EBM Flashcards
What is pellegra and what did it prove for using scientific method?
Symptoms are the 4D’s: diarrhea, dermatitis, dementia and death.
Originally attributed to a microbial infection. Goldberger suggested it was a dietary problem. Victims didn’t eat vegetables, meat, and milk. He did a bunch of tests and they kept getting rejected.
It was demonstrated that niacin (vitamin B3) deficiency caused pellagra.
What is H pylori
Infection by this bacteria causes ulcers. They were met with ridicule. They infected themselves and showed that the bacteria caused ulcers.
What is EBM
its the conscientous, explicit, and judicious use of the current best evidence in influencing decisions about the care of individual patients.
What is RCT (randomized control trial?)
It means taking your sample and randomnly assigning a control and what your testing.
Bradford hill (waksman) discovered the treatment for strep throat, streptomycin.
Archie Cochrane said treatments are only effective if they do more good than harm via RCT
GIve an example of each.
Diagnostic question,
prognostic question,
treatment question
prevention question
- does my patient possibly have this diagnosis? what is abnormal? Should I perform a test and what are the risks?
- what should this patient expect following treatment, surgery, etc..
- what is the best treatment for the patient and which treatment has the lowest risk of harm?
- what can I or the patient do to prevent these conditions.
What is a population and what is a sample, difference.
A population is all the people in a setting with certain defined characteristics that is of interest. The sample are subjects from the population.
You are trying to make inferences from the sample about the population.
A representative sample looks like the population and a biased sample does not (lower external validity)
Exclusion vs inclusion criteria
Inclusion criteria identify who will be in the study and exclusion criteria identify who will not.
Controls variability (better internal validity) but limits generalizability (external validity)
Inclusion: 30-50 year olds who are either coffee drinkers or non coffee drinkers.
Exclusion: smokers.
Probability sampling: when each member of the population has an equal chance of being included in the research. Describe the three kinds of probability sampling.
Simple random sampling: the most common form of probability sampling
-every person in the population has a random chance of being selected.
Stratified random sampling: to ensure variables are expressed. Each trait is then randomnly sampled.
Ex. if you just randomly sample school age children, you may just end up with a biased result of kindergartners. SO you take kindergarten, 1st grade, 2nd grade and randomly sample from each.
Systematic sampling - where patient data is ordered and you set a parameter for the selection of the data of interest. Every 10th data.
What are three non-probability sampling. What are some problems?
Purposive sampling - you are picking patients that will likely support your hypothesis. Really biased sampling.
Convenience sampling - you are choosing patients based on their availability for your study, asking people to volunteer for studies and taking that as your sample. There may be external factors, like poor people may join for money.
Quota sampling: like convenience sampling but includes prereqs. You would have an ask to volunteer but include parameters to get a representative sample of wht you want.
What are extraneous variables? and examples?
Variables in the study that may effect the relationship between the independent and dependent variables. Diet, amount of sleep, medications.
What is selection bias and how is it different from sampling bias?
What is measurement bias
Selection bias, extraneous factors, measurement bias can all confound your study. What is confounding?
Selection bias occurs when comparisons are made between nonequivalent groups. Ex. when the control group and treatment group are littered with extraneous variables that are not standardized. The hospital procedures, etc.
Measurement bias occurs when the methods of measurements differ between groups.
Confounding: an extraneous variable correlates with both the dependent and independent variable. You are allowing a third variable to influence the effects.
Selection bias is about the groups you assign within the sample.
Sampling bias is about your selection from the population.
What is the Central Limit Theorem
If you have a normal distribution (with a mean and a random, independent variable) and your sample is large enough, you can be 95% confident that it will match the population distribution.
What is the sample distribution? (mention SD’s) and variables m and s
if your sample size is large enough, the sample distribution will be bell shaped.
One standard deviation includes (34.1 x 2), 68.2 percent of your data. Two standard deviations includes about 95%
s=standard deviation
m = population mean.
