From study objective to analysis plan

From study objectives to
analysis plan
Helen Maguire
don’t get bogged down -or stuck …
its logical
the keys to successful research
• get the research question crystal clear
• write a clear outline (concept paper)
• involve stakeholders
• talk to people
• revise again and again
• get the research question crystal clear
avoid
X
• too ambitious/ unfocussed
• unsound unscientific basis for hypothesis
• not clear what impact the findings will have in the
field ..
• no clear plan of next steps
• not ethical
• lacking appropriate expertise
The ad-hoc approach to
conducting an epidemiological study
Before data collection
• I want to do a study
– I am not clear about the
objectives
• I prepare a
questionnaire
– I am not clear about
what information I need
• I collect data
– I am not clear what I will
use for what
After data collection
• I come back with data
– I realize they are difficult
to analyse
• I analyse the data
– I realize it is difficult to
interpret the results
• I interpret the results
– I realize it is difficult to
use them
Sound familiar?
The life cycle of an epidemiological
investigation
Identifying
data needs
Involving the
programme
Spelling out the
research question
Formulating
recommendations
Formulating the
study objectives
Drawing conclusions
Planning the
analysis
Preparing data
collection instruments
Analysis
plan
Analysing data
Collecting data
analysis plan:
road map
epietmobile
1. choose a design to identify key indicators
2. identify parameters (variables) needed for
indicators
3. prepare the analysis
4. estimate sample size
Objectives
analysis plan:
road map
1. choose a design to identify key indicators
2. identify parameters needed for indicators
3. prepare the analysis
4. estimate sample size
Design and indicators
• what sort of designs do you know?
• what do you need to consider to help you decide
which one to use?
choosing a suitable study design
Epidemiological
studies
Experimental
(intervention)
Observational
(non-intervention)
Data from groups
Descriptive
Analytic
Ecological study
Data from
individuals
Descriptive
Cross sectional study
Analytic
Cohort study
Data from groups
Data from
individuals
Community trial
Clinical trial,
individual field
trial
Case control
study
what to consider when choosing
a study design
• is the study descriptive or analytical?
– are you comparing groups?
– are you estimating a frequency?
• is the outcome (e.g., disease) acute or chronic
– prevalence data for chronic disease
– incidence data for acute outcomes
• is it common or rare?
– case control for rare outcomes
– cohort / cross sectional for common outcomes
Design and indicators
testing a hypothesis
estimating a quantity
• determine whether hepatitis C is more common in
people who inject drugs (PWID)
– hypothesis testing
– crude objective, smaller sample size
• estimate the relative frequency of hepatitis C
infection in PWID vs others
– quantity estimating
– more elaborate objective, larger sample size
Objectives
clearly identify the study population
• this is different from the sample you will study
examples - what design?
• who is most likely to get wound infection after
appendicectomy?
• what lifestyle factors are associated with acquiring
hepatitis C?
• what experiences do patients with rare disease such as
MDR TB have using NHS services?
what design
• who is most likely to get wound infection after
appendicectomy
– cohort study
• what lifestyle factors are associated with acquiring
hepatitis C
– case-control study comparing those with hep C and
those without
• what experiences do patients with rare disease such as
MDR TB have using NHS services
– qualitative study or survey with open ended questions
analysis plan:
road map
1. choose a design to identify key indicators
2. identify variables needed for indicators
3. prepare the analysis
4. estimate sample size
Parameters
what indicators ?
• who is most likely to get wound infection after
appendicectomy
– cohort study
• what lifestyle factors are associated with acquiring hepatitis
C
– case-control study comparing those with hep C and those
without
• what experiences do patients with rare disease such as MDR
TB have using NHS services
– qualitative study or survey with open ended questions
estimating the relative frequency of meningococcal
carriage in children of parents who smoke vs others
• … from the objectives:
– analytical approach: compare two groups
– chronic condition: prevalence data
– common condition: survey
• study design:
– analytical cross sectional study
• indicator:
– ratio of prevalence of meningococcal carriage among
children of smokers vs others
Design and indicators
what’s needed to calculate the indicator?
