---start--- epidemiology 1/9/98 [the usual amusing parable is being offered...I am not writing it down] bottom line - freud and jung used two different techniques to look at the same thing, and so will we today. if a disease is more common among animals with certain characteristic than those without the characteristic, we say there is an association between the characteristic and the disease. today we'll try to measure that association - how do we detect and measure an association b/w a characteristic and a disease? if the characteristic is present in population before disease occurs and there is a demonstrable association, it is called a risk factor. we can use three indices to detect and measure associations: 1. incidence ratio: ie/io -> incidence in exposed group divided by incidence in unexposed group. Rarely done because it is so hard to measure true incidence. 2. relative risk: RR: ce/co -> measure the cumulative incidence in the exposed group and divide that by the cumulative incidence in the unexposed group. (cohort study) 3. Odds ratio, OR: (case control study) often a good estimate of the incidence ratio. RR and OR are the most common ways of measuring associations. OR is often a good estimate of the incidence ratio. RR is only a good estimate of incidence ratio if the incidence in the exposed group is small, though. So, RR and OR will only have similar values if incidence in exposed group is small. but all are good indices of association whatever the magnitude of the incidence. The point of all this - students often ask him what it is. Well, if you can find a risk factor and get rid of it you can decrease disease incidence. Disease is multicausal. You can interrupt one cause and reduce the risk of disease. Just because one applies these indices and manages to detect an association b/w a risk factor and a disease, that doesn't mean the risk factor is part of the causal chain leading to disease. You have to be careful about this! There was an undeniable link b/w owning washing machines and having lung cancer in great britain a while ago - but owning a washing machine didn't cause lung cancer. At the time, you had to be very wealthy to own a washing machine, and the people who owned them could also afford lots of cigarettes. So an association could yield simply a marker, not a cause or a risk factor. *** Two ways to look at the association - one is by using the cohort study to look at the relative risk, one is by calculating the odds ratio in a case control study. You will NEED TO KNOW how to do these studies in order to answer some exam questions!!! ****** Prospective Cohort Study: animals without disease--->NON EXPOSED group + exposed group --> monitor both for disease + - exposed A B not exposed C D RR= A/(A+B) ------- C/(C+D) a cohort study begins with a cohort of uninfected animals. this is the first step. it's the order of the steps that distinguishes this from a case series study, really. So cohort studies start with uninfected animals. Then we divide them into two groups, trying to do a number of things - first, figure out your putative risk factor (smoking, coffee, leash wearing, whatever), and assign animals into exposed and unexposed groups according to this single risk factor you are interested in - size of herd, breed of cat, diet, whatever. Then you monitor for however long - in people often over 20 yrs - and you record in each group how many individuals succumbed to the disease of interest. suppose you divided your group into animals from large herds and animals from small herds, and then 95% of the animals in large herds were females and only 5% of them were male. would that be a problem? Yes! that could easily have an effect on disease incidence. So when you make your groups you try to make sure that they are as similar as possible ("matched") with regard to every other factor but the risk factor you are looking at. At the end of your observation period, or when your funding runs out, you calculate the cumulative incidence in each group. In the exposed group, A individuals acquired the disease, and B individuals did not, so the cumulative incidence in that group is A/(A+B). The cumulative incidence in the unexposed group is similarly C/(C+D) and if you divide the former by the latter you get the relative risk. Now, when you make these tables, always use the same format - always but the exposed group on the top and the unexposed group on the bottom, and the positive/diseased animals on the left and the negative/healthy animals on the right. ** this is important ** top row is exposed, diseased is the left hand column. if there is an association b/w the risk factor and the incidence of dz, the RR should be greater than 1. If the RR is significantly greater than 1, we assume we have measured and detected a real association. If the RR is equal to 1, or not significantly different from one, we say there is no association. If RR is less than one, significantly less than 1, we make the