1. Introduction: The Use of Cases in Medicine

Despite having a somewhat dubious reputation as a form of anecdotal evidence, case reports remain exceedingly popular forms of communication and publication in medicine. A survey in the early 2000s documented publication of approximately 40,000 new reports per year and noted that case reports represented 13.5% of all publications in core medical journals (Rosselli and Otero Reference Rosselli and Otero2002). There is no evidence that publication of cases has subsequently decreased, particularly given that a number of peer-reviewed, open-access journals dedicated to cases began publication over the past 10 years, allowing even more rapid and easy exchange of information via cases (e.g., see Kidd and Hubbard Reference Kidd and Hubbard2007). Hence, although the preferred publication venues may have changed for this genre of medical publication, it is indisputable that cases remain an important form of research literature in this field.Footnote ¹

So what is a ‘case’? In this paper I utilize the term in a generic sense to include any case reports published individually or as a case series. Cases typically provide narratives of how a particular patient’s course of illness, diagnosis, and treatment occurred. However, they are not ‘about’ that particular patient as such. Instead, they are aimed at future potential patients and, in particular, attempt to expedite identification of subsequent cases; suggest a particular treatment as efficacious (even if only clinically demonstrated but not yet fully understood), or document prognosis; outline factors that could lead to prevention in the wider at-risk population; and allow refinement of diagnostic conditions. What scientific work can cases do? As argued elsewhere (Ankeny Reference Ankeny, Howlett and Morgan2011), cases serve as ‘vehicles’ for gathering facts and putting them into contact with each other. Think of a train caravan composed of a series of different types of cars joined together, but where it is not yet clear which cars actually will make the trip, in which order, or what in fact is the ‘engine’. In addition, as will be argued in this paper, the process of sorting through the details contained within a case helps clinicians and researchers refine hypotheses as to the underlying cause(s) of a particular condition.

Even if we grant that cases have these clear general purposes at their core, cases nonetheless appear to be limited in their ability to provide evidence about causal relations for several key reasons: First, their reliance on individual instances or very delimited sets of patients (sometimes mistakenly described by skeptics as ‘unique’) seems to result in limited sources of evidence as it has been conventionally construed, particularly in our era, which has seen medicine become dominated by randomized clinical trials (RCTs) as the best source of evidence according to the tenets of evidence-based medicine (EBM).Footnote ² Cases are observational, not experimental, and the circumstances under which observations and documentation of the condition of interest occur are often (wildly) uncontrolled and nonideal, given that they are generated within the context of medicine as practiced in a clinical setting. Second, cases tend to provide extremely detailed accounts of particular patients, given that one of their primary roles is as an early communication device for ill-understood phenomena, and thus often seem to obscure underlying causal patterns, reflecting medicine’s role as what has been termed ‘a science of particulars’ (Gorovitz and MacIntyre Reference Gorovitz and MacIntyre1976). In particular, the relatively amethodological and highly narrative structure of cases results in a lack of filtering of those attributes most likely to be causally relevant for explaining the phenomena observed (on the routinized narrative form present in medical cases including some historical reflections, see Hurwitz Reference Hurwitz2006).

This paper approaches these issues by exploring the potential causal power of cases particularly given their prominence in published medical literature: rather than developing an abstract analysis of their limitations, it focuses on a recent series of cases in the medical literature in order to explore their role in causal attribution in medicine. It is argued that the way in which these cases are synthesized to generate causal attributions and testable predictions supports Caroline Whitbeck’s (Reference Whitbeck1977) version of a manipulability view of causation.Footnote ³ As such, cases often are essential first steps within contemporary research practices that provide working candidates for further tests of causal relations using more conventional methodologies such as RCTs or other experimental methodologies, or retrospective case controls. This paper illustrates the reasoning processes utilized when case reports are synthesized and provides further support for Whitbeck’s claims inasmuch as it shows that manipulability is a key model of causal relations at use in these biomedical practices focused on diagnosis.

