Dissertation Ideas

From BrothersBrothers


To Obtain

  1. The Gold Standard
  2. Article: Evidence-based medicine, clinical uncertainty, and learning to doctor. Timmermans
  3. The Body Multiple
  4. The Ontogeny of Information: Developmental Systems and Evolution - This is a reference from Dupre's book, talking about the role of DNA as "the master molecule", viewed as the "unmoved mover."

To Read


Conversation with Ellen:

  1. Take as the overall theme: Care of patients in particular provider-patient relationship (i.e. what we do and how we decide what to do). What does decision-making look like in this setting, with integration of personal knowledge and experience with patients and generalized data? How do introduction of technologies, including WGS and the informatics tools that are used to deliver those results, influence the way we navigate those influences?
  2. This theme is then the basis for addressing the more specific problem of dealing with genetic risk estimates, especially with respect to (1) determining when to tell them to patients and (2) determining which interventions to perform in response to them
  3. Since there is no synthesized model for medical decision-making, and EBM is clearly problematic here (due to no evidence, and due to conceptual problems with applying it in such a setting), a helpful technique will be to analogize from other practices in medicine, perhaps framing these as heuristics for dealing with uncertainty.


  1. I could do a chapter (or maybe a separate article) on genetic risk, taking my own risk for MI, and explore using different tools (the Mayo risk estimate tool, genetic test, family history alone, cholesterol markers alone, etc) to demonstrate that risk estimates vary, and that the formalization of risk in a specific number over-represents the certainty of such an event happening.
  2. Kuhn's definition of Disciplinary Matrix provides a way to frame what medicine is as a conceptual community as opposed to a practice (exemplars, values, etc), also Polanyi's "tacit knowledge"
  3. Clearly, practical reasoning (Aquinas and MacIntyre) is very important here, as well.


  1. Misunderstandings about risk derive from more fundamentals blindspots about role of mathematical models and values in science
  2. Scientific theories, framed as mathematical models, are what Berkeley call "mathematical hypotheses." The world does not exist as numbers of mathematical equations; mathematical models are derived from observation and gain support as they successfully represent, and sometimes predict, the way a particular natural event occurs.
  3. As Kuhn observes, paradigm shifts in science occur when paradigms, often based on specific mathematical models, are viewed as failing to respond to problems. This has traditionally assumed to mean that the model fails to predict or represent certain features of the field of study (such as the Ptolemaic models' increasing complexity as it required modifications to fit observed patterns in the solar system).
  4. There are a number of reasons that such a model may fail, however. A model might fail because it predicts a outcomes based on a certain set of variables. When a new variable comes to be recognized as having an effect on such systems, mathematical models that don't account for that variable could be viewed as failing. Possible examples include something like cervical cancer, when HPV came to be appreciated as the significant cause or stomach cancer + H. Pylori. Obesity would be a great example of the way that various models repeatedly fail as factors thought to be important fail to carry explanatory power: first endocrine, then psychological, then much later genetics.
  5. Ice cube melting is an example of a process that can be modeled quite successfully. This is because the important variables are assumedly all now known, and ice cubes can be represented quite effectively in models, since their characteristics are quite homogenous. (this is from Gorowitz and MacIntyre)
  6. However, more complex phenomenon like hurricanes cannot be modeled with complete accuracy. This is not, however, because the important variables are not known. On the contrary, the appreciation of physics related to air flow over warm water is quite well understood. In this way, the general features of hurricanes can be modeled effectively. But the reason that specific hurricanes cannot be modeled successfully is that inputs into such models in the real world are large in number, complex, and difficult to collect through available technologies. So the paths of hurricanes are represented as probability paths.
  7. Human risk for disease is similar. Many of the causes of many disease are well appreciated. The cause of CF has been known since ____. However, the course of health and wellness are extraordinarily complex such that in many cases the variables that are important in the development of a range of disease are not all known, even though some important inputs may be known. In addition, the histories of humans are even more complex than that of earthquakes, and infinitely more so than ice cubes (again, these examples are those of Gorowitz and MacIntyre). The ability to predict the onset of disease in an individual, then, varies on the basis of the comprehensive of available models in terms of the number of variables it can take account of, and also the inputs that are present in the history and ongoing life story of patients.
  8. Risk estimates for individuals emerge within this context of incomplete knowledge of biological systems and incomplete knowledge of particulars. This fact is demonstrated by the availability of co-existing models to predict disease. For example, MI risk estimate tools. The online Mayo Tool says one thing, 23inMe says another, etc. Each of these models is incomplete in the number of variables it is able to account for, and even the inputs it takes into account are snapshots, they can't account for the history of individuals, such as changes in cholesterol over time.
  9. A significant problem with risk estimates, then, is that they represent predictions for the future in mathematical terms in a way that explicitly communicates a probability (Your risk for having an MI is 20%), but lacks the ability to also communicate the success of the model, or, in other words, the probability of the probability. Disease risk predictions, like weather predictions, become confused with other scientific mathematical models. Even these models are "mathematical hypotheses," albeit ones that have succeeded in representing the pertinent systems accurately through a long history of experiments. But because that historical success comes to support a conflation between the natural reality and the mathematical modeling of that reality, a type of misunderstanding about the nature of scientific models becomes imported into disease risk models.