assumption that the risk factor in fact reduces the incidence of disease - it is a protective factor. disadvantages of this study - if you have a rare disease, it's very difficult. you need a very large number of animals to start with. you have to follow the animals until disease occurs - if there is a lot of lag or latency as in diseases associated with aging or something, you have to follow the cohorts for a long long time, and therefore you need a lot of funding! if you have a cohort study that lasts even 4 or 5 years, you can guarantee that some animals will disappear from the study for unrelated reasons like owners moving or animals dying unexpectedly. there are some ways to get around the problems. we described a prospective or concurrent cohort study done in real time. you could do a historical cohort study - plunder the hospital records. go back ten years and find animals and follow their medical records over ten years or something. a prospective cohort study of risk factors for feline leukemia: it was suspected that infectious anemia was a risk factor for feline leukemia 594 cats w/o infectious anemia were followed 297 cats with infectious anemia were followed results: feline leukemia + - exposed yes 6 291 no 1 593 cumulative incidence of exposed gruop 6/(6+291) = .0202 of unexp group: 1/(1+593)= .0017 RR .0202/.0017 = 11.9 -> this appears to be much greater than one! it is necessary to examine the 95% confidence interval to be sure that this RR indicates that an association exists. confidence interval is 1.06-131.47 there is always some uncertainty in an experiment - you measure that via the confidence interval. If 95 times out of a hundred times you do the study you get a result somewhere within the interval, then you know that you are probably correct. A concurrent cohort study of risk in neonatal calves with various levels of serum gamma globulin: gg % cohort size deaths cum inc 1.1-6.2 73 12 16.4 6.3-12.0 73 3 4.11 12.1-19.3 73 2 2.74 19.4-46.7 74 1 1.35 a historical cohort study showing risks of being a veterinarian in the US based on cause of death in 5016 white men 1947-77 RR of suicide was 1.7 RR of MVA was 1.44 RR of skin cancer 1.61 and so forth. Case Control Study: 1. select cases --> 3. split into exposed (A) and unexposed (C) 2. select controls --> 3. split into exposed (B) and unexposed (D) + - exp A B unexp C D Odds ratio is AD/BC the first thing you do in a case control study is *find cases*. then, by some set rule, you identify control animals without the disease of interest. Often you select 2x as many controls as you have cases. Then, and only then, do you divide up the animals in the case group and the control group according to their exposure to the factor of interest. You assign the letters as per the notes above, and create your square, and calculate the odds ratio as the proportion of cross products. the case control studies are much more efficient than cohort studies - can use even w/rare diseases, because you start with the disease -you don't start with a whole bunch of healthy animals, and hope enough of them get the disease. then, to make the statistics work, you select controls per some rule. if you are studying colic in horses, you could take every colicked horse that came into NBC, and then to make controls, you could say that you would always take the next horse that came into NBC that was of the same age, breed, gender, and use, and didn't have colic. This is how they matched the cases and the controls. Or you could take the next two horses, so you have extra controls. This is a subtle technique that makes the statistics come out right.*** Read the handout on the second page where it says "remember" and remember that stuff.*** the rest of the stuff there is background reading. the last page of the handout is just for reference also and will not be on the examination. exam may ask why confidence intervals are important, btw.*** so case control studies are efficient - use smaller numbers of animals, use hospital records, often historical. many studies these days report results in terms of OR because these studies are cheaper and more likely to get funded. example of a real case control study of risk factors for UTI in female dogs do diagnostic or therapeutic measures increase the risk of UTI? a) catheterization UTI + - exp 13 65 unexp 1 77 OR = 15.4 (1.96-120) so, it's greater than one with a wide confidence interval, so catheterization does look like a risk factor for UTI b) administration of DES + - exp 10 68 unexp 4 74 OR = 2.72 (0.79-9.39) how do you interpret that? the confidence interval isn't necessarily greater than one. case control study to assess vaccine efficacy they are delivering bait with a vaccine and a marker in it so they can tell which animals are vaccinated. then they can catch the foxes and see what's up with them. rabid not rabid unvaccinated 18 30 vaccinated 12 46 what is odds ratio? 2.33 i think he said how do you interpret it? ---end----