A brief word about methods: why focus solely on published accounts of cases, if my main interest is in clinicians’ and researchers’ actual reasoning processes?Footnote ⁴ Both pragmatic and principled rationales underlie this strategy: clearly the direct observation of the generation of novel case reports in real time would be difficult, as it would be impossible to know in advance where an ‘interesting’ case will occur, or whether a particular case is likely to gain traction in the field and result in follow-up and subsequent interrelated cases. Among the tens of thousands of cases published each year, many lurk in the literature awaiting future uptake for years or even decades. More importantly, the use of published case reports explicitly underscores a key point in this paper, namely, that this form of publication is not primarily a clinical phenomenon: by the time a patient ‘becomes’ a case, typically she or he is no longer a patient, having generally died or recovered. Published case reports are an important but neglected part of biomedical research practice that warrants critical exploration.

2. Mysteries of the Heart

Before presenting the details of the set of cases that will serve as the focus for this paper, it is important to clarify some issues associated with the methods utilized here for case choice. These cases have not been chosen because of any documented claim about their typicality; in fact, as noted previously, many cases never accrete over time and simply lie dormant in the published literature as n’s of one, unlike this set of cases. However, the pattern that is typical when cases come to have importance in medical practice and research can be seen in this set of cases, namely, that a number of cases are articulated over time with explicit reference to one another, drawing on some of the assumptions and conclusions in the previously articulated cases. These cases also represent a specific subtype, namely, cases where a condition is fairly delimited and comes to a fairly clear resolution (typically either in death or in recovery) and where the authors propose hypotheses about associated causal factors as part of the process of generating a diagnosis.Footnote ⁵ Hence, this subtype of cases is an excellent starting point for an examination of the potential power of cases to allow causal attributions and ultimately to help reveal causal relations.

A case report in the Annals of Emergency Medicine (Forman, Aizer, and Young Reference Forman, Aizer and Young1997) described a young woman with anorexia nervosa (including binge eating/purging) who was seen in an emergency room in the northeast United States following the ingestion of a large amount of caffeine (20 g) during a suicide attempt, and who was readmitted 18 hours later to hospital after suffering a myocardial infarction. Although it was well recognized that caffeine in higher doses can produce toxic effects, including arrhythmias and hyperglycemia, and caffeine toxicity had previously been reported in several other cases observed in emergency rooms in the context of suicide attempts involving caffeinated diet pills (e.g., Zimmerman et al. Reference Zimmerman, Puliam, Schwengels and MacDonald1985), symptoms of the type observed or severity in this case had not previously been reported in someone without other specific coronary risk factors.Footnote ⁶ The authors propose that the high dose of caffeine together with low estrogen levels (related to the eating disorder) acted in a synergistic manner to constrict the patient’s coronary arteries, ultimately resulting in the infarction, and they note both the need for clinicians to be aware of the possibility of myocardial infarction from caffeine overdose and the wide-ranging health consequences of eating disorders. Although not extensively cited in the years immediately following publication, this case gained attention in subsequent years both in publications of other case reports involving unusual cases of heart dysfunction with some relation to high levels of caffeine intake and in reviews on the effects of caffeine (e.g., Pelchovitz and Goldberger Reference Pelchovitz and Goldberger2011), particularly once the link between highly caffeinated energy drinks and these symptoms was observed, as discussed below.

A subsequent case report (Cannon, Cooke, and McCarthy Reference Cannon, Cooke and McCarthy2001) described a 25-year-old woman with preexisting mitral valve prolapse who developed intractable ventricular fibrillation after consuming a “natural energy” guarana health drink (containing approximately 10 g caffeine/L, equivalent to more than 60 times the concentration of caffeine in cola beverages) and died in the emergency room despite numerous resuscitation attempts. Her underlying health condition is presented by the authors of the case as a key part of the (causal) story, particularly as she had been cautioned by her doctor to avoid caffeine, given the risks presented by the prolapse. However, the cause of death in this case is clearly taken by them (and defended by the coroner in the postmortem investigation) to be the caffeinated drink: “malignant arrhythmias may occur in the absence of cardiac stimulants, but the association of arrhythmia with drugs that increase cardiac irritability is well known. . . . In this patient’s case the Coroner found that the high level of caffeine was associated with the development of an intractable arrhythmia.” The researchers also conclude that “this case highlights the need for adequate labelling and regulation of such products.” This case became relatively highly cited soon after its publication, particularly in the context of numerous separate case reports describing fatal intoxications with caffeine, and resulted in a recall of the particular product associated with this incident in Australia, where the death occurred.

