General Flashcards
(272 cards)
1
Q
examples of interventional study designs
A
N-of-1 randomised trials
systematic reviews
randomised controlled trials
2
Q
examples of observational designs
A
cohort studies
case-control studies
cross-sectional studies
ecological studies
3
Q
examples of descriptive designs
A
case series
case reports
4
Q
functions of good research design
A
- enable a comparison
- allow the comparison to be quantified
- determine the temporal sequence of the risk
- identify the risk factor for the disease
- minimise third variable effects: bias, confounding
5
Q
features of cohort studies
A
- exposure measurement occurs before outcome ascertainment; subjects are recruited based on their exposure
- controls do not have the exposure
- exposure is measured at or near the beginning of the study
- all groups are followed through time
- the outcome is measured when it occurs; cases arise during the study
6
Q
in a cohort study, how might exposure be measured?
A
questionnaire, blood and tissue samples, pre-existing records
7
Q
in a cohort study, how might outcomes be ascertained?
A
- follow-up: is time consuming and expensive, can lead to false diagnoses
- pre-existing records: e.g. registries, medical records
8
Q
can risk be directly calculated in cohort studies?
A
Yes: absolute risk, attributable risk (risk difference), relative risk (risk ratio), survival curves
9
Q
advantages of cohort study designs?
A
- for rare exposures and common outcomes
- rigorous epidemiological design (able to directly measure incidence)
- provide temporal sequence
10
Q
disadvantages of cohort study design?
A
- expensive
- time consuming
- case and controls may differ on important outcome predictors “need to have good case and control selections so that confounding isn’t coming into the sample selection: once you select your sample, you can’t go back and repeat because the timeframes are long”
- can be susceptible to bias and confounding
11
Q
What are the types of bias present in cohort studies?
A
- confounding
- sampling bias (selection bias: a sample is collected in such a way that some members of the population are less likely to be included than others)
- migration bias (type of sampling, and selection bias: excluding subjects who have recently moved into or out of a study area)
- measurement bias (measurement of exposure or outcome is not similar between groups of patients studied)
- misclassification bias (measurement error)
(effect modification)
12
Q
How do prospective and retrospective cohort study designs differ?
A
Re when the cohort is assembled, in the past for retro, present for prospective
13
Q
Advantages of prospective cohort studies
A
- able to collect life and demographic data not available on medical records
- able to set up a standardised way of measuring exposure and degree of exposure to risk factors
14
Q
disadvantages of prospective cohort studies
A
- long duration
- expensive
- loss to follow up
- cannot be used for rare diseases
- inefficient because many more subjects need to be enrolled
15
Q
advantages of retrospective cohort studies
A
- more efficient than prospective cohort study because data is already collected
- cheaper than prospective cohort study because there is no need for long follow up
- faster because patient outcomes have already been collected
16
Q
disadvantages of retrospective cohort studies
A
- long duration
- expensive (less expensive than prospective)
- loss to follow up
- reliance on records; cannot examine a patient characteristic not already recorded
- measurement of exposure and degree of exposure may not be standardised; may not get a standard exposure measurement across different sites, or from patient to patient
- problem with time-dependent exposures; difficult to find temporal sequence
17
Q
Define relative risk/risk ratio
A
- ratio of risk in exposed to risk in unexposed persons
- “how many times more likely are exposed persons to get the disease relative to non-exposed persons?”
- relative risk = Risk in exposed / risk in unexposed where a risk ratio of >1 means the exposure increases the risk of disease while, =1 doesn’t change the risk of disease
18
Q
What are attributable and relative risk calculated from?
A
the incidence/absolute risk of an outcome in an exposed and unexposed group
19
Q
4 components of reporting risk
A
exposed group, unexposed group, the relative risk and the outcome e.g. “Oestrogen users have 1.27 times the risk of developing breast cancer compared to oestrogen non-users
20
Q
Calculate Population-attributable risk (PAR)
A
PAR = attributable risk x prevalence of the exposure in the population
or PAR = incidence rate in population - incidence rate in unexposed
21
Q
Define Population-Attributable fraction (PAF)
A
PAF = (population attributable risk / incidence in total population) x 100 or what fraction of disease in a population is attributable to exposure to a risk factor PAF = (incidence in total population - incidence in unexposed)/ incidence in total population x 100
22
Q
Prognosis
A
prediction of the course of disease following its onset
23
Q
Define prognostic factors
A
patient characteristics that are associated with the outcome of the disease
24
Q
Differences between risk and prognosis
A
- risk factor relates risk factor to disease, prognostic factor relates disease to outcome
- risk factors deal with healthy people, prognosis deals with sick people
- risk factors the outcome is usually disease onset, prognosis the outcome is the consequence of disease e.g. suffering, death, disability, complications
- risk factors usually for low probability events, prognosis factors are for relatively frequent events
- factors may be different; variable associated with an increased risk are not necessarily the same as those marking the worse prognosis
25
What is the difference between clinical course and natural history?
clinical course = evolution or prognosis of a disease that has come under medical care and has been treated in a variety of ways that affect subsequent course of events
natural history = prognosis of a disease without medical intervention
26
important features of prognosis studies
patient selection, time of entry into the study, follow-up, outcomes
27
What are the ways prognosis can be reported?
absolute risk, relative risk, odds ratio, 5-year survival, case-fatality, disease-specific mortality, response, remission, reoccurrence
28
How can you be sure a finding isn't due to chance / or is causal?
When confounding variables are absent, where there are no selection and measurement biases, with p values and hypothesis testing
29
How do you tell if a finding isn't due to chance/ or is causal?
hypothesis testing and p values
30
What is random error?
poor precision by chance alone
31
What is systematic error?
measurements differ from the truth in a systematic way - a degree of error is always inevitable e.g. blood pressure
32
When does selection bias occur? What does it result in?
occurs when comparisons are made between two groups that are different in ways other than the main factor under study
the difference between the two groups affects the outcomes of the study
33
What is the problem with using volunteers in studies?
Volunteers are generally more health conscious compared to the rest of the population (selection bias)
34
Types of selection biases?
- volunteers
- low response rates - control group, might not be as interested
- ascertainment or detection bias - an individual's chance of being diagnosed with a particular disease is relating to whether they've been exposed (not detecting the exposure correctly)
- healthy worker effect; those able to work are healthier than the broader population (excluding disability, elderly, severely ill)
- loss to follow up
35
How to control selection bias in case-control studies?
strong case definition, objective ways to assess this
appropriate control selection
high participation rate in both groups
clearly defined inclusion and exclusion criteria
36
How to control for selection bias in cohort studies?
maximise retention of the cohort (exposed and non-exposed)
where/if possible follow up those who dropped out
clearly defined inclusion and exclusion criteria
37
When does measurement bias occur?
if the methods of measurement are not similar among the groups of patients; either measurement of the exposure or outcome
38
What are types of measurement bias?
- recall bias - differences in accuracy of recall of memory e.g. someone inaccurately memory, mothers of children with leukaemia (was mum exposed to X-ray when pregnant mother might be more likely to recall that event than a mother with children without leukaemia)
- reporting bias - incomplete medical notes, some information considered unimportant
- interviewer or observer bias; errors due to individuals taking the measurements
39
Ways to control measurement bias?