  1. Dupre's book provides an account of the different frames for thinking about genetics/genomics. There seems to be a potential harm in implementing genomic understandings into a world that still treats of DNA in the frame of classical genetics.


  1. Is evidence-based medicine a form of theoria or a form of phronesis? It appears to be an attempt for theoria to bridge the gap from the theoretical to the practical, using a skeletal account of phronesis that involves only deciding which evidence is most pertinent to a particular patient.
  2. While the problems with this account vis a vis phronesis are obvious, it's problems vis a vis theoria are even more important. EBM requires professional judgement on the part of the provider - Which evidence applies here? What are the implications of studies that don't quite fit? This is closely linked with another concern - that EBM does not generate desirable outcomes in all cases. Evidence generated in the ways currently used are based on relatively heterogeneous groups. The failure of EBM is based on two elements: the concern that available evidence isn't applicable, and the concern that available evidence is for heterogenous groups, and thus imperfectly predicts outcomes.
  3. Personalized Medicine has emerged from a deep anxiety about this shortcoming of EBM. Personalized Medicine is based on the commitment that (1) the application of evidence should be relatively straightforward (minimize professional judgement and, thus, contingency) and (2) the evidence applied should be for homogenous groups into which the patient fits perfectly. This is why the term "stratified" is sometimes used.
  4. The problem with Personalized Medicine, though, is that attempts to "fix" EBM but brings no new conceptual innovation to the situation. Personalized Medicine is, in fact, EBM repackaged. It could even be said that Personalized Medicine is a more fundamentalist version of EBM.
  5. One question, then, is whether PM effectively addresses the practical (i.e. translational) goals of medicine based on theoria, or are practical reasoning approaches more appropriate. Put more directly, can PM do something EBM can't, or does PM, like EBM, require practical reasoning to succeed?


  1. Personalized Medicine is a fundamentalist form of EBM.
  2. The anxiety about judgement in medicine inherent in PM (and to a lesser degree in EBM) drives a marginalization of judgement into the "art of medicine" - a marginalization that renders judgement as something for which an account cannot be given.
  3. But the goal of PM and EBM is to render all elements of medicine "measurable," in the strain of "metrics"
  4. However, a key part of judgement is "know how." We must know how to use evidence, physiology, observations of patients, patient preferences and self-reported history, patient perceptions of their own bodies and the causative role of different factors, and economic factors to decide how to act.
  5. A key element of this "know how" - how we learn and how we refine it - is feedback from the world. Our patients, colleagues, and spouses tell us in their words and actions when we have made mistakes - when our judgement went wrong. They also, of course, show us when we did well.
  6. However, the account of PM and EBM (but not, interestingly, physiology) treat this account of experience with individual patients as mere anecdote. Not only can we not learn from anecdote, but we emphatically should not learn from anecdote. The "metrics" are what matter. This account, therefore, not only robs us of our ability to provide an account of judgement, but drains us of our ability to improve our judgement.
  7. Joe here provides an Stout's account of know-how: learning to convert the rules of soccer into the practice of being a soccer ref. The application of the rules is at first mediated (I have to think about the rules and how they apply), but eventually they become automatic, and I know when I see an infraction. However, whether this type of know-how is mediated or immediate, it is still susceptible to feedback, from the coach who objects. When that happens, the ref needs to re-examine the judgement. This is from Chapter 12 of Tradition and Democracy.