In the late 2000s, a case report appeared (Berger and Alford Reference Berger and Alford2009) describing a 28-year-old man who collapsed after motocross racing in Queensland, Australia, having consumed excessive amounts of energy drinks (seven to eight cans, equaling a total of approximately 640 mg of caffeine) over a 7-hour period. On the surface, this case appears to be more causally ‘clean’ than the previous ones described, given the absence of other confounding conditions such as underlying health conditions. But note that this clarity is in part artificial, a result of the building up of these cases over time as a body of knowledge: in principle, the condition could have been due to other case-associated factors such as excessive exercise or exertion, which are also known to be correlated with heart dysfunction in some circumstances. The researchers describe a complex causal story: since the drinks in question also included taurine, for which toxicity information is somewhat limited, the researchers note that the heart abnormalities observed may have been related to “a combination of excessive ingestion of caffeine- and taurine-containing energy drinks and strenuous physical activity” (43). But the researchers’ final conclusion is that caffeine (delivered via caffeinated beverages) is the key cause on which future research, monitoring, and even policy should focus: “We accept that this is a single case, which does not and cannot establish causality. However, in the context of concerns reported in the media in recent years relating to similar events overseas, and in the presence of a plausible pharmacological mechanism, we believe that the potential dangers of these caffeinated energy drinks should be highlighted, and monitoring for future adverse events should be conducted” (43).

What can be established based on this short summary of this interrelated set of cases? First, on the clinical side, it subsequently has become nearly received wisdom that high doses of caffeine consumed by seemingly otherwise healthy individuals can result in heart abnormalities and that various types of highly caffeinated substances pose significant public health risks, particularly among young people who may unknowingly ingest them in large quantities (see, e.g., Wilson et al. Reference Wilson, Kado, Samson and Miller2012; for a review, see Gunja and Brown Reference Gunja and Brown2012). Second, from a philosophical point of view, this case series reveals a typical pattern of evidence gathering by accretion: in the recent medical literature, the cases summarized above (and others that are closely related, sometimes together with available data generated using nonhuman experimental organisms) are often cited as key sources of evidence supporting the claim that high doses of caffeine can cause heart abnormalities. It is this second use of cases—as a means of generating hypotheses regarding plausible causal associations for particular medical conditions—that warrants more exploration.

3. Whitbeck’s Theory of Causation in Medicine

In a classic but now relatively neglected paper, Caroline Whitbeck (Reference Whitbeck1977) explores what she terms the ‘disease entity’ model of causation in medicine. She begins by reminding us about the goals of medicine and the corresponding types of inferences that are of most interest in cases of disease, namely, prevention, effective treatment, and prediction.Footnote ⁷ In connection to these goals, she argues in favor of a manipulability view, namely, that causal relations are best analyzed in terms of the production or prevention of some state of affairs by manipulation of another (see also Collingwood Reference Collingwood1940; Gasking Reference Gasking1955; and for a more technical examination, see Woodward Reference Woodward2003).Footnote ⁸ Thus, in medicine, since practitioners typically are interested in curing diseases where possible, they focus on the (causally related) sufficient conditions for cure and “speak of it as ‘the’ factor which cured the disease” (Whitbeck Reference Whitbeck1977, 628). This attribution is similar where prevention is the main goal: practitioners focus on the (causally related) necessary conditions, avoidance of which would have prevented the development of the disease condition.

