view it cli realiytas more objective and scientifically rigorous but often demands large, rigid trials that are difficult to conduct.85 Bayesian analyses are pertinent for detecting minimal differences in low power RCT or those that are small scale but with robust previous data (for example, retrospective studies). These situations are common in oncological surgery. Twenty years ago, the study by Stahl et al.88 raised an excellent question. Should patients with esophageal cancer undergo surgery after chemoradiotherapy? RCT compared chemotherapy-chemoradiotherapy followed or not by esophagectomy surgery (SBRT). This study presented many of the biases frequently encountered in surgery, and which decrease the power of a study, such as crossovers, excessive initial hypotheses (often produced to facilitate the feasibility of the study), or early interruption because of a lack of inclusions. In frequentist statistics, the probability that research is “true” depends on its probability of being true prior to the study (Table 2) , the level of significance, and the power of the study89,90. All the biases in the study by Stahl thus compromised the identification of a real therapeutic effect, and the final frequentist analysis did not find any difference in survival between the treatment arms, despite the signs of efficacy from the surgery (best local control). A complementary Bayesian analysis of Stahl’s study91 showed that esophagectomy surgery had a high chance (>70%) of decreasing the absolute risk of death by approximately 5% at 5 years. Both the Bayesian and frequentist methods ultimately lead to the same reasonable conclusion: surgery has a small or minimal benefit. But the medical treatment decision for esophagectomy, shared between the surgeon and the patient, is simpler to establish with Bayesian results than with the dichotomous and inappropriate frequentist conclusion. Bayesian and frequentist statistics frequently produce concordant conclusions, and combining them provides a better understanding of therapeutic effects while strengthening the overall evidence92. The choice between these frameworks should therefore be guided by the clinical question and the available prior knowledge, rather than by doctrinal preference. Table 2Table of the probability of a real benefit of a treatment following a major trial in terms of a priori probability, power and alpha risk. Table taken from the report by the Société Française de Pharmacologie et de Thérapeutique (the French pharmacology and treatment society)90.Table 2 dummy alt textA priori probabilityPowerAlpha risk (unilateral)Probability of a real benefit following a positive trial70%80%2.5%98.7%70%50%30%79.5%20%80%2.5%88.9%20%80%40%33.3%20%50%2.5%83.3%20%50%20%38.5% Table of the probability of a real benefit of a treatment following a major trial in terms of a priori probability, power and alpha risk. Table taken from the report by the Société Française de Pharmacologie et de Thérapeutique (the French pharmacology and treatment society)90. Qualitative studies Qualitative studies arise from a constructivist paradigm in which knowledge reflects realities shaped by individual experiences, representations, and social contexts93. Rather than measuring average effects, these studies seek to understand how patients, surgeons, and healthcare teams perceive disease, risk, innovation, and care pathways. These studies use a subjective approach to the perceptions and experiences obtained from non-statistical data. They are typically based on interviews, focus groups, ethnographic observation, or thematic analysis of clinical narratives, and are particularly useful for addressing questions that remain inaccessible to purely quantitative designs. In surgery, qualitative approaches can illuminate how new procedures are adopted, adapted94, or resisted within teams and institutions; how surgeons experience uncertainty95, responsibility, and technical failure; and how patients weigh survival, postoperative recovery, function, and quality of life when making treatment decisions. They are also relevant to contemporary issues such as frailty, physiological reserve, functional recovery, and the interpretation of quantitative intraoperative technologies, where measurable outcomes alone may not fully capture the lived reality of care. Qualitative studies should therefore not be regarded as competitors to scientific evidence, but as a complementary source of understanding. Qualitative research helps explain why an intervention succeeds or fails in practice, how professional cultures shape innovation, and how patient values influence the clinical meaning of benefit. Qualitative studies do not replace scientific evaluation, and refutation in Popper’s sense is not possible for qualitative data. Rather, they reflect reality and broaden it toward a conception of truth in surgery that is more context-sensitive and centred on the individual. Such studies contribute to the generation of working hypotheses for scientific or institutional purposes. Conclusions A scientific reality in surgery is progressively constructed on two equally important pillars: rationalist thought — "the grand scheme of things," in Claude Bernard's words which seeks to understand physiological phenomena and formulate coherent theories; and empiricist thought "the law of averages," as Pierre-Charles-Alexandre Louis described it which subjects these theories to systematic testing through observation and quantitative data. RCTs remain the gold standard for scientific validation in clinical research, but the evaluation of surgical techniques is often poorly adapted to conventional trial designs, obliging both methodologists and surgeons to develop complementary approaches. Methodological pluralism should therefore be understood not as a slogan but as an operational framework in which each type of question is matched to an appropriate combination of methods. Early surgical innovation is best assessed through IDEAL staging, prospective technical reporting, and observational studies; comparative effectiveness relies on randomized trials, including expertise-based, or pragmatic variants, complemented by Bayesian and registry-based analyses; implementation, adoption, and diffusion are studied through registries, big-data resources, and qualitative investigations; and shared decision-making is informed by both quantitative risk estimates and qualitative insights into patient values and preferences. This structured complementarity aligns the four dimensions of surgical truth: mechanistic, probabilistic, clinical, and institutional and requires dedicated academic, funding, and regulatory structures to support the full evaluation of surgical innovation from mechanistic rationale to real-world implementation. Ethical Standard This review article did not involve studies with human participants or animals. Source of funding The authors did not receive any financial support for this work. Data availability No new data were created or analysed in this study. Data sharing is not applicable to this article. CRediT authorship contribution statement Frédéric Dumont: Writing – review & editing, Writing – original draft, Validation, Resources, Methodology, Formal analysis, Conceptualization. Charlotte Bourgin: Supervision, Methodology. Timothée Vignaud: Validation. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Bibliography Associated Data Data Availability Statement No new data were created or analysed in this study. Data sharing is not applicable to this article.