by Raywat Deonandan, PhD

My Notes
  • Required.
Save Cancel
    Learning Material 2
    • PDF
      Slides 12 ValidityReliability Epidemiology.pdf
    • PDF
      Download Lecture Overview
    Report mistake

    00:00 Now we don't know that smoking causes cancer, at least not until we've explored it more fully. The reason we can't be reliably certain experimentally that smoking causes cancer is because no one's ever applied a randomized controlled trial to the relationship between smoking and cancer. That would look like this, we'd have a study population that we randomize into two groups, one group would have to smoke and the other group would have to not smoke and we'd follow them for many years and we'd see which group develops more cancer. That's not ethically feasible or permissible, so it's never been done. As a result another set of criteria or perspectives need to be applied to that scenario to establish whether or not it's likely that smoking causes cancer. And today we're pretty confident that in fact smoking does cause lung cancer.

    00:48 Now the set of criteria I'm talking about are called the Bradford Hill criteria named after Sir Austin Bradford Hill. There are other criteria as well but his are the most famous and the most widely applied. They're the foundation for causal proof in epidemiology in the last few decades, they've also bled into other disciplines like anthropology and social science. What Bradford Hill did was he set out a series of conditions, wherein a relationship must satisfy them to be considered likely causal. Not all relationships satisfy all of his criteria, but we like to say that if they satisfy most of them or a lot of them or the important ones, that relationship is likely a causal relationship, so let's go through them. The first one is strength and by the way I want to say that even though I'm using these words in descriptions of Bradford Hill's criteria, other people may describe them differently, but we're talking about the same thing. By strength I mean effect size, how big is the relationship statistically. Now we've covered in another lecture relative risk and odd ratios, that's what we're talking about here, how big are the relative risks or odds ratios for the association between our potentially causal factor and the outcome, that's strength. Consistency, do I see the relationship regularly. And specificity, is it likely that this one exposure only causes this outcome, or this one outcome is only viewed in the presence of this exposure. This one's a hard one to observe and it isn't a hard and fast rule, but we'd like to see it if we can. Temporality, this one is critical, so it tells us, does the cause proceed the effect, does the smoking come before the cancer, if the cancer came before the smoking, it's pretty much impossible that the smoking caused the cancer. Biological gradient, this is also called a dose-response relationship. That's when the more of one causes more of the other, meaning, the more I smoke, the more likely it is I'm going to get lung cancer. Plausibility, doesn't make sense, doesn't make sense biologically, does what we understand about the biological sciences and medicine and physiology, does that play into and satisfy the relationship that I'm watching or observing. Coherence is the relationship that I think I'm watching, does it fit into what I know about science as a whole, is it coherent, was what I know about this particular disease or outcome or phenomenon. An experiment is it possible to conduct an experiment. Now if I could conduct an RCT, that would pretty much satisfy whether or not my risk factor is causal for my outcome, but sometimes we can't do it on people, but we can do it animals, so do we have the experimental animal data for example. And analogy, one is difficult to describe, but, can we devise an analogy or analogous relationship between my outcome and perhaps another exposure, if so, then it's possible that my exposure is not unique and possibly unlikely. In other words, have I explored other theoretical possibilities for the relationship that I'm observing.

    04:04 Okay now let's use all of Bradford Hill's criteria to explore the relationship between smoking and lung cancer. First strength, as we've had explored in another lecture, there is a strong relative risk association measured through cohort studies between smoking and lung cancer, a very high relative risk, so strength is satisfied. Consistency, do we see this relationship a lot, we see it all the time, pretty much every study of smoking and lung cancer shows a strong relationship. Specificity, well this is where this particular example sort of falls apart, many things can cause lung cancer and smoking can lead to many different outcomes. So the Bradford Hill criteria don't satisfy that one checkmark, specificity, when it comes to smoking and lung cancer. Temporality, yeah smoking definitely precedes lung cancer, we can definitely show that in a cohort design where we know the exposure came before the outcome. Gradient, absolutely, so the more you smoke the more likely or the more probable you are to have lung cancer. Plausibility, does it make sense? Yeah, we can show in the laboratory setting that lung tissue exposed to the carcinogens of tobacco smoke does tend to become more cancerous. Coherence, is this coherent with other things we've observed? Absolutely, by understanding of this particular branch of science shows that it is coherent. Experimental evidence, yes we've exposed laboratory animals to smoke and they have developed cancer. And lastly analogy, we have explored some other things that perhaps could be causing lung cancer, so this one is a bit more vague and I would say that the Bradford Hill last criterion is not satisfied in this example. But that's okay, the majority have been. And using this framework, we can be reliably certain that there is likely a causal relationship between cigarette smoking and lung cancer.

    About the Lecture

    The lecture Causality by Raywat Deonandan, PhD is from the course Validity and Reliability.

    Included Quiz Questions

    1. Gradient relationship
    2. Temporal relationship
    3. Biological or scientific plausibility
    4. Strength of association
    1. Anecdotal evidence
    2. Data analyses of experiments and observational studies
    3. Clinical observations
    4. Multiple studies with different designs and investigators coming up with same conclusions
    5. Examine the quality of the data and validity of studies
    1. Confounding variable
    2. Coherent variable
    3. Moderating variable
    4. Mediating variable
    5. Systematic variable

    Author of lecture Causality

     Raywat Deonandan, PhD

    Raywat Deonandan, PhD

    Customer reviews

    5,0 of 5 stars
    5 Stars
    4 Stars
    3 Stars
    2 Stars
    1  Star