- clear, precise and objective definition
- appropriate choice of measurement definition
- quality control
- is the measurement accurate and precise
40
What are special biases associated with screening tests?
- Lead-time bias: (Huntington's disease) overestimation of the survival time due to the backward shift in the starting point for measuring survival that arises from early detection procedures
- length-time bias: (slow vs fast growing cancer) a systematic error due to the inappropriate selection of long-duration cases
- compliance bias: because screening requires a patient to present and be regularly screening the type of patient presenting for screening is generally more compliant than those who are not presenting at screening
41
What is channelling bias?
type of selection bias. Drugs with similar therapeutic benefits are prescribed to groups with different prognoses: e.g. a new drug may be prescribed to a group with pre-existing morbidity, so a disease state may be incorrectly attributed to use of the drug.
42
What is data completeness bias?
type of measurement/information bias. Missing data can lead to different results recorded for the same finding.
43
What is surveillance bias?
type of measurement bias. "the more you look, the more you find" some patients are followed up more or have more diagnostic tests performed on them than others
44
What is partial verification bias?
measurement bias. In the 1980's CT scans were obtained when there was a headache present because clinicians assumed that headache increased the likelihood of haemorrhage as a cause of the neurologic deficit. If CTs were more likely to be obtained in the evaluation of acute neurologic deficits when headache was present, then a study of headache as a predictor of haemorrhage would result in higher sensitivity and lower specificity than if headache played no role in obtaining the gold standard CT
45
What is publication bias?
(Selection bias). When not only the quality, but the results and hypotheses tested influence whether or not a paper is published. Can bias systematic reviews. e.g. research with significant results are more likely to be published.
46
What is referral or verification bias?
Measurement bias. The results of a diagnostic test affect whether the gold standard procedure is used to verify the test result.
47
What is reporting bias?
Selection bias. Selective revealing or suppression of information by subjects
48
What is social desirability bias?
Measurement bias. The tendency of survey respondents to answer questions in a manner that will be viewed favourably by others.
49
What is spectrum bias?
measurement bias. The performance of a diagnostic test may differ between groups due to the different mix of people in each group.
50
Cumulative incidence equation?
number of people who develop disease in a specified period/ number of people at risk of getting the disease (at the start of the period)
- measures the proportion of (a group of) people who develop disease during a specified time period
51
Incidence rate / incidence density equation?
number of people who develop disease/ number of person-years when people are at risk of getting the disease
- measured how quickly people are developing a disease
52
equation for attributable risk?
incidence in exposed - incidence in unexposed
53
equation for risk ratio?
cumulative incidence in exposed / cumulative incidence in unexposed
54
sampling
process of selecting a representative part of the whole population for the purposes of determining parameters or characteristics of the whole population
55
3 characteristics to think about when determining a population
1. members of the population should be susceptible to the outcome
2. population relevant to questions begin asked
3. sufficiently described so that you can decide to which people the results of the study applies
56
purpose of statistics
draw inferences from samples about the underlying population or show how representative a sample is of a population - sometimes the degree to which the sample is representative of the population
57
Sampling error
differences between the sample results and the population characteristic being measured, needs to be kept small
58
factors contributing to sample error
bias and random sampling error
59
what is random sampling error
variations within the sample due to the natural variations seen between individuals - occur purely by chance
60
ways to handle random sampling error
- appropriate sampling techniques (e.g. random selection)
- appropriate sample size (power = statistical terminology): the larger the sample size, the greater variation is measured, the closer the sample value is to the true population variable (the more precise the measurement)
- appropriate statistical calculations measuring error occurring by chance (confidence intervals and p-values) - how much the association is there by chance
* can't just say the sample isn't big enough - re study critique - didn't do a para calculation, what about the techniques to select the sample? ways of measuring cause and effects?
61
What is a target population?
the collection of individuals about which inferences are desired
62
What is the study population?
the collection of individuals from which the sample is drawn
63
What is the sample?
the collection of individuals on which the investigation is conducted
64
What are the two main sampling methods
non-probability methods and probability methods
65
What are non-probability methods
- where the probability of choosing someone is unknown
- cheaper, but unable to generalise, potential for biases
e. g.
- availability sampling (people are easy to find),
- quota sampling
- convenience sampling: made up of people who are easy to reach e.g. pollster interviews shoppers at a local mall
- deviant case sampling (those who are extremes of that population are chosen)
- typical case sampling (averages are selected)
- snowball sampling (one person tells other people, other people tell other people - e.g. recruiting through Facebook, particularly good in communities hard to break into like drug addicts)
- expert sampling (people who know the most about it are selected - or point to various populations)
- critical case sampling (e.g. using small island nations to analyse the effects of climate change and apply it to larger landmasses)
66
What are probability methods
the probability of being sampled is known e.g. random sampling methods in RCTs
67
Importantly, when can statistical inferences be made?
Only when random samples are used.
68
What is simple random sampling (SRS)
akin to pulling names out of a hat
69
What is systematic sampling
taking the ith number of a list - less efficient than SRS
70
What is stratified sampling
taking SRS from mutually exclusive and exhaustive subdivisions of the population: more efficient than SRS
71
What is cluster sampling
taking SRS from defined clusters; less efficient than SRS
72
What is a type I (alpha) error
is the probability of finding a difference with the sample compared to the population, when there isn't really one.
- usually set at 0.05 or 5%
- OR the probability of rejecting the Ho when it is true
=false-positive conclusion
73
What is a type II (beta) error
the probability of not finding a difference that actually exists between our sample compared to the population, when there really is one. usually set at 0.2 or 20% depending on the study design, again an arbitrary allocation; or 20% chance of missing a true difference is considered reasonably acceptable
- OR probability of accepting Ho when it is false (power; power = 1 - type II error (beta)
=false-negative conclusion
74
what are the four factors sampling size is dependent on
1. level of significance required (p values, CI intervals)
2. difference between groups you wish to detect (large or small difference)
3. the variability of the estimate
4. cost
75
as the sample size increases... (sample size is dependent on four factors: significance, difference, variability, cost)
- the level of significance becomes smaller
- the difference becomes smaller
- the variability increases
- the cost per unit is reduced
76
related information generated and sample size
in general, the information generated is a squared function of sample size:
- to get twice the information, you need four times the sample
- to get three times the information, you need nine times the sample
77
clinical considerations about sampling and samples
- numbers are subordinate to method: a study with rigorous design and a small sample size is superior to a study with poor study design and a larger sample size
- methods and numbers are subordinate to cost: cost is always a factor
- all issues are subordinate to ethical considerations: need to make sure the sample is big enough that an effect can be detected if it is there or prove that it is not there + small enough that we are not asking more participants than is necessary to undergo the inconvenience of being a part of the study
78
What are the two approaches to chance?
- hypothesis testing (null, alternate hypothesis) - dichotomous outcome, either statistically significant or not e.g. uses the p value
- estimation: estimate a range of values in which the true value is likely to occur e.g. uses confidence intervals
79
If an RCT is testing whether Drug A is better, what are the four possible conclusions?
