EBP - Summarising Data and RCTs

variable

characteristics for which information is recoded

categorical variable

have non-numerical values, with each value associated with a category

Ordinal/ Ordered Categorical

can assign numbers to categories / ranked

Nominal/ Unordered Categorical

can not assign numbers to categories

Binary/ Dichotomous

two categories only

numerical variables

have numerical/ measured values

continuous

can potentially take an infinite number of distinct values

Discrete/ Count

can only take integer values

can numerical variables be converted into categorical?

yes for simpler summary of information

ordinal or binary

never nominal

can categorical variables be converted into numerical?

no

ALSPAC

avon longitudinal study of parents and children

due date april 1991- december 1992

graph for single categorical/discrete variable

bar chart

pie chart

making variable ordinal improves graph

graph for single continuous variable

histogram

has group intervals

if converted to ordinal it can be done as bar/pie chart

graph for continuous outcome X categorical/discrete exposure

box and whisker plot

compare distributions

graph for a pair of continuous variables

scatter plot

graph for a pair of categorical/discrete variables

no graph

presented in table

prevalence

proportion with the outcome at a particular point in time

incidence

proportion of new cases with the outcome that occurred in a specific time period

new diagnosis and immigration of ill people

what does prevalence and incidence not include

cure

death

emigration

normal distribution

symmetric distribution described by mean and SD

gaussian

skewed data

positively skewed (long tail to right)

negatively skewed (long tail to left)

measures of location

mean median mode

also known as measures of central tendency

when is the mean suitable?

continuous

normal distrubution

when is the median suitable?

continuous normally distributed

continuous non-normally distributed

discrete variables, with extremes/ outliers

when does the mode not exist?

if data has been measured precisely so each value can be distinguished

measures of variability

standard deviation, interquartile range

also known as measures of spread

standard deviation

measure of spread of observations around the mean

varience

SD^2

when is SD suitable?

continuous approximately normally distributed variables

when is IQR suitable?

continuous non-normally distributed or discrete variables

why is the range not that useful?

relies on two extreme values - unrepresentative

95% reference range

= mean +/- 1.96\*SD

likely values for an individual in population / ‘clinical normality’

limitations of clinical experience

patients may get better anyway

placebo effect

no controls

follow-up is a haphazard

small numbers of patients

placebo

phenomenon where patients symptoms can be alleviated by an otherwise ineffective treatment, as they expect the treatment will work

overcoming limitations of clinical experience

randomly allocate patients to a an active treatment group or control group

randomised control trial

random allocation

every patients has the same chance of ending up in either group

why are RCTs at the top of the hierarchy of evidence?

best evidence of associations

confounding eliminated

temporal sequence known so no reverse causality

bias minimised

in an RCT what is the control group?

placebo or the current standard treatment

in an RCT what is the exposure

whether or not they receive treatment

PICO model for clinical questions

patient/problem/population

intervention

comparison/control/comparator

outcome

3 steps of an RCT

planning

conducting

analysing

planning an RCT

establish H

define EC I C BC

calculate SS PC

specify O

establish a hypothesis

define eligibility criteria

define intervention

define control

define baseline characteristics of interest

calculate sample size/power calculation

specify outcome measures

why should the eligibility criteria not be too restrictive?

results of the study will only be generalisable to a limited range of patients in clinical practice

how to check successful random allocation?

done by comparing baseline characteristics/confounding factors across treatment groups

similarities between the groups suggests successful randomisation

why do we need to do a power calculation?

what is power?

need to calculate a sample size that is large enough to detect a difference between intervention and control group if it genuinely exists but not waste time and resources

probability of detecting a difference

why and how do we need to specify the outcome measures?

judge efficacy of the treatments must be specified

definition of, how and when outcome is measured

conducting an RCT

obtain EA

seek IC

screen for E

generate RAS

ensure COR

specify B

obtain ethical approval

seek informed consent

screen for eligibility

generate random allocation sequence

ensure concealment of randomisation

specify blinding

where to obtain ethical approval?

local or regional research ethics committees

ethics committees can refuse X and request Y

permission

amendments

clinical equipoise

uncertainty that the treatment is better than control

selection bias

bias due to systematic difference in the characteristics of participants selected to take part in a study and those who are not

lack of concealment of randomisation leads to this

blinding

single vs double

unawareness of treatment allocation

participant and/or outcome assessor is unaware

performance bias

behaving differently due to awareness of treatment

eg placebo

overcome by blinding

recall bias

behaving differently due to awareness of treatment when self-reporting

detection bias

change in measurement of outcome due to awareness of treatment

overcome by blinding

interviewer bias

change in behaviour when conducting an interview to measure outcomes due to awareness of treatment

analysing an RCT

check randomisation successful

check loss to follow up

main comparison analysis

loss to follow up bias

difference in proportion of drop out between treatment and control group

eg if those with side effects or bad outcome withdraw treatment effect if overestimated

intention to treat

participants are kept in their allocated groups for analysis even if they swapped groups, ideally they are also included in analysis even if they drop out

per protocol

only those who adhere to treatment and do not drop out are included

why is ITT preferable?

minimises confounding

maintains sample size

represents clinical practice

why does ITT minimise confounding?

maintains comparability of groups in terms of baseline characteristics

strengths of RCTs

no reverse causality

minimised confounding (confounders evenly distributed)

selection/performance bias minimised

measurement bias minimised

weaknesses of RCTs

loss to follow up bias

large sample sizes/long follow up times can be ==expensive and time consuming==

efficacy of treatment under trial conditions may be different from effectiveness in normal practise

not suitable for all exposures eg unethical

efficacy

does the treatment work under ideal conditions