Causal inference from clinical experience

Philosophical Studies https://doi.org/10.1007/s11098-024-02264-x Causal inference from clinical experience Hamed Tabatabaei Ghomi1 · Jacob Stegenga2,3 Accepted: 1 December 2024 © The Author(s) 2024 Abstract How reliable are causal inferences in complex empirical scenarios? For example, a physician prescribes a drug to a patient, and then the patient undergoes various changes to their symptoms. They then increase their confidence that it is the drug that causes such changes. Are such inferences reliable guides to the causal relation in question, particularly when the physician can gain a large volume of such clinical experience by treating many patients? The evidence-based medicine movement says no, while some physicians and philosophers support such appeals to first-person experience. We develop a formal model and simulate causal inference based on clinical experience. We conclude that in very particular clinical scenarios such inferences can be reliable, while in many other routine clinical scenarios such inferences are not reliable. Keywords Causal inference · Clinical experience · Simulation · Evidence-based medicine, EBM · Multi-armed bandit problem · Computational philosophy · Philosophy of medicine 1 Introduction Maria does not feel well, so she visits her doctor, who diagnoses a disease and prescribes a drug. Maria returns home and starts taking the drug. Time passes, and Maria’s symptoms change. Maria then returns to her doctor, who determines that either Maria has improved or she has not improved, and the doctor’s confidence in the drug accordingly increases or decreases. We want to know how reliable such * Hamed Tabatabaei Ghomi Jacob Stegenga 1 Department of Philosophy, King’s College London, London WC2R 2LS, UK 2 Department of Philosophy, Nanyang Technological University, Singapore, Singapore 3 Department of Philosophy, University of Johannesburg, Johannesburg, South Africa Vol.:(0123456789) H. Tabatabaei Ghomi, J. Stegenga inferences are when deployed across a series of such interactions, representing the putative knowledge that the physician might gain from their clinical experience. Specifically, we want to know if (and if so, under what conditions) first-person clinical experience is a reliable basis for inferences about the general effectiveness of interventions. The so-called evidence-based medicine movement says that such inferences are not reliable, because of phenomena such as the placebo effect and expectation bias that influence both a patient’s experience and a physician’s evaluation of that experience. Instead, says the evidence-based medicine movement, we need to test such causal relations using methods such as randomised trials that minimise the effect of such biases (Howick, 2011). Historically, inferences about the effectiveness of interventions based on clinical experience have led us astray, claims the evidence-based medicine movement. This is part of the rationale for maintaining randomised trials and meta-analyses of such trials at the top of so-called evidence hierarchies, and relegating physician expertise and judgement to the bottom of such evidence hierarchies. Standard guidance from evidence-based medicine methodologists is to assess the effectiveness of interventions only with randomised trials, with explicit guidance that evidence from any other study design, including case reports, can be ignored (Blunt, 2015). The view that clinical experience is an unreliable guide for inferring the general effectiveness of interventions has been widely asserted for decades (e.g.Choudhry et al., 2005; Guyatt et al., 1992; Meehl, 1986). On the other hand, many physicians and patients routinely make such inferences, and some physicians and philosophers have argued that appeals to first-person clinical experience can be reliable evidence for making inferences about the effects of interventions. Tonelli and Shapiro, for example, argue that clinical experience provides what they call experiential knowledge, and such knowledge is important for treatment decisions and assessing the response of treatments—they claim that expertise developed through first-person clinical experience can inform physicians how to “deploy therapeutic decisions in an optimal and individualized manner” (Tonelli & Shapiro, 2020, p. 76). They also claim that “assessing the effect of an intervention is also highly dependent upon the experiential knowledge of the clinician” (p. 76). Similarly, Healy (2011) argues that physicians routinely can make reliable singlecase causal inferences about the effects of interventions based on clinical experience. This view is reflected in some large surveys of physicians’ attitudes to clinical experience (Dewitt et al., 2021) and Cartwright (2017) has articulated a variety of kinds of evidence that can be used to warrant single-case causal inferences. Sometimes the appeal to causal inferences based on clinical experience is made in the context of responding to sceptical arguments about medicine; for example, Stegenga (2018) offered such a sceptical argument by appealing to small effect sizes from randomised trials and biases in those trials, and Healy (2020) responded by explicitly asserting that physicians can reliably observe the effects of drugs based on clinical experience. Healy claims that “(e)verything we have in medicine is built on professional and patient anecdotes. Every discovery of a benefit or other effects of drugs comes from this. Other evaluative techniques, and especially randomized controlled trials (RCTs), are less accurate and less objective” (2020, p.1). Yet that view clashes with the position of evidence-based medicine, aptly summarized by Howick, who claims that clinicians are routinely led astray when assessing the effectiveness Causal inference from clinical experience of interventions based on clinical experience “due to the natural course of illness and the placebo effect” (2011 p. 164). The aim of this paper is to offer some insight on this polarized debate, and to suggest, albeit in preliminary terms, a path through the debate which respects the concerns of both sides, while ultimately offering a partial but not full vindication of the evidence-based medicine view regarding causal inference from clinical experience. Assessing the reliability of causal inference based on clinical experience by empirical evidence is of limited value, because precisely what is in question is the reliability of one mode of evidence (first-person experience) compared to others (randomised trials, for example). Moreover, empirical evidence about clinical experience is limited to the range of events that one can observe in actual clinical practice, which would be limited in scope for various contingent reasons about the practice one observed. In this paper, instead, we develop a formal model of causal inference from clinical experience. We then use a computer simulation to generate data based on this model, with the (...truncated)