- Drug A is better than usual care and this is the conclusion of the study
- *Drug A is the same and has similar effects to usual care and the study concludes that a difference is unlikely
- *Drug A is the same and has similar effects to usual care, but it is concluded that drug A is more effective
- Drug A is more effective than usual care, but the study concludes that there is no difference between drug A and usual care
80
What is used to estimate the effects of random variation?
statistical testing
81
When bias is absent, what is responsible for statistical uncertainty?
random error
82
steps in how to approach a clinical question?
- construct an answerable question
e. g. 'what is the relationship between coeliac disease and the development of neuropathy'
- set decision threshold (comes from clinical experience)
e. g. a 20-30% increased risk of neuropathy is clinically important (5% level of significance)
- gather information
e. g. data collected on small intestine biopsies and matched to neuropathies between 1969-2008
- process information (mechanical act of dealing with large amounts of information - select the right graphical representation of the data, correct selection of statistical tests, cleaned the data)
e. g. risk of neuropathy HR = 2.5 with 95% CI (2.1-3.0), so the results are statistically significant because a HR of 1 = no effect and the CI doesn't include 1: p value of
83
What is inference?
drawing conclusions from data
84
What is statistical inference?
drawing conclusions from quantitative or qualitative information using statistical processes to describe and arrange the data and, in some cases, to test the suitable hypotheses
85
What are the two major forms of inferences?
estimates (point and interval - ways to describe the data) and hypotheses (use statistics to test the veracity of a hypothesis)
86
what is a point estimate?
use of sample data to estimate population parameters
- or the best estimate about the population from the data arising within the sample e.g. mean, SD, proportion
= 1 representative figure *the true effect size is unlikely to be exactly that observed in the study (random variation) so for any point estimate, need to have a summary measure to give the statistical precision of that point estimate
87
What is interval estimate?
- range of reasonable values containing the parameters with a certain degree of confidence
= estimate + and - [confidence x variability]
use of sample data to provide a range of reasonable values intended to contain population parameters with certain degrees of confidence
- the more narrow a CI, the more confident one can be about the true size of the effect
= the true value is more likely to be closer to the point estimate than the outer limits of the interval estimate
- 95% CI, 5 times out of 100, the true effect size could fall outside the limits or CI; gives statistical precision of the estimate
88
For data that is individual observations, what measure of spread is used?
Standard deviation
89
If measurements have been taken from different sample populations, and you want to graph the distribution of the grouped means, what measure of spreads are appropriate and inappropriate?
appropriate: standard error of the mean
inappropriate: SD
90
When is SD used?
to describe the spread in the frequency distribution
- spread of the individual data; but doesn't tell what is happening in the population, doesn't extrapolate data from a sample to a population
91
What is SEM?
standard error of the mean = standard deviation for the means
- helps to estimate the probable error of the sample mean's estimate of the true population mean
- standard error = SD / [(square root of N) - 1] where N is sample size; the larger the sample size, the smaller the standard error
92
Normal distribution values?
1 SD = 68%
1.96 SD = 95%
2 SD = 95.4%
3 SD = 99.7%
93
Equation for confidence intervals?
CI = mean (-1.96 x SEM), (+1.96 x SEM) = 95% CI, 95% sure the interval will contain the true population mean
94
When might confidence intervals be used?
to characterise the statistical precision of:
- any rate: incidence or prevalence
- diagnostic test performance
- comparisons of rates: RR, OR, HR
- other summary statistics
to get statistical significance at 0.05: if the point corresponding to no effect (e.g. RR =1 or treatment effect = 0) falls outside the 95% CI for the observed effect, the results are statistically significant
95
How should a confidence interval be interpreted?
- there is a 95% chance that the true population effect lies within the stated confidence limits
- if the study was repeated 100 times, 95 times out of the 100 the true effect would lie within the expressed confidence limits
- we are 95% sure that what we have measured in the sample is true for the population
* not there is a 95% probability that the parameter lies within the CI - obvious, we are 100% the sample lies within the 95% - not about accuracy of data, but that the data in the sample will be representative of the population
96
benefits of interval estimates?
more useful in clinical studies than hypothesis testing
- gives how much uncertainty is inherent in the sample statistic because it gives the interval on either side
- formula captures the uncertainty in a sample statistic by defining an interval likely to capture the population parameter with a specific degree of confidence
- the width of the interval estimate depends on:
- level of confidence (higher level, wider interval) e.g. 99% CI?
- variability (sample size: smaller sample, wider interval)
- standard error (larger standard error, wider interval)
97
Does statistical significance also mean clinical significance?
Statistical significance isn't always clinically significant; clinically, the effect might be really small, the range might be too wide to have much meaning; depends on the question asked, the patient themselves, clinical experience
98
What is hypothesis testing?
- drawing of inferences between competing hypotheses as a prediction of the examination of the data is going to show
99
What are the two types of hypotheses?
- null hypothesis (Ho): the hypothesis to be tested
- alternative hypothesis (H1): the hypothesis contradicting Ho - usually stated in the hypothesis
* all testing is done against the null hypothesis: we reject or fail to reject Ho
100
what are the steps in hypothesis testing?
- Construct Ho
- construct H1
- determine level of significance
- collect data
- *calculate test statistic*
- calculate p-value
- compare p-value with level of significance
- reach decision
101
What should be considered in choosing a test statistic (test statistic is step 5 of hypothesis testing)?
use the table in lecture 21. Predictor = exposure (was the data binary, categorical, ordinal or continuous)
+ outcome (binary, categorical, ordinal or continuous)
- what type of data is it, and is the data predictor or outcome data
102
What is the p-value? (step 6 of hypothesis testing)
- the p value is a quantitative estimate of the probability that a difference in the treatment effect in a particular study could have happened by random error assuming that there is no difference between the two groups
- the calculated level of significance at which we would be indifferent between accepting and rejecting the *null hypothesis* given the sample at hand: the smaller the p value, the less likely there is no difference between the two groups
- the probability that the test statistic or relationship would be as extreme or more extreme as the observed value if the null hypothesis were true
- "if there was no difference between treatment groups upon repeated studies, what proportion of the trials would conclude that the difference between the two treatments was at least as large as that found in this study" a large p value supports the null hypothesis that there is no difference between treatment groups
103
What is statistical significance in terms of p value?
- p value of
104
Comparing CI and p values, what are their advantages?
CI: emphasise size of the effect, allows the reader to see the range of plausible values and determine the clinically meaningful results, provide information about statistical power
p values: tradition, convenience, sometimes CI aren't feasible
105
what is statistical power?
the probability of finding a difference when a difference really exists
- or the probability that a study will find a statistically significant difference when a difference really exists
- where power = 1 - beta
= a determinant of sample size
106
What does a t-test do?
compares means of a continuous variable in two research samples e.g. treatment and control groups
107
What are the two types of t-tests?
- A student t-test: used if the samples come from two different groups
- a paired t-test: used if the samples come from the same group (e.g. systolic BP measured in a group of women before and after exercise)
108
what is a z-test?
works like a t-test but compares the proportion of two groups
109
When will an ANOVA test be used (analysis of variance)?
dependent variable is continuous, independent variable is categorical (nominal, dichotomous, or ordinal)
110
What is an N-way ANOVA test?
if more than 1 independent variable is used, N is the number of independent variables
111
What test for continuous predictor (age) and a binary outcome (cancer)?
logistic regression
112
What test for continuous predictor (birth weight) and continuous outcome (systolic blood pressure)?
