WEEK 1 Flashcards
Hierarchy of evidence
Evidence based medicine: integration of best research evidence with clinical expertise and patient values
Systematic reviews
Critically appraised topics [evidence synthesis]
Critically appraised individual articles [articles synopses]
—filtered information
Randomised controlled trials
Cohort studies
Case controlled studies. Case series/reports
—unfiltered information
Background information/expert opinion
Case series
Tracks subjects with a known exposure
Report on characteristics of a group subjects with a particular condition
Eg if several people with oesophageal cancer drink hot tea does drinking hot tea causes cancers?
Cross-sectional study
Snap shot/one time point. Ask people questions about outcome and exposure
Measures prevalence health outcomes or determinants in population at one point in time
Only describe relationships not causality
Case control studies
Compare people with condition to without
Retrospective look backward in time to identity exposures
Confounders: another variable associated with outcome and independent variable, is this variable driving the relationship instead. Statistical adjustments
Reverse causality: the outcome is causing them to use the risk factor eg oesophageal cancer causing them to drink hot tea
Useful if you have a rare outcome , recruit knowing outcome
Cohort study
Longitudinal study
Take a large group of people
Sharing a common characteristic followed over long period time to study and track outcomes
Identify group before develop disease
Identify causes
Take lots of other important measurements
Need to adjust and account for confounding factors
The RCT
Highest quality evidence for a single study but might not be most appropriate study design to use
Measure effectiveness of a new intervention or treatment compare to alternative
Participants randomly assigned to different treatment groups helps reduce selection bias and balances known and unknown factors that could influence outcome
Systematic reviews and meta analysis
Systematically identify all RCTs or other studies
Combine the results
Provide comprehensive overview of existing evidence while meta analysis analysis data to get a more precise estimate of effect size, summary statistics
Forest plots
What is data
A set of values of qualitative or quantitative variables about one or more persons
Information about a person or group of people such as:
-demographics- eg age, ethnicity
-health information- eg heart rate, temperature
-diagnostic related information- eg blood test results
-disease registry- eg cancer status
What is statistics in practice
Collect data:
-surveys/patient notes
-for different individuals
-for different characteristics (called variables)
Collate data:
-database
-each individual constitutes a row in database and each variables constitutes a column
Summarise data:
-averages or location
-spread or variability
Types of data
Quantitative or numerical:
-represents quantity
-measures of values or counts and can be expressed as numbers
-how much, how many, how often
Qualitative or categorical:
-represents quality
-measure of type
-what category
-report numbers and percentages
Numerical data- continuous
Numerical data- measures of values or counts and can be expressed in numbers
If data can take any numerical values within a possible range then it can be described as continuous
Examples:
-age
-height
-weight
Numerical data- discrete
Numerical data- measures of values or counts and can be expressed as numbers
If the data can only take a value of a whole number within a possible range then its discrete
Examples:
-number children in family
-number GP visits per year
-number of teeth
Categorical data- binary or dichotomous
Categorical data- describe characteristic, they are mutually exclusive ( can only belong in one group) exhaustive (every person falls into a group)
If there’s only two possible categories its binary
Examples:
-alive/dead
-disease present/absent
-success/fail
Categorical data-ordinal
Categorical data- describe characteristic. They’re mutually exclusive, exhaustive
If there’s 3 or more possible categories, and they can be logically ordered or ranked then its ordinal data
Examples:
-academic grade
-clothing size
-cancer stage groups
Categorical data- nominal
Categorical data- describe a characteristic. Mutually exclusive and exhaustive
3 or more possible categories and there’s no logical ordering then it can be nominal
Examples:
-eye colour
-ethnicity
-marital status
Summarising continuous data
Plot data on graph
Need to be able to describe:
-values for average/location
-measures of variability
-distribution shape
Histogram
Summarising continuous data- average and variability
The average value and variability in data is described using:
-mean and standard deviation: useful when no outliers but if there are then the mean will be pulled in the direction of the outliers
-median and IQR: more robust to outliers
Summarising continuous data- averages-mean
Idea of quoting a single value which is most “typical” for measurement
Computed by summing the observations and dividing by sample size
Useful for counts/measurements, depending on shape
Outliers can make mean atypical
Summarising continuous data- variability- standard deviation
How much does data vary from average
Standard deviation explains how far on average a measurement is from the mean
Small standard deviation= numbers close together
Large standard deviation= numbers are spread out
Standard deviations are affected by outliers
Summarising continuous data- averages -median
Identified as half way value when the data is put in rank order
Often used for counts and measurements
Not affected by outliers
Not affected when extreme values are unknown
Summarising continuous data-percentiles
0th percentile=
25th percentile
50th percentile= median
75th percentile
100th percentile
IQR- interquartile range
-75%-25%
Summarising continuous data- averages- mode
The mode is the most commonly occurring category
-very rarely used
For numerical data may state the number of modes when describing distribution
Eg unimodal or bimodal
Summarising continuous data- distributional shape
Negatively skewed away from zero leaning, negative direction towards 0
Normal no skew perfectly symmetrical distribution- mode, median and mean all same place
Positively skewed - leans towards zero. Positive direction away 0
Identifying shape of distribution helps us determine what summary measurements should be reported
Normal distribution- mean and standard deviation reported
Skewed distribution- median and IQR reported
Baseline tables
Descriptive tables
Frequencies and percentages for categorical variables