Parametric vs non parametric
Parametric means normal distribution, you assume a mean and a standard deviation.
Non parametric makes no assumptions about the shape or form of the probability distribution. We’ll see data tends to be a normal distribution.
Describe a chi square statistic
It is the only test for studies using categories (no measurement, just number of cases that fall under yes or no) It is non parametric, it uses categories, you compare the observed numbers to the expected numbers to calculate your chi square statistic. The larger it is, the more related they are. Null hypothesis is rejected.
What is the method of a chi square statistic?
To calculate expected. You take one box, multiple the total in the row with the total in the column divided by the total for each box.
Then its ((observed - expected)^2)/expected and sum all the values. If your chi square statistic is larger than the critical value accounting degree of freedom, then your categories are related.
How does a case control and a retrospective cohort differ? Case control study vs cohort study.
The sampling is what is different. Case control, you would identify people with the condition.
Retrospective cohort would first pick a group that were related to each other (like a class of juniors) then identify from among them the condition. And gather information about their life
Cohort studies measure how many people develop disease out of a total. Incidence in both groups
Case control studies look at people who already have the disease and determine the odds that the diseased group was exposed.
Case-control is you take people who have the condition and you look into the past and ask if they were exposed to something.
Calculated by (those with the disease who were exposed/ those who have the disease and weren’t) / (those who don’t have the disease and were exposed/ those who don’t have the disease and weren’t exposed)
Cohort is a group of people who have something in common and then observed for a period of time. (people all have propensity for breast cancer) they differ in the variable of interest. Study a “predictor variable” (something that will likely lead to a condition)
Cohort studies look at how many people develop the disease out of the total. It looks at relative incidence of disease. Not good for chronic diseases (super long studies), not plausible experiments, incidence of head trauma on bike accidents wearing and not wearing helmet, Can’t make them do that.
You calculate a risk ratio. Incidence of disease in those exposed and incidence in those not exposed (the baseline).
Ex. 20 in every 100 minors develop COPD. 4 in every 100 surface workers develop COPD.
(20/100) / (4/100) = 5. Five times as likely to get COPD. If CI includes 1, no effect.
What happens if CI includes 1?
It has no effect, CI can not include 1.
Describe three kinds of sampling bias.
Healthy- user bias: select participants that are health conscious
Berkson’s bias - select a population from an impaired or diseased group
Exclusion bias - excluding participants who have a certain characteristic.
What is investigator bias and how is it avoided.
Investigator knows the expected results and treats the groups differently, influencing the results.
Allocation concealment: people do no know which group subjects are in. Random, and subjects given numbers
Investigator blinding: do not know which group they are providing the treatment to (don’t know what the treatment is) and taking measurements to.
What is the hawthorne effect? What is double blinding?
What is recall bias?
People change their behavior in a study, they want to please and perform as expected. Make sure subjects don’t know which group they are in and investigator is blinded. But if there is an obvious change this doesn’t really work.
Patients and the researchers are blinded.
Single blinding usually just refers to participants being blinded.
Recall bias is a threat to internal validity of a case control study. People can not remember the exposure or event of interest.
what is absolute risk. difference between odds ratio and relative risk
incidence is the number of new cases over a period of time.
Absolute risk is the number of people with the condition/total number of people.
Odds ratio:
Exposed vs not exposed for both scenarios
People with the disease, exposed over not exposed
People without the disease, exposed over not exposed
A/C // B/D
Risk ratio makes more sense.
Incidence in people who are exposed over incidence in people who are not exposed.
Exposure compared to baseline
A/(A+B) vs (C/C+D) exposed vs not exposed.
Ranks the hierarchy of evidence.
Systematic review>RCT>Cohort>Case control.
Systematic review is a summary of all the research out there.
Describe RAndomized Control study and what to look out for.