Pearson correlation; linear regression
113
What is correlation?
a measurement of the strength of the association between two study variables
- strength of a linear relationship
- often referring to the Pearson's product moment correlation coefficient, denoted by 'r'
114
What is regression?
- ability to get line of best fit: a predictive model
derives a prediction equation for estimating the value of one variable given the value of the second
- determines line of best fit; slope and intercept
- data occurs in pairs of measurements (X, Y)
115
What does the correlation coefficient 'r' do?
defines both the strength and direction of the linear relationship between two variables
- correlation coefficient is an index between -1 and +1
- when r = -1, there is a perfect negative linear association
- when r = +1, there if a perfect positive linear association
- when r = 0 there is no linear relationship
* r doesn't always indicate causality or clinical significance
* r is affected by extreme values - normal curve is useful
116
What is the general equation for the sample and population regression line:
Y with a hat = b0 + b1X where X is the predictor, b1 is the slope and b0 is the y-intercept
Y = Beta0 + betaiX
117
What is the method of least squares?
the statistical technique for finding the regression line
118
How does a regression line test statistical significance?
does the slope beta 1 differ from zero?
| "does the slope of the population regression line beta 1 differ from zero"
119
What is the coefficient of determination (r^2)?
defines the proportion of the total variation in values of the dependent variable (Y) that can be attributed to its linear relationship with the independent variable (X)?
120
Equation for r^2 or the coefficient of determination?
r^2 = sum of squares due to regression / total variation in Y
121
Compare the number of variables dealt with in correlation vs regression
Correlation deals with 2 variables, regression is limitless
122
Other types of regression not covered in Year A but in journal articles...
```
y = binary: logistic regression
y = count: link function called Poisson Regression
y = time to event: Proportional Hazards Regression
```
123
What is a common statistical test used in clinical research when hazards ratios are used?
Cox proportional hazards: time-to-event outcome
124
What is survival analysis?
statistical picture of the survival of a group of patients in the form of a graph showing the percentage survival vs time
125
Purpose of survival analysis?
report prognosis in such a way that information can be inferred about the likelihood that patients with a certain prognostic factor will experience an outcome at any point in time
126
When is a Kaplan-Meier analysis used?
with any dichotomous outcome that occurs only once during follow-up
127
What are the best estimates, for a given set of data, for the probability of survival for members of the cohort?
points of the survival curve
128
In a survival curve, the rate of outcomes (decreases/increases/remains unchanged)
Can be any: can remain unchanged because the rates of survival are being applied to a diminishing group of people
129
When might a hazard ratio be used?
Hazard ratio essentially analogous to relative risk with a time component - the cumulative relative risk over a time period across all groups. E.g. might compare different prognostic factors and the chances of survival for each (multiple survival curves graphed to compare prognostic factors)
- the chance of an event occurring in the treatment arm/ the chance of an event occurring in the control arm
130
What are the features of a case-control design?
- subjects are recruited based on disease: outcome ascertainment occurs before exposure measurement
- exposure for both cases and controls has occurred at the time of recruitment.
- controls = similar individuals without the disease
- the frequency of past exposure is compared between cases and controls
- relative risk is estimated.
131
What are some important design elements of a case-control study?
- Case selection 1. ideally, cases (those with the disease) are going to be incident cases. Presence and duration of the disease. This is because the risk factor for a disease is whether or not you have it and duration of disease (how long you have the disease before you recover/die). It's difficult to separate what the effect of the duration of the disease is. 2. Can enrol prevalent cases: used for rare diseases, looks like a cross-sectional study. Problem of determining temporal sequence for exposure-disease relationship. Can also get survival bias (missed severe cases who died before recruitment). 3. Case definition: crucial to case selection. Must only select people who are at risk of exposure (e.g. only war veterans in studying depression on war veterans).
- control selection. 1. Must be as similar as possible to cases except for absence of disease.
132
Equation for prevalence when prevalence of the disease is low (
= incidence x duration
- or the number of people with disease at a given point in time / total number of people in the population
* assumes a stationary population
133
What are the approaches to control selection?
- population approach: randomly drawn from population
- cohort approach: *nested case-control*
- hospital and community controls: hospital controls taken from the same hospital but with a different outcome. Community controls are selected from the community that surrounds the hospital. Weakness of community controls; anyone could go to a hospital.
- multiple controls per case. Improves the ability to determine a difference in risk profile. Get an improvement for every additional control per case up to 3 or 4, then there's no improvement.
- matching: cases and controls are matches according to a particular characteristic that may be (strongly) related to the outcome. To reduce variation between groups except for risk factor.
- negative controls: no-comorbidities.
134
How might exposure be measured in case-controls?
- thoroughly completed medical records or stored samples from lab analysis. Questioning of cases and controls or appropriate proxies (good interviewer and questions; recall and record bias).
135
Advantages of case-controls? (similar to retrospective cohort study)
- can look at multiple exposures. e.g. people with lung cancer. Can look at exposure to smoking, bacteria, air pollution, work exposure.
- rare outcomes, common exposures.
- relatively fast
- relatively inexpensive
136
Disadvantages of case-controls? (similar to retrospective cohort study)
- dependence on appropriate control selection.
- prone to bias.
- not good for measurement of rare exposures.
137
What is a nested case-control study?
A case-control study/ a group of cases and controls that have been drawn from a cohort population that has been followed for a period of time.
- Might look back at past medical records for additional information about exposure (on top of the findings of the cohort study).
138
How might risk be estimated in case-control?
Can't get a risk as subjects are recruited based on outcome.
Can compare relative exposure frequency in cases and controls.
Can use odds ratio
139
Define odds ratio?
The odds that a case is exposed/ the odds that a control is exposed
or the proportion that the event does happen/ the proportion that the event does not happen
140
What does an OR > 1 mean?
the frequency of exposure is higher among cases than controls: the greater the odds ratio, the stronger the association
141
What does an OR
the frequency of exposure is lower among cases than controls: indicates protection
142
The is OR approximately equal to RR?
when the incidence of disease is low (typically less than 10%). It depends on the size of the RR to actually determine how low the incidence must be before the OR is approximately equal to the RR.
143
Which studies rely particularly on accurate diagnosis of disease?
Both cohort and case-control studies rely on accurate diagnosis of disease.
- Assess diagnostic test.
- Ability to correctly interpret a diagnostic test.
- Ability to conduct a sequence of tests in order to rule in and rule out possible diagnoses until the correct diagnosis is reached.
144
What are the two ways the effect of an intervention can be measured?
- Observational: in a cohort study, intervention = exposure
| - Experimental/Interventional: clinical trials
145
What is a clinical trial?
prospective investigation comparing the effects of an intervention with a control
gives more information about causation than observational studies
146
Define intervention re RCTs?
factors introduced into the lives of the research subjects and expected to change their outcomes
147
Define controls re RCTs?
the status quo for research subjects that is not expected to change their outcome
148
Describe RCTs
- Select a sample from the population
- Individuals are randomly assigned to groups
- One group is given the treatment, the other is control
- Check for outcomes in both groups.