  1. Reading about Carl Rogers in "A History of Pastoral Care"
  2. This text draws attention to the liberal underpinnings of Carl Rogers' counseling method - that it was a reaction against legalism, and that one of the critiques of this from within pastoral counseling circles was that communities of meaning provided a context for doing more than just moving forward a person's self-realization. ...the realization of selfhood occurred only within a community of language and commitment, which would provide guidance and direction" (306).
  3. Some of the stances on Personalized Medicine follow in these liberal footsteps - in reaction to open possibilities about ways that people might think about things, there is a reaction against paternalism toward nondirectiveness. Ravitsky and Wilfond come in here - they might derive meaning from it, so we should tell them.
  4. But this denies the possibilities of what a provider engaging in phronesis - through the interaction with the patient the provider should not be nondirective, as if telling someone everything with no care is somehow less directive than telling them nothing. Rather, phronesis allows the provider to determine what, in this setting, is the most helpful thing I can do for this patient - what can I do to help him or her attain human flourishing, which is inclusive of most forms of health. The provider is therefore directive, and the selection of data to reveal to the provider and the decisions about how to use the information is all based on this orientation.


This is from a grant I reviewed: One common approach to increase trustworthiness is to increase transparency (O'Neill 2006); However, information sharing per se does not necessarily increase transparency or trust. Provision of raw data with no interpretation can sometimes obfuscate rather than enlighten (O'Neill 2006; Archon Fung 2007) reducing a patient’s ability to make an informed decision. It can also shift responsibility to the patient explicitly or implicitly which, particularly in an acute care setting, may not be appropriate. Rather than focusing just on sharing informational content, O’Neil suggests that a more interactive model of communication (O'Neill 2009) is necessary to engender well-placed trust. The Royal Society’s recent report on ‘Open Science’ recommends ‘intelligent openness’ where shared information needs to be not only accessible but ‘understandable, assessable and useable’(Royal Society 2012). Relating these criteria to acute medicine might engender a clinically useful dialogue between clinician and patient. Ideally communication should foster an environment where patients are supported and equipped to ask appropriate questions in order to make their own decisions or to trust their doctors to make those decisions. Sharing of medical records is one possible way of achieving this.


