So calling this a Web of Causation, and there is an article by Nancy Krieger in Social Science and Medicine in 1994, entitled The Web of Causation Has Anybody Seen Spider. So that's what we're looking for. We're looking for exactly what causes what. Because if we know exactly what causes what, then we can prevent the disease or perhaps, treat the disease. And this is the difficulty of studying things that appear to be related, and appear to be associated with raised risk or lower risk, but then we attribute that to actually the cause of the disease. So causality is important because it leads, can lead to prevention if we know it's going on. We have to notice that we have several different metrics to evaluate evidence about cause. So for example, the strength of a relationship. So for a depressive disorder and for smoking, we found in the case control study that the relative odds was 9 or 10. That's a strong relationship. Sometimes the relative odds is just 2 or 1 and one half. And we speak of that as being associated, but we're less sure that it's actually a cause. We have a dose-response relationship. We saw that in the cohort study on the smoking, where the people who smoked more had higher risk of lung cancer than people who smoked less. The more of the cause, the more of the result, that's the dose-response relationship. And that contributes evidence to causality. It's nice that the results are consistent. So, if we do in East London, as George Brown did on stress and depressive disorder, is it also true in Scotland, where he has done work. And is it true in Manhattan and North Carolina and is it true for white folks and black folks and so forth, that's consistency. When it happens repeatedly, that gives us support for the idea that their might be a cause. Temporality is important, we have to be sure that the cause occurs before the disorder. We talked about that in terms of divorce and depressive disorder. Specificity is helpful also. So, if we're thinking stress causes depressive disorder, it would be helpful, or perhaps convincing, if we had a single causal change from stress to depressive disorder. Well, as it turns out stress is related to lots of different disorders. Anxiety disorders and even recurrence of schizophrenia. So, that's a issue of specificity. We can make a theory about a cause a lot more convincing if it one cause has one outcome and, then plausibility. This is often the term used as biological plausibility. That is, can we make a theory which involves a chain of reactions, especially involving the body and medicine, which is plausible. So how would stress affect depressive disorder? Well, it might actually change the hormones. It might change the brain chemicals and we can actually make different theories about how brain chemistry changes in reaction to stress, and that is so-called biologically plausible. The main trick is confounding. That's the main threat to our study of causation, and I'm going to give you the absolutely simplest example of confounding, which you will understand, I think, right away. But here is, and this is from Lillienfeld and Stolley's book. So we have Florida and Alaska. And in Florida there are 131,000 deaths in a given year and 2,000 in Alaska. Florida has a lot more people in it, 12 million people versus 520,000 in Alaska. But still the mortality rate in Florida is 1,000 and it's only 390 in Alaska. So now, do we conclude from this that Florida is raising risk for death and is a causal factor for death? Florida is causing deaths. Do we actually conclude that? You, you already know that's not true, right? So, what we're going to do is, we're going to, I'm going to show you the technique of age adjustment. And, asking you the question, is Florida dangerous to health? We're going to look first at the relationship of age to death in the United States. Then we're going to look at the relationship of age to Florida and Alaska. And then there's a procedure called Age Adjustment, which I'll show you. And we'll see which part of causality this helps with. So, everybody knows that when you get older you're more likely to die, and this is the death rate in the United States in 2003. Very low for people under the age of 45 or 50, and it gets up to be very high at the age of 85. Here's the age distribution in Florida and Alaska. This is the percentage of the population that are under 5, and 5 to 19, and so forth, and Alaska has more young people in it. And then on the far right where I've drawn the, the red ellipse, you can see that Alaska has many fewer old people and Florida is loaded up with old people, they go there to retire. So now we're going to adjust mortality rates for Florida and Alaska. And the way we do this is, we calculate specific rates of death by age. So, we have a rate of death for the under five age group in Florida of 284 per thousand, and you see it's not much different for those in Alaska, 274 per hundred thousand. And, for the five to 19 age group, we see that the rates are very similar. And then we move up for example to the 65 plus and we see that the rate of death for 65 or older folks is 4,425 per hundred thousand, pretty similar to Alaska, 4,350. So now if we take the US population and our standard population in 1988, there were 245 million people in the United States, and 18 million of them were under 5. So we apply those death rates from Florida to our standard