149
Advantages of RCTs
- most scientifically rigorous design
- provides the best evidence of causation
- can give data when observational studies cannot
- clear cut aims
- clear cut endpoints
- clean and precise (well matched groups, clear inclusion and exclusion criteria)
- deals with incidental outcome-related factors, many other sources of bias
150
Disadvantages of RCTs
- the setting tends to be artificial (trade-off between closely matched groups and being able to apply that to a broader community)
- adherence to protocols may be difficult
- inclusion and exclusion criteria are still relevant
- costly and resource-intensive
- may require many subjects
- ethical concerns preclude many experiments 1) patient sacrifice, 2) no patient can be exposed to significant risk
151
What are some limitations of RCTs
- the existence of several RCTs does not mean that a clinical question is settled
- practical limitations may reduce feasibility
- rare conditions
- reluctance to allocate treatments randomly
- long time before results are known
- practice may be well-established
152
What are some important design elements of RCTs
- Ethics. Under what circumstances is it ethical to assign treatment at random, rather than as decided by patient and physician. Beneficence, non-maleficence, respect for autonomy (informed consent), justice (equity, impartiality, fairness). Equipoise: the randomisation is ethical if there is no compelling reason to think that one of the allocations is better than the other. They must know that they can withdraw from the study at any time. The trial must be stopped if there is enough evidence that it is effective, if there is significant harm, or if there is futility in continuing.
- subject eligibility. Subjects are required to meet rigorous criteria, promotes homogeneity and strengthens the internal validity at the expense of external validity. Inclusions involve applying strict diagnostic criteria. Exclusions usually include co-morbidities, survival, contraindications, refusals and non-compliers
- intervention. 1) generalisability. Is this intervention likely to be implemented in usual clinical practice. 2) complexity. does the intervention represent the normal complexity of the real-world treatment? 3) strength. Is the intervention sufficiently different from currently available alternative management strategies?
- comparison groups. 1) no intervention group, 2) comparison group that is part of the group (Hawthorne effect - patients adjust their behaviour while being observed, patients who volunteer often work hard to make sure good results are obtained), 3) usual care group, 4) placebo treatment, 5) an alternative intervention - comparison may be patients who receive the current best treatment
- outcome measurements. objective measurement?, If there are composite outcomes, is the effect size of each component assessed separately?, are the outcomes clinically important (not just statistically)?
153
Hawthorne effect
cumulative change in the outcomes for a number of things. The total effects of treatment are the use of the contributions made by the natural history, the Hawthorne effect (effect that people in the study put in effort to get a good outcome), usual care, placebo effect and a new intervention *RCTs must be able to filter out all the above effects except for the new intervention
154
How is treatment allocated in RCTs?
- randomisation: provides the best opportunity to create truly comparable groups (baseline characteristics in Table 1 - ideally prognostic factors)
- measures differences in the groups that may occur after randomisation: patients disease status, compliance, cross-over (patients move from one intervention group to another), co-interventions (patients receive a variety of interventions other than the one being studied - patient decides to take up an exercise regime)
- blinding: can occur at various different levels in the study design
155
What is blinding?
Strategy used to ensure participants are unaware of the treatment group they are in. Ensures behaviour and reporting does not systematically affect the results. Blinding can occur at 4 different places in an RCT: 1) after randomisation and before the treatment: investigators allocating participants to group aren't aware which treatment is being assigned next, 2) patients or participants are unaware of the treatment group they are in, 3) physicians caring for people in the study may be unaware of the treatment the patients are receiving - patients are treated equally regardless of their intervention 4) assessment of the outcome: outcome measurements are assessed by people unaware of the treatment group the participants are in; preventing bias in measurement or analysis
156
Treatment effects can be measured in 3 ways?
1. Relative risk reduction
2. Absolute risk reduction
3. Number needed to treat
157
What is relative risk reduction? What is the equation?
the amount by which the treatment has reduced the relative risk.
RRR = (control event rate - treated event rate / control event rate)
or 1 - RR; where RR > 1 = relative risk increase (RRI)
158
What is the absolute risk reduction? What is the equation? (same as attributable risk - different contexts)
- the absolute difference in rates in the control group compared to the treated (experimental) group
= control event - treated event rate (benefit)
- absolute risk increase = treated event rate - control event rate (harm)
159
Define number needed to treat? What is the equation?
- number of patients who would have to be given the experimental therapy in order to prevent one adverse event (e.g. death or other complication) from occurring, in a specified period.
NNT = 1/ (control event rate - treatment event rate)
NNT = 1/absolute risk reduction
*number to treat should be rounded up, for whole people
160
Define number needed to harm? What is the equation?
number of patients who, if they received the intervention would lead to 1 additional patient being harmed, during a specified period.
NNH = 1 / (treatment event rate - control event rate)
NNH = 1 / absolute risk increase
161
Define intention to treat analysis
- Principle of analysing outcomes of the RCTs according to the groups that participants were originally assigned in regardless of whether they underwent the intervention.
- Advantages: provides information to the clinician about what treatment choice is best at the time the decision is made? - do you offer the treatment? Important that groups are analysed according to how they were originally randomised in order to retain full benefit of randomisation.
- Disadvantage: can mask the effectiveness of the treatment.
162
What are some types of RCTs?
parallel group studies, clustered studies, cross-over studies, factorial design, N-of-1 trials, sequential trials
163
What occurs in a parallel group study (RCT)
two or more groups are treated separately or concurrently
164
What is a clustered RCT?
naturally occurring groups are randomised and outcomes are counted in patients according to the treatment their group was assigned e.g. randomising patients according to the ICUs that they were admitted into, might be different infection rates in different hospitals.
165
What is a cross-over design?
Patients are divided into 2 or more groups. Each group starts with a different treatment. After a time (washout period), treatment assignments change. Effectively, each patient is exposed to the intervention and the control. Requires fewer patients.
e.g. drugs with a short half-life.
166
what is a factorial design?
E.g. Group 1 = surgery A drug A, group 2 = Surgery A, drug B, group 3 = surgery B drug A, group 4 = surgery B, drug B
Allows the evaluation of many combinations of treatments simultaneously. Complex, but cost effective.
167
What is a N of 1 trial?
= random collection of periods of active and placebo therapy for a single patient. (washouts, random order)
- according to what is best for the patient
- must be blinded (patient and physician)
- evaluates the efficacy of the treatment for long term use
- treatment must be reversible to allow for repeated treatment
- outcome is often patient preference or symptom score
- useful for determining treatment where the activity of disease is unpredictable e.g. migraines, asthma, fibromyalgia
168
What is a sequential trial?
- patients are allocated to whichever treatment has been most successful in previous patients, not commonly used
- usually reserved for dangerous therapies, vulnerable patients, or deadly or rare conditions
- random allocation to group A or B; A has the better outcome so then the randomisation is repeated but weighted so that more participants are in group A then the outcome is measured
169
What is a systematic review?