THe best way to measure cause and effect. You sample a group from a population then randomly assign people to experimental or control groups. (Random assignment - something that can’t be done in observational studies)
First concern, getting your population of interest. You don’t want a sampling bias. You may want exclusion criteria to just specify to your population of interest. You want to minimize differences within your sample that may interfere with your conclusion (socioeconomic status, occupation, etc)
IN the example they did a run in experiment. It is a small experiment you do before starting your real one to exclude people that would mess up your data. In this case it was compliance of taking aspirin.
Random assignment is important because you hope you equally distribute extraneous factors into each group so they end up washing out. You could you stratified randomization where you take groups with a known potential variable and randomnly assign them to each group. You just want to remove selection bias.
Control for extraneous factors. You want to check because even random could result in inherent biases. There were no differences between the groups in terms of smoking, incidence of diabetes, age..etc.
Two ways to deal with drop outs or treatment changes?
Intention to treat: you include data on those who switched or dropped out but still keep their original group assignment. You want to increase variability to increase external validity.
According to treatment received: you ignore those who dropped out and for those who switched treatments, people will belong to groups that they were in when the study ended.
How does observational differ from experimental?
No random assignment, can’t manipulate independent variables, hard to exclude all extraneous factors.
What is risk factor exposure? And what are ways to quantify it?
Exposure means the person, before becoming ill, has come in contact with the factor in question. So on the charts, its people who develop the disease and don’t develop the disease.
Ever been exposed, current dose, largest dose ever taken, cumulative dose, years of exposure.
Dose response curve are different based on the factor of interest. For smoking you would do pack years and a total cumulative does. Radiation event may just be ever exposed.
What are the 6 kinds of environmental risk factors?
Chemical agents: workplace exposure, cleaning products, industrial chemicals, pesticides, environmental smoke.
Physical agents: radiation, noise, vibration. (waves..)
Biologic agents: infectious agents, bacteria virus, allergens)
Psychosocial agents: depression stress, trauma
Mechanical agents: bodily harm from physical exertions, heavy lifting, repetitive motions like texting, injuries.
Life style risk factors (behavioral) = smoking, taking drugs, sun exposure, unsafe sex.
(biological, chemical, mechanical, physical) look like immediate causes
(psychosocial, lifestyle) seem like distant causes, more remote causes (factors related to education, less prenatal care, malnutrition)
WHat is latency?
The time between exposure to event. Most conditions that lead to morbidity have a long latency so the cause is hard to deduce.
Immediate, days, years, or decades.
Lung cancer from smoking is 25-30 years.
Common exposure to risk factors made it hard to distinguish an effect they were having as was the case for smoking.
It is only by comparing patterns of disease with and without these risk factors (control) could you deduce. It would be most useful if you took a group of people with the same propensity of getting a disease and then exposed one group to a risk factor and the other not. Unethical. You would have to do observational. See how the disease comes up and draw conclusions.
Low incidence and small risk. Multiple causes and effect.
Many diseases have low incidence 2- 3%. Even if it increases risk by 20% for example, incidence is still tiny. this makes it harder to also experiment based on such rare diseases.
Small risk: refers to how the effects of individual factors for chronic diseases are small. Because the effect is so small, you need to study a large number of people to detect the differences and it will be harder to even figure out a connection and start testing it.
Most diseases are not one to one. one factor can influence many diseases while one disease may have many causes.
What does it mean risk factor may not be causal.
Schizophrenia doesn’t cause lung cancer.
Those with schizo are more likely to develop lung cancer
This is because they smoke way more.
Schizophrenia is a marker for increased likelyhood for lung cancer but is indirectly connected.
Additive and multiplicative effects?
Some risk factors may add onto each other for increasing the probability of getting a condition. BUt sometimes it is multiplicative and exponentially increases your risk. As is smoking ad asbestos on lung cancer.
50/total got lung cancer when smoking and asbestos worker. 10/total got lung cancer only through smoking. 5 times more likely. Total 50 times more likely that a non smoker (1/ total).
