Variation Flashcards
What is observation?
Actual data result in front of you
What is tendency?
True or underlying trend or relationship
Relationship between observation and tendency?
Observed is best estimate of true tendency. Testing hypothesis to do so
What is a hypothesis?
A statement that an underlying tendency of scientific interest takes a particular quantities value.
When testing the hypothesis, use and state the null as your hypothesis.
What is calculated probability?
P-value
What does a small p-value indicate?
Hypothesis is incompatible
If p-value equal to or below 0.05?
Reject hypothesis (null) at 95% confidence level as observations are statistically significant hence data is inconsistent with the hypothesis and strong evidence against hypothesis.
At P=0.05, there is 5% chance that it is due to chance (the null) whereas 95% chance that A and B are connected. 0.05 is the cut off, hence a lower p-value the better.
Is p-value > 0.05?
This does not prove the null hypothesis (even is P=0.999), just can not reject null hypothesis.
There is a greater than 5% chance that the effect is due to chance hence significant value
Bones to pick about p-value?
P<0.05 is arbitrary as nothing much changed
Sample size affects it
Statistical significant not alway clinically relevant
What is the 95% confidence interval?
We can be 95% sure that the true value of any measure lies between
(Observed value/error factor) - (Observed value X error factor)
How to calculate error factor for incident rate or prevalence rate?
2(square root (1/d))
e
d= events observed in population
How to calculate error factor for incidence rate ratio?
2(square root (1/d1 + 1/d2))
e
d1 & d2= events observed in each population
How to calculate error factor for SMR?
2(square root 1+O)
e
O= observed number of events in population
Method to doing an IRR and statistical relevance question?
State you H0 - null that there is no difference
IR for both group
IRR to give value X
Error factor of value X
Upper and lower CI
H0 is the value 1, hence if 1 features in CI there is
insufficient evidence to reject the null hypothesis as we can not say with confidence that men are more likely to have a stroke than women as p>0.05
Define cohort study?
Group of people with know exposures / personal characteristics
Process of concurrent cohort studies?
Start with 2 groups disease free people of same characteristics. 1 group with exposure, the other without.
Follow up over time for set period of time.
Calculate IRR and how many & which diseases occurs in p-y form.
Analyse and interpret
What is another name for concurrent?
Prospective
What is the historical cohort study process?
Start with 2 groups disease free people of same characteristics. 1 group with exposure, the other without. Classify initial exposure using records
Follow up using historical records until a set time.
Calculate IRR and how many & which diseases occurs in p-y form.
Analyse and interpret
What is another name for historical?
Reterospcetive
Why are cohort studies good?
Comparative study
Measures co founders
Can see the effects of a specific exposure
What is internal comparison in cohort study?
When you subdivide the level of exposure to give relative risk.
Compare exposed and unexposed within cohort.
What calculation is used in internal comparisons)
Incidence Rate Ratio
Characteristics of internal comparisons?
Very different sized cohorts - small exposed means larger error factor
Large study required
Confounders
What is an external comparison?
Compared those exposed to external reference population - eg the general public
Calculation in external comparison?
Standardised Morbidity Rate
Characteristics of external comparison?
Complicated to standardise as the expected # cases. The cohort study members increase age hence complicated to work out age-specific rates.
Can study serious diseases using large cohorts
Limited population data
Healthy worker effect
What is the healthy worker effect?
A selection bias.
Study involves workers/employed to reference population. Workers tend to be healthier than unemployed hence can work.
Any test involving exposure to something at work.
Comparison, should always be against other working people to prevent bias
What is used to calculate the expected #cases in external comparison of cohort studies?
Lexis diagram
What is a lexis diagram?
Shows the age-sex rates observed for given age group and calendar period
What are the age-sex rates in lexis diagram limited to?
Routine data source
What does an arrow on a lexis diagram mean?
Period of follow-up for an individual in the cohort study
What does each square of a lexis diagram represent?
The p-y of observation for a given age group& calendar period
When are external comparisons good?
Good when able to use sub-cohorts as limited data
Limitations of external comparisons?
When no incidence data
Might only have mortality data
No comparible ie HWE
Limited data on reference pop
How are cohort studies better that case-control?
Ranges of outcomes
Can study rare exposures
Exposure proceeds outcome
Positives to cohort?
Detailed assessment of exposure, outcomes and co founders
Better than case-control
Better for conditions that fluctuate with age
Negatives to cohort?
Time and resource intensive as large
Risk of high loses to follow up hence survivor bias
Not good for rare outcomes as too few cases
Unknown cofounders can’t be adjusted for
What is non-responder bias?
The characteristic of non-responders may be different to those responding to surveys he be despite a larger sample, a low response rate does not necessarily result in more of a representative of the pop.
What type of survey sample gives the most representative response?
Large sample size and large response rate
When creating a survey you need to think about?
Definition of disease and the types Method of diagnosis Sampling frame Sampling proportion Sampling technique Response Interpretation of data
Smaller the survey?
The higher the chance variation
Out of median and means, which is less influenced by extreme values?
Median is less influenced
Disadvantage to mode?
Misleading in skewed distribution
Can routine and administrative data make errors?
Yes.
Either interpretation of data or coding error.