- the identification, selection, appraisal and summary of primary studies addressing a focussed clinical question using methods to reduce the likelihood of bias
- or an unbiased summary of available evidence of a given topic
- reproducible
170
What are the methods used in systematic reviews? (6)
- Define the question
- conduct the literature search
- apply inclusion and exclusion criteria
- assess the quality of the studies
- conduct analysis
- summarise the evidence
171
What is important in 'define the question'
- Use the PICO method
- specify inclusion and exclusion criteria
- state the methodology: time period of the review, language (might include only english written articles - expenses associated with translation), publication restrictions (using peer-reviewed journals, using popular news sources?, databases used?)
172
What does PICO stand for?
```
fundamental principle in evidence based medicine. An answerable question. Careful inquiry into precise healthcare problems.
P: patient/population
I: intervention / exposure
C: comparison / control group
O: outcome
```
173
What is important in 'conduct a literature search'
- Decide on information sources: databases accessed, experts consulted, funding agencies, pharmaceutical companies, hand-searching, personal files, Cochrane database, citation list or retrieved articles
- identify titles and abstracts
174
What is important in 'applying inclusion and exclusion criteria'
- usually done by a minimum of 2 or 3 researchers
- pre-set inclusion and exclusion criteria
- apply inclusion/exclusion criteria to titles and abstracts: obtain full articles for eligible titles and abstracts, apply inclusion/exclusion criteria to full articles, select final eligible articles - assess agreement on study selection
175
What is important in 'assessing the quality of the study'? (critical appraisal)
- data abstraction: participants, interventions, comparison intervention, study design - make sure they all fit with the methodology
- results
- assessment of methodological quality
- assess agreement on validity assessment
176
What is important in 'conduct analysis'?
- determine methods for generating the pooled estimates
- generate the pooled estimates (where appropriate)
- explore heterogeneity, conduct subgroup analysis
- explore the possibility of publication bias (publications that only publish positive results)
177
What is important in 'summarise the evidence'
- Forest plots: shows the point estimate of effectiveness and CI for each study in the review
- boxes represent the point estimates with the size of the box being proportional to the size of the study (the bigger the box the greater the participants in the study)
- vertical line marks the point where neither the intervention nor the placebo was more effective
- horizontal line marks the CI
178
What are some of the features of a Forest plot?
summarises
- number of studies: the rows show the number of studies met the inclusion and validity criteria
- what studies and when: the first column represents the study and the year of publication
- pattern of effect sizes: the point estimates taken as a whole, show what the various studies reported for effect sizes.
- precision of estimates: the vertical line through the boxes.
- the effects for the "big" studies: the large statistically precise studies (seen by narrow CI and large boxes) deserve more weight than the small ones
179
What is a meta-analysis?
statistical technique for quantitatively combining the results of numerous studies measuring the *same outcome* (- studies must be similar-) into a single pooled or summary estimate
- combines the results from individual studies to increase power
- must have homogeneity of subjects, interventions and outcome measures
(-often in SR)
180
What are other types of systematic reviews?
- literature review: an overview (not reproducible, can be different), not comprehensive
- a narrative/unsystematic review: discussion piece, doesn't look at quality of study or do quantitative analysis
181
What is the Cochrane collaboration?
worldwide collection of systematic reviews for medical interventions using RCTs
- set up by Archies Cochrane (doctor/epidemiologist); by not having a repository where summaries of these studies were, the profession was not taking advantage of the relevant RCTs, had no systematic process to put them into practice
- study resources are always limited, if there were properly designed evaluations in one repository, then the studies could be applicable and used in the medical profession
182
What does public health have to offer?
- descriptions of the health status of populations
- causation
- evaluation of interventions
- natural history
183
define epidemiology
study of distribution and determinants of health-related states or events in specified populations, and the application of this study to control health problems
184
define clinical epidemiology
- defined as the science of making predictions about individual patients by counting clinical events in groups of similar patients and using strong scientific methods to ensure that the predictions are accurate
the application of epidemiological knowledge, reasoning and methods to study clinical issues and improve clinical care
185
What is meant by frequency?
How often does the disease occur?
186
What is meant by abnormality?
is the patient sick or well?
187
What is meant by risk?
What factors are associated with an increased risk of disease?
188
What is meant by prognosis?
what are the consequences of having the disease?
189
What is meant by diagnosis?
How accurate are tests used to diagnose disease?
190
What is meant by treatment?
How does treatment change the course of disease?
191
What is meant by prevention?
Does an intervention on well people keep the disease from arising? Does early detection and treatment improve the course of disease?
192
What is meant by cause?
What conditions lead to disease? What are the origins of disease?
193
What are the 5D's or important events in clinical medicine?
death - bad outcome if untimely
disease - a set of symptoms, physical signs and laboratory abnormalities
- discomfort - symptoms such as pain, nausea, dyspnea, itching and tinnitus
- disability - impaired ability to go about usual activities at home, work or recreation
- dissatisfaction - emotional reaction to disease and its cause, such as sadness or anger
194
What is evidence based medicine?
- conscientious, explicit and judicious use of current best evidence in making decisions about the care of individual patients
the application of clinical epidemiology to the care of patients
195
What are the three key parts to making a clinical decision?
best available evidence, clinical expertise and patient preferences
196
What are the two fundamental principles of Evidence Based Medicine?
1. EBM builds a hierarchy of evidence
2. Evidence alone is not sufficient to make a clinical decision. Also needs to be considered:
- benefits vs. risks
- inconvenience
- costs
- alternative management strategies
- patients values and preferences
197
Skills needed to practice EBM?
diagnostic expertise, in-depth background knowledge, effective searching skills, effective critical appraisal skills, ability to define and understand benefits and risks of different treatments, in-depth physiologic understanding allowing application of evidence to the individual patient, sensitivity and communication skills required for full understanding of the patient context, ability to elicit and understand patient values and preferences and apply them to management decisions
198
Basic principles of EBM?
variables, numbers and probabilities, populations and samples, bias and confounding, validity and chance
199
How can you describe a disease?
Person - Who is becoming ill
Time - when are they becoming ill
Place - where are they becoming ill
200
Characteristics of time?
- changing or stable?
- clustered (epidemic) or evenly distributed (endemic)?
- time-trends: point source, propagated, seasonal, secular, combinations
201
Characteristics of place?
- geographically distributed or widespread (outbreak, epidemic, pandemic) offshore (tsunami), travel
- climate effects (temperature, humidity, combined effects) e.g. probably not zika virus in Australia because of different mosquitoes
- urban / sub-urban squatter/ rural
- relation to environmental exposure (water, food supply)
- multiple clusters or one?
202
Characteristics of person?
age, gender, socioeconomic status, marital status, ethinicity/race/genetic profile, behaviour/habits
203
Define case definition?
a set of diagnosis criteria that must be fulfilled in order to identify a person as a case of a particular disease
- critical component of descriptive epidemiological study
* provide background, clinical criteria, laboratory criteria - do these fit with the patient
* * there are layers in case classification: probable and confirmed
204
Define variables?
measurements of characteristics which have the potential of varying among observations or in the same observation over time.
205
Define data?
collection of observations and variables on which inferences are drawn
206
What is nominal/categorical data?
Groups are mutually exclusive.