  1. Distinction between epistemological and physical probability: We can assume that our risk for developing cancer (for example), based on all relevant effects at play in our bodies (prenatal, environmental, behavioral, dispositional) is a physical reality. That is, there is a "true" probability that we will develop cancer. This is never 100%, since there will still be purely stochastic effects (i.e. the chance mutation of a particular gene at a particular site).
  2. However, in our current state of knowledge, we can't know this "true" possibility. We are more interested in epistemological probability, that is, what can we know about our probability of developing cancer.
  3. The "belief" we may develop about our chances of developing cancer are based on all of the available data. Leibniz, with his background in jurisprudence, first introduced us to the idea that our epistemological probability is an imperfect approximation of our physical probability, but that we are forced, as we are in a court of law, to develop epistemological probabilities from the evidence we have available to us.
  4. In this way, there is a certain logic to obtaining more information about our risk for developing cancer using genetics - we could argue that failing to obtain this evidence is like failing to admit all relevant evidence in a court of law.
  5. The rationale for generating and revealing genetic information is, up to this point, entirely sound. In fact, when we add to the rationale that the information is already "known" in a computer somewhere, and all we need do is report it, then the justification for inputting that information into our formation of an epistemological probability for developing a disease becomes overwhelming.
  6. However sound this rationale may be in the abstract, when we place it into concrete situations, we begin to see how it may not lead to the more accurate "epistemological probability" that we seek. There are a number of reasons for this, and we should take each in turn:
    1. Genetic reductionism causes us to mistake the genetic among a particular population as the true physical probability. That is, it can cause us to ignore the significant contribution from a whole range of other factors in the combined pathophysiology of a condition.
    2. Insofar as we are able to be non-reductive in formulating a model that we believe represents the pathophysiological condition of a single human, we will be faced in many circumstances with combining probabilities from different sources, since as the number of factors we consider increase, the larger the total sample size we must pull from in considering the interaction among multiple factors. That, in order to generate a frequentist estimate of the risk for developing lung cancer in a particular patient, there are only to ways to eschew genetic reductionism:
      1. We can obtain a sample large enough to contain a significant number of individuals who are like our individual in terms of alleles at (for example) 10 genes, personal smoking behaviors, second hand smoking behaviors, family history of lung disease, occupational exposures, and exposure to environmental pollutants. We could then look at the risk of developing lung cancer among the population of study participants who are identical to our patient in all of these respects, and attribute that risk to our patient. The less reductive we are, the larger our sample size must be.
      2. Instead of looking only at patients who are identical to our patient, we could generate a statistical model that combines the various effects of each characteristic (risk of allele A on Gene A vs risk of allele B on Gene A) and extrapolates a risk for our patient who has a relatively unique set of characteristics. This approach not only make the probability we generate more conceptually opaque, it also renders our estimates more speculative. Is the combination of risks alleles in gene A and gene B additive, non-additive, or synergistic? What about gene A, gene B, and gene C? What about among smokers? Insofar as we understand the causative dimension of these effects we will be able to build better models. But it remains the case that our model is forever insufficient in comparison with the "true" physical probability. And we are forced to consider what the marginal benefit will be to our current model from incorporating additional genetic risk information. Notwithstanding the axiom that "more knowledge is good", when it comes to statistical models, this is not the case. In some cases, additional information may add no additional precision to our model.
    3. Failure to recognize distinction between epistemological and physical probabilities
      • Regardless of whether we are reductive in thinking about the physical probability or nonreductive, NEEDS MORE HERE
    4. Effectiveness at altering belief
      • In our earlier discussion about statistical models, I observed that our frequentist estimation of risk is at risk for being "opaque". What does this mean? Leibniz was convinced that epistemological probability is objective. That is, even though we could not generate certainty by taking the available data (**ex datis**) into account, we could generate a conclusion that is "incomparably more probable" than the opposite, and that everyone, using this same data, could come to the same conclusion. Regardless of whether it is possible to generate a frequentist account of empirical probability that is theoretically objective, the field of psychology has demonstrated convincingly that assessments of risk are far more complex than frequentist accounts of the same. That is, the beliefs of actual persons about their risks for developing a disease are developed through a series of experiences, and this process of developing beliefs has its own set of rules and guidelines. For example, psychologist speak of "anchoring." That is, our beliefs develop as we encounter new information or experiences in our lives, but on the event of encountering new information, the new beliefs we develop are nearly always "anchored" to the belief we held before that experience. As a practical example, assume that I believe my risk for breast cancer to be very low, and I encounter a genetic result indicating that my risk might be high compared with the general population. My possible responses to this information are huge, but in general my beliefs will tend to be changed incrementally. That genetic result may be able to alter my perceived risk, but likely not as much as a frequentist account would consider reasonable.
      • Anchoring is just one example of a tremendously complex set of effects that come into play when we speak of how we form beliefs about our expectations of the future, or what we might call our "perceived risk".
      • Could discuss here "future self", change with age, importance attributed to effects (including genetic effects), anxiety, etc.