But what then guides etiological attributions, including the selection of a particular entity as ‘the’ cause? Whitbeck notes that practitioners appear to follow a series of rules of thumb: first, they have a decided preference for proximate as compared to remote causes, and second, they tend to prefer those factors that preexist in the environment prior to contact with the patient’s body, which then ‘acts upon it’. The first preference clearly reflects the historical priority of a particular type of disease entity, such as an infectious agent, which has dominated medical thinking and practice certainly since the advent of the germ theory but has a long history even prior to it. She notes that the second preference is less intuitive than the first and ties back to the instrumental nature of medicine: it causes us to focus on environmental factors that can be recognized and eliminated before they can cause disease in the body (or the elimination of which may cause a disease condition to abate). Most importantly, Whitbeck argues that as theories of medicine shift, what clinicians or researchers identify as the causal agent also changes, as well as even whether considering one thing as the agent is the right strategy.Footnote ⁹

What are the implications of the manipulability view of causation in this context? First, and perhaps most importantly, there is no need to eliminate all of the necessary conditions in order to prevent something: practitioners need only choose one. Obviously, the key question relates to which one they select: often this is the most accessible, obvious, unusual, or similar. The choice is often quite random, in the sense of not necessarily being logical or grounded in solid scientific evidence that would allow them to weigh the relative importance of these conditions. Hence, practitioners’ interests in clinical medicine are decidedly instrumental in two ways: they focus on what needs to be prevented or removed in order to prevent disease (and similarly for cure), and they also are focused according to the interests of the speaker and what she or he wishes to control (or thinks is controllable in practice).

4. Cases and Causation

In this section I return to the set of cases described in section 2, in order both to support a manipulability view of causation underlying the processes of causal attribution and to return to the question of what type of epistemic work cases in fact do for those who practice medicine or do medical research.

Instead of manipulability, why not simply focus on necessary and sufficient conditions when providing an account of how causal relations are attributed based on medical cases? Or perhaps we instead should favor some form of a causal network model to describe the processes of causal tracing that appear to occur when practitioners articulate and accrete cases. The latter suggestion is especially tempting, particularly given the way in which individual cases appear to reveal causes in an iterative manner and as part of a causal narrative. This process is compatible with the argument that in the process of working through the details of a disease condition, particularly when talking through the disease process with a patient as part of the clinical encounter, practitioners rely on a causal network instantiation model (Thagard Reference Thagard1998) to produce a narrative-style explanation of why someone got ill. In each of the individual published cases described above, we clearly see development of a detailed narrative that outlines various putative causes and systematically excludes certain ones as irrelevant based on additional information.

So, for instance, in the first case detailed above (Forman et al. Reference Forman, Aizer and Young1997), details are provided about other aspects of the case that clearly are designed to rule out other potential causes, such as personal and family histories and use of substances other than caffeine (such as illicit drugs), none of which contain any apparent ‘red flags’ that might have been associated with the unusual cardiac symptoms. As more information is gathered within a particular case, certain hypotheses are narratively ‘tested’ with regard to what appear to be necessary/sufficient causal conditions. Such analyses are essential first steps that provide working hypotheses about causal attribution that can ground further tests of causal relations. Clearly, the processes of ruling in and out various possible causal factors also reflect aspects of practice relating to what misleadingly is sometimes termed the ‘art’ of the case: experienced practitioners engage in a detailed, systematic sorting and weighing process of various factors in ways that are highly specific, require particular skills, and derive from both medical education and training, especially in differential diagnosis (Hunter Reference Hunter1991), as well as the long history associated with the particular format, language, and contents in which cases appear in the published literature.Footnote ¹⁰

Similar processes occur when a set of cases is articulated and brought together over time. Often a new case is put in dialogue with similar cases that have preceded it, with focus in particular on the putative cause. However, sometimes new cases ‘overlook’ relevant past cases where, for instance, their authors might not see the previous cases as closely related because of surface dissimilarities, particularly in relation to the articulation of causes (or may simply be unaware of the previous cases).Footnote ¹¹ For example, the fatal case involving a woman with a previously recognized heart condition (Cannon et al. Reference Cannon, Cooke and McCarthy2001) does not make reference to the first case (Forman et al. Reference Forman, Aizer and Young1997) of the woman with an eating disorder. The subsequent literature connects these to each other once the ‘red herrings’ of the preexisting conditions come to be de-emphasized in the case descriptions in favor of arguments for the causal significance of high doses of caffeine even in healthy individuals.