Neither rank nor magnitude.
e.g. religion, race, sex, blood type
207
What is ordinal data?
at least 3 options.
rank, not magnitude.
e.g. small, medium, large; pain rating; heart murmurs grades I-IV
208
What is interval data?
- ratio (interval) value has a true zero e.g. temperature, time, length, volume
both rank and magnitude.
e.g. age, body weight, degrees celsius
209
What is validity/accuracy?
degree to which the data measures what it is intended to measure.
e. g. physical examination, standard protocol is supported with another finding (blood test) to see if they are both in agreement
- accuracy of measuring pain, depression difficult
- results are grouped around the true value
210
What is reliability? (or reproducibility or precision)
extent to which repeated measurement of a stable phenomenon (measured by different people, instruments, times) get similar results
- results are grouped around a narrow range
211
When might variation occur?
- biological variation within and between individuals
| - measurement variation between instrument and observer
212
Define, adv and disadv of mean?
= sum of values of observations / number of observations
- well suited for mathematical manipulation
- affected by extreme values
213
Define, adv and disadv of median?
= the point where the number of observations above equals the number below
- not easily influenced by extreme values
- not well suited for mathematical manipulation
214
Define, adv and disadv of mode?
Most frequently occurring value
- simplicity of meaning
- sometimes there are no, or many, most frequent values
215
What are mean, median and mode measures of?
central tendency
216
Define, adv and disadv of range?
from lowest to highest values in a distribution
- includes all values
- greatly affected by extreme values
217
Define, adv and disadv of SD?
the absolute values of the average difference of individual means from the mean
- well suited for mathematical manipulation
- For non-Gaussian ("normal") distributions, does not describe a known portion of observations
218
Define percentile, decile, quartile
the proportion of all observation falling between specified values
- describes "unusualness" of a value. Does not make assumptions about the shape of the distribution
- not well suited for statistical manipulation.
219
2 ways of describing data? (reproducibility and accuracy)
Central tendency (how accurate) and dispersion (how reproducible)
220
What does a positively skewed curve look like?
the tail is towards the positive end
221
What does a Kurtic distribution look like?
widespread with big peak, seen in >20,000 patients
222
What does a Platykurtic distribution look like?
widespread, flat top, seen in
223
Define normal distribution?
describes the frequency distribution of a repeated measurement of the same physical object measured with the same object
*clinical distributions are not normal therefore 2SDs doesn't collect 95% of variables
224
How is the distinction between normal and abnormal?
not distinct, there may be overlap between normal distribution and abnormal distribution
225
How to classify 'abnormal'?
unusual, associated with the disease, treating the condition leads to a better outcome
- functional decline: ability to do everyday activity vs. U-shaped BMI against mortality rate (lower BMI, higher mortality due to muscle wasting from sickness)
- depends on whether BMI relating to mortality or BMI relating to functional decline?
226
Difference between linear regression and correlation?
Linear regression finds the best line that predicts Y from X. Correlation is almost always used when you measure both variables. It rarely is appropriate when one variable is something you experimentally manipulate. Linear regression is usually used when X is a variable you manipulate (time, concentration, etc.)
227
Difference between primary, secondary and tertiary prevention?
Primary prevention aims to prevent disease or injury before it ever occurs e.g. health promotion re drinking, smoking.
Secondary prevention aims to reduce the impact of a disease or injury that has already occurred i.e. have early disease e.g. screening detects presence of disease in healthy people.
Tertiary prevention aims to soften the impact of an ongoing illness or injury that has lasting effects - targets those with fully symptomatic disease e.g.surgery on people with coronary heart disease to prevent future events.
228
What is the general proportion used in expressing clinically relevant frequencies?
- numerator is the number of patients experiencing the event
| - denominator is the number of people in whom the event could have occurred (population)
229
What is point prevalence?
prevalence is measured at a single point in time for each person
230
What is period prevalence?
prevalence present at any time during a specified time period
231
equation for incidence?
number of people who develop disease in one year / average number of people in the population
232
What are other ways of measuring frequency?
```
mortality (death rate per 10^n)
morbidity
case-fatality rate
survival rate
cause-specific
attack rate
```
233
What are ways burden of disease might be expressed?
```
life expectancy
potential years of life lost (PYLL)
quality-adjusted life years (QALY)
health-adjusted life expectancy (HALE)
disability-adjusted life years (DALY)
```
234
Define risk?
the likelihood or probability of a positive or negative event occurring
235
Define risk factors?
characteristics associated with an increased risk of an event occurring
236
Define exposure?
exposure to a risk factor means that a person, before becoming unwell has either come into contact or manifested a certain characteristic that clearly increases the probability of a particular outcome (disease), that for the purposes of the study is the characteristic of interest
237
How might exposure be measured?
duration of exposure, current dose, largest dose taken, total cumulative dose, years of exposure, years since first exposure
238
Define latency period?
time between exposure and first manifestation of disease
239
Often, the causes of disease are obscured in poor recollection or complex interaction between physiology, behaviour, society and exposure
- common exposure to risk factors: cigarette smoking, high salt high fat diets, alcohol exposure
- low incidence of disease: e.g. lung cancer most common cancer but still a very low incidence - average GP clinic will see relatively few cases of lung cancer, difficult to draw inferences from infrequent events
- to detect a small risk, the number of people observed needs to be large enough to see the risk and see difference in exposed and unexposed group
- for common risks, difficult to disentangle the independent effects of risks, hard to correlate between exposure and disease
- multiple causes and effects: usually not a close 1 to 1 ratio between a risk and disease *one risk factor might contribute to multiple diseases e.g. stroke
240
Define cause (of a disease or event)?
is an event, condition or characteristic (or a combination of these factors) which plays an important role in producing the health outcome (disease)
241
What is a sufficient cause?
```
the factor (or combination of factors) that if present, will inevitably result in disease
*each sufficient cause has a necessary cause as a component
```
242
What is a component cause?
a factor that contributes to disease causation, but is not sufficient to cause disease on its own
243
What is a necessary cause?
```
any factor (or combination of factors) that is required for the development of disease. Disease can never be present when the factor(s) is absent.
*have to have the factor, but they may not have the disease
```
244
Types of factors in causation?
Predisposing: age, sex
Enabling/disabling: low income or nutritional status
Precipitating: exposure to risk factor
Reinforcing: repeated exposures
245
What is descriptive epidemiology?
branch of epidemiology that describes the amount and distribution of diseases within a population
246
Features of ecological studies?
- special because of time or geography
- examine relationship between groups of individuals with exposure to a putative risk factor and an outcome
- exposures are measured at the population, community or group rather than the individual level
- provide information about an association: may generate or support existing hypotheses
- prone to bias: ecological fallacy - relationships observed in groups won't necessarily hold true for the individual
* can be quickly done, difficult to interpret, not good for causation, may establish links
247
Features of cross-sectional study design?
- the observation of a defined population at a single point in time or during a specific interval. **Exposure and outcome are determined simultaneously.
- quick and relatively easy
- useful for determining: prevalence of risk and frequency of prevalent cases
- prone to bias
- not useful for rare outcomes or exposures, events of short duration
- no information on temporal relationship between exposure and outcome
248
Features of case-control study design?