Means and standard deviations or medians and IQRs as appropriate for continuous data
Comparative studies
The majority of epidemiological studies compare outcomes between two or more groups
-randomised controlled trials RCT
-comparative cohort study
-case control study
Treatment vs exposure effects
Treatment effect- usually from RCT
Exposure effect- usually from observational study
The same statistics can be used in both cases
-interpretation differs
Common statistics used for treatment and exposure effects
Analysis depends on outcome type
Binary outcome:
-relative risk RR
-risk difference RD
-odds ratio OR
Continuous outcome:
-mean difference MD
Describing risk and probability
In 10000 control patients 1200 had CVD event in 1 month
Probability 1200/10000=0.12
Percentage 12%
Risk is 12 per 100
1 in (100/12) ~8
Odds= 1200/(10000-1200=8800)=0.14
Risk ratios/relative risks
Relative risk= the probability (risk) of event on treatment/ probability (risk of event) on control
Risk ratio=1 : risk equal in intervention and control arm
Risk ratio >1: risk outcome greater in treatment arm
Risk ration <1: risk outcome less in treatment greater in control arm
Interpretation relative risk
Express as percentage increase or decrease in risk
(RR-1) x100
Eg 3.1-1 x100=210% increase in risk on treatment versus control
0.6-1x100=-40% increase=40% decrease in risk on treatment versus control
If exposure instead say risk in exposed group versus unexposed group
Express risk as percentage of risk in other arm:
Eg RR=3.1-> risk on treatment is 3.1x100=310% of risk control
Or RR=0.6x100=60% of risk of control
Risk in intervention arm is RR times the risk in control arm:
-risk on treatment is 3.1x risk in control
-risk on treatment is 0.6 x risk in control
Risk difference
RD= probability of event on treatment- probability of event on control
RD=0 risk of outcome equal in treatment and control
RD>0 risk of outcome greater in treatment
RD<0 risk of outcome less in treatment greater in control
Risk difference interpretation
Say absolute risk 2.4% lower in treatment than control
Receiving aspirin rather than control leads to absolute reduction of 2.4% in risk of MI
24 out of 1000 patients given aspirin rather placebo would avoid MI within 1 month (risk measured in 1 month)
Number needed to treat
The number of patients who on average need to be treated to prevent one event that would otherwise occur
NNT is 1/absolute risk difference
If RD=0.024 (ignore sign) NNT=1/0.024=42
Treating 42 people (for a month) will on average prevent one event
Relative versus absolute effects
Relative measures are commonly reported
Relative measures can look large where event is rare
Absolute measures less susceptible to misinterpretation
NNT is effective way to communicate to lay population
Odds- another way of measuring chance
Odds: number with event/number without event
If 9 had outcome event and 1 didn’t in trail
-odds are 9/1=9
If 1 had outcome and 9 didn’t odds=1/9=0.11
Odds ratio
Odds ratio= odds of event on treatment/odds of event on control
Computes the odds in treatment divide by odds in control
Similar to RR for rare events
Can be calculated in case control studies where RR can’t: less intuitive
OR=1 odds equals in intervention and control
OR>1 odds of outcome greater in treatment arm
OR<1 odds of outcome less in treatment arm greater in control
Say if OR=0.78
The treatment reduces odds to 0.78 78% of odds of control
Reduces odds of event by 22%
Odds of event on treatment are 0.78 times odds of control
Issue with odds ratio in RCT
Odds difficult to interpret
So odds often interpreted as risk
This can lead to an overstatement of effect size
Mean difference
Comparing a continuous outcome between two groups
Mean in group 1- mean in group 2
Compute mean in treatment group subtract the mean in control group from it
Treatments and exposure effect summary
Use treatment and exposure effects to compare outcomes between groups
Choice of measure depends on type data in outcome variable
Key comparative effect measures
Binary outcome:
-relative measures
—risk ratio/relative risk
—odds ratio
-absolute measures
—risk difference. Number needed to treat
Continuous outcome:
-mean difference
Cohort study a and d
A cohort study follows a group of individuals over time to determine the incidence and risk factors of certain outcomes. Subjects are typically categorised based on exposure to a specific factor or intervention
Advantages: can study multiple outcomes for a single exposure. Provides temporal info enabling inference about causality
Disadvantages: can be time consuming and expensive, susceptible to loss to follow up which can introduce bias
Case control study
Compares individuals with a specific condition (cases) to those without (controls) to identify potential exposure factors. It’s retrospective in nature assessing exposure in the past
Advantages: efficient for studying rare conditions, less time consuming and costly compared to cohort studies
Disadvantages: prone to recall bias, as it often relies on participants memories. Does not provide incidence rates (only odds ratio)
Systematic review
Synthesises evidence from multiple studies on a specific question, often using rigorous and predefined criteria. It may or may not include a meta analysis
Advantages: provides a comprehensive understanding of existing research on a topic. Reduces biases through systematic methods
Disadvantages: quality is dependent on the included studies. Potential for publication bias (over presentation of significant results)
Randomised controlled trial
Assigns participants to different interventions randomly aiming to determine the effect of the intervention. It’s considered a gold standard for clinical research
Advantages: minimises confounding factors due to randomisation. Allows for causal inference
Disadvantages: can be expensive and time consuming, potential ethical issues, especially if denying treatment
Case series
Describes characteristics, treatments and outcomes for a group of patients with a particular condition or treatment. It lacks a comparison group
Advantages: useful for rare diseases or novel treatments, relatively quick and easy to produce
Disadvantages: cannot establish causality due to lack of a control group. Prone to selection bias
Cross-sectional study
Assesses individuals at a single point in time to determine the prevalence of an outcome and its associated factors. It provides a “snapshot” of a population
Advantages: relatively quick and often less expensive than longitudinal studies. Good for assessing prevalence
Disadvantages: cannot determine causality due to its cross sectional nature. Susceptible to confounding, as exposure and outcome are assessed simultaneously
