Study design and summarising data Flashcards
What are RTCs?
Randomised Controlled Trials
* Intervention study
* Choice of treatment/intervention allocated randomly
* Typically randomised to new vs current or placebo treatment
* ‘Gold standard’ in research studies
Why are Randomised Controlled Trials random ?
- Ensures patients’ characteristics don’t affect which treatmentthey receive
- Unbiased
- Treatment groups balanced
- Any differences in outcome can be attributed to treatment received
- Fair test of efficacy
- Usually use computer-generated random list
How does Blinding in Random Controlled Trials take place?
Why?
- Treatment is concealed from either (single) or both (double) patient and assessor
- Reduce bias e.g. conscious or unconscious bias
- Double-blind: need placebo drug or sham treatment, not always possible
- Randomisation makes blindness possible
What is the analysis in randomisation controlled trials?
Intention-to-treat: analyse in randomised groups, even if don’t comply or if switch treatments
* Ensures balance remains - VERY IMPORTANT
* Fair test of offer of treatment
Per protocol analysis sometimes used as well , particularly if a large number of patients are not managed as planned.
Limitiations of Randomized controlled trials
- Can be expensive and difficult to organise
- Some interventions are impossible or unethical to randomise to
e.g. impact of lead on children’s intelligence
What are Cohort studies?
- Observational study- no intervention
- Subjects observed in natural state (real world)
- Investigates causes or factors associated with disease (or condition)
- Selects group of healthy individuals
- Follow-up to monitor disease state and possible risk factors over time
- Disease state (e.g. yes/no) according to risk factor status (e.g.smoking yes/no)
- Usually prospective (but retrospective designs may be used)
What are Advantages of cohort studies
- Data is collected prospectively, so estimates are less likely to be biased than those from a case-control
- RCTs can only investigate potential benefits for ethical reasons but cohort studies can investigate potential harm
- Imbalance between exposed and unexposed can be corrected in the analysis
- Cohort studies can look at exposure to treatment and harm
- eg impact of lead on children’s intelligence
What is the design of a weighted sample study?
- Subjects selected from large population on basis of outcome of interest
- All cases and a random sample of controls are included
- Full data collected on selected sample
What is advantages of weighted sample study?
- Can be carried out prospectively
- More cases from a smaller sample
- Greater power & accuracy without excessive cost
Disadvantages of a weighted sample study
- Sampling frame needed to identify weighted sample
- Analysis must correct for sample weighting
- Rarely used ( need to get a statition on board)
What is case-control studies
- Observational study - no intervention
- Subjects observed in natural state (real world)
- Investigates causes or factors associated with disease (or condition)
- Selects group with disease: ‘cases’
- Choose comparator group without disease: ‘controls’
- Compare cases and controls with respect to possible risk factors
- Usually retrospective
What are the limitations of case-control studies?
- Choice of control group affects comparison
- Data reported by subjects or from records - usually
retrospective so may be incomplete, inaccurate or biased
✓ BUT often quick to do and inexpensive!
✓ Evidence from case-control studies may be used in
planning further research
What are cross-sectional studies?
Observational study
* Subjects observed in natural state (real world)
* Collect data for each subject at one point in time only
* Similar to a cohort study, but without follow-up
What are cross-sectional studies useful?
Useful for:
* measuring prevalence of a disease/condition - e.g. % in population with asthma
* surveys of attitudes/views/behaviours - e.g. patient satisfaction, alcohol drinking
* Not useful for looking at cause & effect
Why summarise data?
- To monitor data quality
- To check for invalid or missing entries
- To describe characteristics of participants in a study
- e.g. 1st table in many research articles
- Before doing a complex analysis
What are the different types of data?
- Quantitative - Continuous or Discrete
- Categorical
What are the different types of quantative data?
- Continuous: lies on continuum – any value valid between range e.g. weight, height
- Discrete: data can only take certain values – usually integers, often counts e.g. number of children in a family
Individuals fall into one of a number of separate categories
* Can be 2, 3 or more categories:
* 2 categories: dichotomous or binary data
* >2 categories: can be ordered or unordered
Gender: male/female – dichotomous
* Disease status: alive/dead – dichotomous
* Stage of cancer: I, II, III, IV – ordered (‘ordinal’)
* Marital status: single, married, divorced, widowed, legally
separated – unordered (‘nominal’)
How are continuous data summarised?
Centre of data
* Mean (arithmetic average)
* Median (middle value when data ranked)
Spread of data
* Range (min, max; useful as descriptor)
* Standard deviation (‘SD’; shows spread of data; same units as data)
* Variance = SD2
* Interquartile range (IQR; middle 50% of ordered data when data split into
4 equal-sized groups or quartiles)
How do you calculate standard deviation?
Need average difference squared (d2)
* Divide sum of d2 by (n-1)
(Reasons for using (n-1) instead of n are complex. We only use n if
have whole population which almost never happens)
* This is called ‘variance’
* Square root of variance is standard deviation (SD) *Here:sumofd2 =52,n=9
* Variance = 52/(9-1) = 52/8 = 6.5
* SD=6.5 = 2.55
When to use which summary?
Quantitative data: spread of distribution
* Continuous data → use standard deviation (SD)
* Continuous data with skew – consider using IQR as well
* Continuous data → consider giving range (min to max) - useful in addition to SD if space allows
What is Unordered categories?
Unordered categories (‘nominal data’)
* Frequencies in each category
* Proportion or percentage
* Don’t clutter with too many decimal places
What is Ordered categories (‘ordinal data’)?
Frequencies in each category
Proportion or percentage
Cumulative proportions/percentages
How are histograms presented
Histogram: distribution of continuous data
* Shows shape of distribution, range, middle
* Areas in rectangles proportional to number in category
How are box plot presented and interpreted?
- median = horizontal line in box
- upper quartile = top edge of the box
- lower quartile = lower edge of box
- Whisker length up to 1.5 times width of box
- Outliers shown as dots
What are the different shapes of distribuition?
“Bell-shaped” distribution
* Roughly symmetrical
* Central peak with tails on both sides
Reported alcohol is positively skewed
* Positive skew - long tail on right hand side
* Common distributional shape in
medicine/biology
Gestational age of her baby is negatively skewed
* Negative skew- long tail on left hand side
* Less common in medicine/biology
What are the charectiristics of a normal distribution?
The Normal distribution has useful features relating to its mean and standard deviation (SD):
* About 95% datalies within mean ± 2SD
* About 68% datalies within mean ± 1SD
* Used for ‘Normal ranges’, t-tests, 51 regression etc.
What do bar charts show?
Why are bar charts used?
- Shows frequency or percentage in each category
- May be quicker to absorb than a table
Why are pie charts show and used?
- Size of slice, defined by angle, proportional to frequency in category
- 2 pie charts show health status in Bristol is similar to whole of England
- Popular but rarely used
in academic presentations – bar charts or tables preferred