population, 18.3, and we project that Florida would have 52,000 deaths and Alaska would have 50,000 deaths under the age of five. So we can do this for each specific age group. Then we just add up the deaths we get, we get 1.9 million deaths in Florida and 1.8 million deaths. This is applying the age specific death rates for Florida and Alaska to the standard population. When we do that, we get an expected death rate of Florida of 812 and an expected death rate in Alaska of 764. So this is age-adjustment. And it's to deal with the confound of age with risk for death. So, we would say that the difference in the death rates between Florida and Alaska is confounded by the age difference. And when we do this adjustment, it eliminates the confound and shows us that it's not the state that is making the death rate be high, it's age in the state. So that technique of adjustment is conceptually similar to a wide variety of techniques of adjustment used in analysis of variance and co variance and linear regression and logistic regression. They are not automatically identical but the concept is the same. We have a potentially confounding variable that we can identify, we adjust for that variable, and then we can better understand whether the causal factor is basically eliminated as a candidate as a cause due to the confounding factor. So, as a conclusion to this study of the web of causation, you can see there are many many different possible confounds. And I'm going to give you just one example that I arrived at after discussion with Carlos Muntaner, my colleague. And so we can consider suicide, which might occur in a truck driver, and we might have the biological explanation for the suicide would be that there is a, a brain hormone called dopamine and it is found depleted in that truck driver, and that is basically the cause of the suicide. But, then we might learn that the truck driver had taken amphetamine and that amphetamine has the biological effect that the rebound is leading to depletion of dopamine. So then we're thinking, well it's not actually depletion of dopamine, it's the rebound from amphetamine. We also think of that as a biological cause, but then we might learn that this truck driver is bereaved. A treasured uncle in his family had recently died and so that might be a confound, and we would call that a psychological possible confound, or cause. And we learn that amphetamine is a drug which is available among truck drivers at truck stops and through dealers because they need to stay awake on the long haul, and amphetamine helps them stay awake. So we might call that a structural level of action for this cause. And we also know that truck drivers work alone, so if they were working in an environment supervised closely, in a factory let's say, or in a store, in a retail setting, the supervisor might notice that they were taking amphetamine and restrict them, or they would be afraid to do it. So we would call that possibly a structural cause also. And we might also note that among truck drivers, everyone takes amphetamine. So we can think the drug is viewed normatively, so it's not stigmatized to take the drug, everybody knows you have to do it if you have real long ride to go. And we might call that a cultural level of action. We also might know that the truck driver was more or less forced or constrained to take this job because his standard living on a prior job was declining, and so he had to take this job as a fallback position. We might think of that as as economic cause of the suicide. But it also might be true that the union had petitioned against long runs. The long truck driver runs, 12 hour, 18 hour runs sometimes required by the company lead to the need for amphetamine and in this constellation of factors might have contributed to the suicide. The union being ineffective is really a political cause and there may be other areas, other countries where unions are more effective, truck drivers don't have to make long runs and they don't have to take the amphetamine and they have a lower rate of suicide. But it's also true that there may be declining resources in the world in general, leading to a declining standard of living and the truck driver feels this decline in resources and so is taking the job in order to retain his standard of living. And so we might call that a world system cause. So you can see the web of causation. Has anyone seen the spider? This is a complicated spider. And this is one act, it's a dichotomous act, a yes or no. This, the person that's committed suicide, they were living, now they are no longer living. And in studying that, and in studying the onset of depressive disorder in general, we often have a, a huge array of causes which we have to sort out. And many cases, there is no obvious solution. But again, just to remind you, our job as a epidemiologist in this particular field of explanatory epidemiology, or analytic epidemiology, is as Jerry Morris said, to suggest clues to ideology. And all of these are potential clues to ideology and maybe the subject of further analyses to uncover the causes of depression and thereby to prevent it in the public's health. So thanks for listening. We'll consider more about these multiple causes of depression in the next lecture when we consider the risk factors for depression, which are available in the current literature, and I look forward to talking with you then.