Thus, when cases are brought together and compared, they often generate testable causal predictions, as similarities and differences are articulated, and information added, until the cause can be identified. In the cases detailed above, the ability to directly test the hypothesis that high doses of caffeine can cause heart abnormalities is limited in terms of RCTs or similar trials, but it can be addressed through a range of other experimental techniques, such as animal models, as well as via case controls and other retrospective and observational methodologies. Hence, the attribution of a putative cause in these clinical settings via cases then creates a basis for medical research to engage in more ‘conventional’ or traditional methodologies.

However, both with the articulation of causes within individual cases and through the articulation of a set of interrelated cases over time, manipulability remains key: caffeine is selected as ‘the’ cause not because it is in fact the only biologic or in principle cause, but because it is something that can clearly be identified as an external entity that caused a change in the patient (the most obvious form of manipulability, according to Whitbeck). Caffeine also is the cause because in the clinical circumstances in question it is the ‘easiest’ thing to eliminate: although other types of clinicians might focus on treating the underlying conditions (such as the eating disorder or mitral valve prolapse), those working in an emergency room setting (and producing these cases) focus on the factor they view as being eliminable (if not in the current patient, then in the broader population via public health interventions, policies, and education).

These cases hence underscore the instrumental nature associated with the identification of causes: alterations in the extent of knowledge within any particular case change what is identified as the cause, but point of view matters. Compare how a psychiatrist might identify the ‘cause’ of these heart abnormalities in the first case with the point of view of the emergency room physicians who wrote up the case. Additional factors in the determination of causes within cases include dominant understandings of disease at any particular moment (e.g., whether caffeine can be harmful if ingested in high enough doses, which was debated even in recent times), available technologies to measure various vital signs, and various normative claims about what counts as typical or abnormal.Footnote ¹² Therefore, what is key in the determination of causes is what can be ‘manipulated’, not as judged in principle but in practice.

5. Conclusions and Implications

But so what? A skeptic will likely say that cases are a strange remnant of an old-fashioned way of doing medicine, or an easy way for those not involved in rigorous research to get published. She also will contend that medicine is weird, or perhaps more precisely nonscientific, especially with regard to the interface between the clinic and the laboratory. Thus, a more accurate and widely applicable account of how causes are established in medical science would need to focus on research using more traditional scientific methodologies such as RCTs and not bother with cases, which anyhow get overridden and ignored once ‘real’ evidence is available.

As this paper has sketched, review of the published medical case literature reveals that these claims cannot be sustained. At a minimum, cases provide a way forward for clinical practitioners who need working causal attributions with some manipulable facts in order to treat subsequent patients who might present with similar patterns of symptoms until better evidence can be produced (if ever), but they also can generate clear evidence about causal hypotheses worth testing by researchers. These causes are specific and instrumental in nature, in large part because much of medical research seeks to support those working in the clinical setting with actual human patients.Footnote ¹³

Furthermore, medicine is far from unique in its focus on pragmatics and applications: upon reflection, many other scientific fields may evidence similar patterns of reasoning and evidence generation when we shift our focus away from theories toward scientific practice. Medical cases often provide a form of substituted evidence in areas where RCTs are not ethically or practically possible; so too might case-based reasoning be in use within fields where more conventional forms of evidence generation are not possible. Examples include situations where the phenomenon of interest is rare (or even ‘unique’) and not easily replicable (say, in the case of accidents such as airplane crashes or structure failure),Footnote ¹⁴ or the early stages of scientific practice before more detailed theories can emerge and be tested. Medical cases also can serve as early-warning systems, for instance, in cases of newly emerging diseases or public health threats, and similarly other sciences interested in various forms of adverse or unusual events (such as toxicology or meteorology) may well utilize variants on this form of reasoning. A bit more far afield, cases in some sense evidence many practical and epistemic qualities associated with the open science movement: inasmuch as they are critical devices for furthering scientific knowledge that is useful in practice in real time, they have some advantages over more traditional forms of scientific publishing (and admittedly some clear disadvantages).

In summary, although medical cases might appear to be detective novels lacking an author who can provide us with the identity of the perpetrator, they can, as argued above, provide evidence for working causal attributions that can ground future clinical treatment, public health policy, and research, particularly when they explicitly detail the reasoning behind their conclusions or causal proposals. Hence, the case can simultaneously serve as a narrative and an epistemological device that can assist practitioners in identifying causes.