(8)流行病学导论-Introduction to Epidemiology-公开课-关爱惟士
(8)流行病学导论-Introduction to Epidemiology


Professor Kaarin Anstey

Kaarin Anstey教授

In epidemiology we look at a particular health problem in terms of the population. Whereas clinicians are just focusing on an individual patient, in epidemiology we want to look at the prevalence, so how common is a particular health condition across the entire population, and at what rate does it increase. So we talk about incidence. So if we’re looking at dementia, we’d say, “how many people have dementia at this point in time?” - that’s the “prevalence.” And “over the next five years, how many people would develop dementia?” - that’s the “incidence.” And then another critical aspect of epidemiology for this is looking at “risk factors,” so things that increase the chance that a person will develop the disease that we’re studying.

在流行病学的研究中,我们是从人群水平来了解某一特定的健康问题。临床医生只是专注于个体患者,而在流行病学中,我们是要观察患病率,即在整个人群中特定的健康问题有多严重。另外还要观察它以什么样的速度在增加,所以我们谈论发病率。因此当我们研究认知症的流行病学时,我们会说,“在这个特定时间点有多少认知症患者?” - 这是指“患病率”。而“在接下来的五年中,会有多少人发展成认知症? - 这是指“发病率”。此外流行病学的另一个关键点是观察“风险因素”,就是增加一个人发展我们正在研究的这种疾病的概率。

I’m going to be talking about the types of epidemiology studies that are conducted, to answer questions of prevalence, incidence and risk. So, to get the prevalence of a disease, say dementia, we need to have a sample that we study that is representative of the population. Ideally, for example, in Australia we would take a random sample of the entire population of Australia, but we’ve got such a large land mass here that’s not possible. So we often do a representative sample of a particular region, and then we extrapolate up to the country population from that. So that’s a single, one-off survey that we could do or study. Often we want to do a longitudinal study, especially if we’re looking at risk factors. So we take a cohort and we do a longitudinal study or what’s also called a prospective study, or it’s sometimes called a cohort study. And that’s where we take this group of people who’ve been sampled randomly from a defined area or population, and we follow them over time, and we test the same people again and again and again. And we look at how their risk factors may change and whether or not they develop dementia longitudinally.


How useful is an observational study or an epidemiology study in the field of dementia epidemiology in establishing a risk and causality, compared to a randomised control trial? To answer that question, we actually have to take a step back. And we have to think about the fact that dementia occurs due to pathological changes in the brain, and these occur over decades. So they’re not occurring over a very short period of time like 12 months or two years. And, so, we can’t actually conduct short-term studies on the causes of dementia at the population level; we need to have that long-term information to see who’s going to go and develop dementia. From that point of view, we’re really left with cohort studies as the main method of studying risk for dementia.


And, secondly, a number of the risk factors that have come out from the research that’s been conducted are things that we couldn’t examine using a randomised control trial methodology. I’d just better explain what a randomised control trial is. A randomised control trial is when you take a sample and you randomly allocate the members of the study to different conditions, and then because of the randomisation, you’re able to adjust for all of the potential factors that may influence the results. So, for example, you might conduct a randomised control trial of a drug that theoretically is thought to prevent dementia, and every person who came in would get a random allocation to drug, or no drug. The problem is, with dementia, again, it takes so long to develop and the brain changes take a long while to accumulate, so we couldn’t conduct a short-term randomised control trial. And, secondly, it wouldn’t be ethical to look at some of the risk factors for dementia in a randomised control trial. For example, we couldn’t ask people to smoke to see if smoking caused dementia. We couldn’t expose people to heavy air pollution, to see if that causes brain damage that’s irreversible. The sort of questions that you’d end up asking are just almost illogical and they’re completely unethical. So we really can only look at these questions using what we call observational studies, where we look at exposure just through normal life, whether the people chose to smoke, whether they lived in an area with heavy air pollution, and then we use statistical methods to try to adjust for all of those potentially confounding factors. And then we follow people up and see if those what we call “exposures,” so exposure to smoking or air pollution or heart disease of whatever, if those things increased the risk long-term of dementia.


So, the question is: how do we evaluate the results of observational studies that show, for example, factor A is a risk factor, and then a similar study in another country might find it’s not a risk factor? There’s a number of approaches to this problem. This is something that we deal with – well, I in particular in my group, at the Australian National University, deal with a lot. First of all, we look at the quality of the research design of the study that found the result, did they adequately adjust for potential confounders? Was the sample biased? How long did they follow up the sample; is it a long enough follow up? Was there a lot of sample attrition leading to sample bias? Were the measures adequate? Did they properly measure the exposure and did they have a proper measure of dementia diagnosis at outcome? So you look at all of these design issues. Was it a big enough sample to give a statistically robust result?


So that’s the first approach, and sometimes that alone will tell you that the result is probably not reliable, because there are methodological flaws in the study or limitations that mean it’s inconclusive. Secondly, what we do in this field of dementia epidemiology is that we consider each cohort study as one study in a sample of studies, and there’s actually a population of these studies, so we assume there is a true finding of an effect. So we do something called “meta-analysis,” and that’s where we bring together all of the different studies on a particular topic. So if we took, for example, smoking, does smoking increase risk of dementia, we would get all of the published studies on smoking and risk of dementia and we’d use statistical methods to pool the results. And that gives us a robust estimate and we can actually look at something called the study bias, the selection or publication bias using statistical techniques to see if we have got a good representation of all of those studies. And, from that, we derive a much more robust estimate of the effect and a standard error around that estimate, and that’s really what we prefer to use, rather than just the result of a single study.