- patients are sampled by outcome; subjects are recruited on the basis of disease (outcome)
- compared to a group (control) without the outcome with respect to the specific exposure
249
Advantages of case-control study design?
- good for rare outcomes and common exposures
- relatively fast
- relatively inexpensive
- assess multiple exposures
250
Disadvantages of case-control study design?
- depend upon appropriate controls selection
- prone to bias
- not good for rare exposures
251
Define controlled clinical trial?
participants are assigned to groups and the investigator specifies the exposure for each participant group. Followed over time and outcome is measured and compared between the two groups.
252
Define randomised controlled trial?
A clinical trial where the assignment of participants to a group is at random
253
What are measures of association?
a wide variety of coefficients that measure the statistical strength of the relationship on the variable of interest
e.g. absolute risk, attributable risks, relative risks, odd ratios, population attributable risks, population attributable fraction
254
What are the four ways to measure risk?
1. Have been exposed and have the outcome
2. Have been exposed and don't have the outcome
3. Haven't been exposed and have the outcome
4. Haven't been exposed and don't have the outcome
255
What is absolute risk?
the probability of an event in a population under study
256
Define attributable risk? What are its synonyms?
- the additional risk (incidence) of disease following exposure, over and above that experienced by people who are not exposed
- the risk of disease in exposed individuals that can be attributed to the exposure
- attributable risk = incidence in exposed - incidence in unexposed
= risk difference / absolute risk reduction (ARR) or absolute risk increase (ARI)
257
Define epidemic
the occurrence in a community or region of cases of an illness, specific health related behaviour or other health-related events, clearly in excess of normal expectancy
258
Define endemic
the constant presence of a disease or infectious agent within a given geographic area or population group
259
Define pandemic
the epidemic affects a large number of people and crossed international boundaries
260
Define cluster
an aggregation of relatively uncommon events (or disease) in a space and/or time in amounts that are believed or perceived to be greater than could be expected by chance
261
How does an epidemic come about?
- new appearance of, or sudden increase in, an agent from the environment or from an infected source e.g. change in migration patterns of birds
- arrival of new susceptibles e.g. decreased immunisation
- introduction of effective route of transmission e.g. climate change bringing new mosquito species
262
steps in outbreak investigation
- prepare for fieldwork: prepare for the unexpected, immediate response, limited investigation window, research risk factors and previous outbreaks of the disease, make administrative arrangements like supplies, equipment and personnel
- establish the existence of an outbreak (provisional case definition) e.g. is there a common cause, are there more cases than expected, are there relationships among cases, or are these sporadic, unrelated cases of the same disease, are these cases of similar but unrelated diseases, what constitutes occurrences clearly in excess of normal expectancy - depends on lots of factors inc. temporal variation; - causes of false alarms, change in surveillance or reporting policy, change in case definition, improved diagnosis, improved diagnosis, errors in diagnosis (false positives), increased public awareness and demand for tests, increased reporting
- verify the diagnosis: take history for clues, establish clear picture of events, speak directly with people who are affected, confirmation using other diagnostic tests
- define and identify cases: establish/refine a case definition - start broadly and simply, clinical criteria, epidemiological components, laboratory investigations, refine with multi-stage definitions as investigation progresses, identify and count cases
- characterise the outbreak: undertake descriptive epidemiological techniques to orient the outbreak in terms of person (demographic characteristics and risk factors - denominator problems: who is at risk), place (maps) and time (epidemic curves - provides a clear temporal picture, can characterise the type of outbreak and help identify putative agents, graph frequency by time): 5 W's What (diagnosis: symptoms, microbiology), who (person: age, sex), When (time) - background rate, time of onset, temporal trends, Where (place) - geographic extent, clustering, Why (causes) - exposures and risk factors
- develop hypotheses: data collection through interviews and questionnaires, is the hypothesis feasible
- evaluate hypotheses
- refine hypotheses: analytic techniques - descriptive statistics, simple measures of associations (RR, OR, RD) with CI, stratified and multivariate analysis, appropriate interpretation of causal relationships, consideration of new hypotheses if needed
- implement control and prevention measures: often happens concurrently, should occur as soon as information is available, actions to prevent further spread are implemented based on suspected or verified hypotheses, often can't wait until all facts are established, media and information outlets are important vehicles: 4 broad control measures - 1) sanitation, 2) prophylaxis, 3) diagnosis and treatment, 4) control of disease vectors
- communicate findings: keep information flowing during an investigation, clearly state the facts, beware of legal implications, aim to improve practice, have a single spokesperson, effective public relations strategy at the time of an outbreak is critically important
263
What does an epidemic curve look like when there is a point source?
epidemic curve has a steep upslope and a gradual downslope
264
What does an epidemic curve look like when there is a continuous extended source?
Epidemic curve has a sharp upswing, followed by a plateau, then a right tail
265
What does an epidemic curve look like when there is a propagated/contagious source?
Serial transmission leads to an epidemic curve with progressively taller peaks
266
Use of cohort studies in epidemics?
- follow a large group of exposed and unexposed people, assume a strong suspicion of the source
- calculate RR from the attack rates in exposed and unexposed people
- attack rate = terms for CI in outbreaks
267
Use of case-control studies in epidemics?
- compare exposure among cases and controls
- analysis involves calculation of OR
- ideal in clearly defined single outcome, rare disease; large or unidentified population at risk of disease, multiple exposures suspected
- disadvantages include choice of controls, recall bias
268
What is confounding?
- when the causal relationship is confused because of the presence of an additional factor of interest that is associated with both the cause and the outcome
- present when you stratify or adjust for a potential confounder and the effect estimate changes by 10%: if the crude and adjusted estimate effects differ by 10%
- e.g. exercise to obesity prevention; possible confounders are age, sex, BMI, smoking, family history, diet
269
Which confounders are unlikely to have much effect on the results of the study?
- if the association between the confounder and either the exposure or the outcome is weak
- if the confounder is rare; if more people in the study will not be exposed or many of the people in the study are exposed
- those confounders to take of are thus relatively common and strongly related to exposure and outcome
270
What are ways to control confounding?
implemented during design phase:
- randomisation: patients randomly assigned into intervention/control
- restriction: limiti the number of patient characteristics
implemented during data analysis:
- matching: match each participant in the intervention group with a participant having the same characteristic in the control group
- stratification: data is analysed and results are presented according to sub-group analysis
- standardisation: mathematically adjust the crude rate for a confounding variable so that the relative weight for that confounding variable is the same in each group.
- multivariable adjustment: mathematically adjust for difference in large number of variables simultaneously e.g. logistic regression or cox proportional hazards
- best-case/worse-case analysis: describe how the results would differ under extreme assumptions of selection bias
271
Features of effect modification (or "synergism or antagonism" or "interaction")?
- occurs when the presence or absence of an additional variable changes the effect of exposure on disease: the variable has an association between the exposure and disease but it increases or decreases the effect
272
How does effect modification differ from bias and confounding?
- bias needs to be removed or prevented
- effect modification is therefore a finding to be reported
- epidemiological analysis is generally aimed at eliminating confounding and discovering and describing effect modification
