Statistics Flashcards
statistics allow professionals
to make well informed decisions
- understanding how accurate and valid data is
data should always be considered in
context
e. g. he later had a blood test which gave him a reading of 113ml of alcohol in 100ml of blood- the limit is 80’
- this doesn’t even make sense
how can stations give advice on efficient data collection?
is it representative?
what can we actually measure? for how long for? outcome? long short?
stages of statistical investigation
1) understand the problem and formulate in statistical terms
2) plan the investigation and collect the data
3) assess the structure and quality of the data
(scrutinising the data for errors, outliers and missing obs)
4) continue the initial examination of the data to describe the data
(use of summary stats, graphs and tables)
5) select and carry out appropriate statistical analysis of the data
6) compare findings with any previous results and acquire further data if necessary
7) interpret and communicate results
the first step in any medical stat investigation should be
- getting a clear understanding of clinical/ biological background to the situation under study
- clarify objectives
- formulate the problem in statistical terms
what can be a statisticians most valuable contribution
the investigator explaining why they wish to do the experiment
types of error
type I, II, III
Type I error
- finding a diff when there is none
- want to avoid this- why p values and cut off values exist
Type II
- failure to find a cure difference
- concluding the two treatment arms are the same
- often due to underpowered studies
Type III
conclusion are not supported by the data presented
- getting the right answer for the wrong question
when visiting the GP immediate question asked are
- When visiting the GP, the immediate question most patients ask are:
- What is wrong with me (diagnosis)?
- How should it be treated (treatment effectiveness)
- What does the future hold? (prognosis)
- With more reflection, patients may ask further questions
- Why me?
o Role of genes, environment, lifestyle etc - Could it have been prevented
these medical questions can be answered using
stats
what is wrong with me
- Consider a women who goes to see a doctor about a lump on her breast, worried it may be cancer
- In fact, she should be cautiously positive because past data tells us that 9/10 breast lumps are not cancers
- We can express our diagnostic uncertainty using conditional probabilities
- Prob (not breast cancer) lump= 0.9
- Prob (breast cancer) lump= 0.1 (10% chance)
- There is only a small chance the lump is cancerous, but breast cancer is serious so the doctor will organise further tests
How should it be treated?
- There are three main ways to deal with cancer:
- Surgery
- Radiotherapy
- Chemotherapy
- Most radiotherapy is given using photon particles but researchers had the idea that neutron particles, which are bigger and heavier, might be better for some cancers
- How could a fair test be conducted?
how could a fair test be conducted
randomised control trial of neutron therapy
explain how a randomised control trial of neutron therapy would work
1) cancer patients needing radiotherapy randomised to either treatment with photon or neutron theory
2) this randomisation prevents bias in who receives treatment and makes characteristic in the samples equally distributed, or there due to chance
3) patients followed up to see how long they lived
4) using survival curves- looking at time for event to happen i.e. ddeath
first clinical trial
- Scurvy affected sailors deprived of fresh food
- James Lind of the Royal Navy 1747
- Wrong theory: putrefaction preventable by acids
- 23 scorbutic sailors divided into 6 groups
- same diet plus:
- a quart of cider
- 25 drops of elixir vitrio
- 6 spoonful’s of vinegar
- half a pint of seawater
- two oranges and a lemon
- a spicy paste plus a drink of barley
- provided evidence for the role of vit C in treating scurvy
RCTs are now regarded as
the gold standard
- exponential growth since 60s
- the least bias due to blinding and randomisation
observational studies
investigators play passive role
- treatment or exposure variable is not under control of the researchers because of ethical concerns or logistical constraints
types of observational study
- cross sectional
- cohort study
- case control
- ecological studies
case-controls can
be prospective and retrospective
negative of case-control
cause vs effect
ecological studies
aggregates units and looks for association
e. g. Hep E looking at pork consumption and how much this correlated with Hep E
- data points are for countries overall
why use observational studies
- cheaper
- quicker
- possible to study groups often excluded from clinical trials
- gives better estimate of what happens in routine practice - patients in trials may be more compliant with treatment
- where ethical consideration prevent conduct of trials
groups commonly excluded from clinical trial
- vulnerable populations
- pregnant
- women of child bearing potential
- children
- those who can be unduly influenced
- ill people
- smokers
- frail and elderly
ethical considerations prevent conduct of trials
- testing a chemical potentially toxic effect
- testing drugs in children when drugs are unlicensed
Case study: Maternal smoking and infant health
- Many women smoke during pregnancy, even though widespread warnings on harmful effects
- Compare birth weights of babies born to smokers and non-smokers to see if they corroborate this warning
a variable
is a set of characteristics that describe an aspect of the patients in a research study
type of variables
binary- discontinuous e.g. smoking status
birth weight
continous
number of pregnancies (0,1,2,2+)
discrete number of values
- ordinal (ordered categories
mother marital status
nominal
types of variables
categorical or quantitative
categorical
ordinal (ordered) or nominal (unordered)
statistical distribution
describes the different values that occur and the frequency with which they occur for a given variable
we can summarise distributions by using
numbers -descriptie stats and graphs
summary statistics :categorical
- frequencies
- relative frequencies
summary statistics: quantitative
measure of location: averages (mean, median and mode)
measure of variation: SD, IQR
Histograms
show distribution, including multiple modes, skewness and tail size
histogram and box and whisker plots
identify outliers
compare location and variation in several groups
box and whisker
scatter plots
display general form of relationship between two variables
bar chart
display frequencies of categorical variables and cross tabulation of categorical variables