• list the indicators that the study will generate
– proportion with carriage for categorical variables, prevalence
rate, prevalence ratios
• remember:
– outcome variable(s)
– “covariates” including
• potential risk factors
• potential confounders
• identify the information needed to calculate the
indicators
– numerators and denominators
• example: number carrying / total children
Parameters
....from indicators to variables
• identify variables that enable you to get your
indicator
– information “meningococcal C vaccination status” can be
collected by review of cards or interview of the mother
• choose the best variable
– review standardized guidelines (e.g., WHO, CDC)
– e.g. how to measure smoking
• plan data collection methods for each variable
–
–
–
–
record review
interview
observation
laboratory data
Parameters
covariate measurement
for meningococcal carriage study among
children of smokers vs others
• potential risk factors
– income
(validated field methods)
– ethnic group
– education
– area of residence
– vaccination against Men C
• potential confounding factors
– age
– sex
Parameters
analysis plan
road map
1. choose a design to identify key indicators
2. identify parameters needed for indicators
3. prepare the analysis
4. estimate sample size
Analysis
rationale for preparing the
data analysis in advance
• focus on the objectives of the study
• avoid multiple comparisons
• avoid comparisons for which the study was not
designed ….(tempting as it is..)
• ensure data collected can be analyzed
– “Other, specify: _____” ?? .. groups that cannot be analyzed
• save time
– filling dummy tables speeds data analysis
Analysis
a word about coding
26
a word about coding
• binary coding (usually):
– “1” is yes
– “0” is no
• gender (usually):
– “1” is male
– “0” or “2” is female
• age (or any categorical ordered var.)
– by percentiles
– by common sense
27
preparing the analysis,
stage by stage
• recoding stage
– example: age into age groups
• descriptive stage
– calculate prevalence or incidence
• analytical stage
– univariate, stratified and multivariable analysis
– prepare empty (dummy) tables (shells) now
Analysis
initial stage of analysis meningococcal
carriage according to smoking status
• recoding stage
– create outcome data with laboratory results
-carriage Yes/No
– recode smoking data
dichotomize quantitative smoking variable (how many do
you smoke a day (0,1,2,3…20))
-smoke Yes/ No
• descriptive stage
– calculate prevalence of meningococcal carriage
Analysis
analytical stage meningococcal carriage
according to smoking status of parent
• univariate analysis
– prevalence of outcome by age, sex and residence
– prevalence of outcome by smoking (potentially examine
dose response effect)
• stratified analysis
– prevalence of outcome by smoking, stratified for age, sex
and residence
• multivariable analysis (adjusted risk/rate ratio)
– logistic regression model
– binomial regression
Analysis
dummy table for meningococcal carriage study –
proportion carriage by factor (analytical stage) *
Prevalence
exposures/factor
exposed
unexposed
Prevalence ratio
(95% confidence
interval)
female sex
XX/XX (xx%)
XX/XX (xx%)
XX (XX-XX)
caucasian
XX/XX (xx%)
XX/XX (xx%)
XX (XX-XX)
age > median
XX/XX (xx%)
XX/XX (xx%)
XX (XX-XX)
smoking parent
XX/XX (xx%)
XX/XX (xx%)
XX (XX-XX)
residence urban
area
XX/XX (xx%)
XX/XX (xx%)
XX (XX-XX)
*variables dichotomized
analysis plan:
road map
1. choose a design to identify key indicators
2. identify parameters needed for indicators
3. prepare the analysis
4. estimate sample size
Sample size
the analysis plan determines
the sample size
• choose the study design
– cohort, case control or survey
• determine the level
– descriptive or analytical
• common mistake
– designing a descriptive study
– making comparisons for which the sample size is insufficient
Sample size
sample size for study on meningococcal
carriage
• study design
– analytical cross sectional survey
• level
– analytical
– need to
• use prevalence ratio for sample size estimation
Sample size
take home messages
•
clarify again - precise focussed objectives
•
choose a design - identify the indicator
•
know the parameter (variable) you want before you
think about how to get information about it
•
know where you go with the analysis
–
•
the planned analysis drives the data needs and not the
reverse
work out sample size from all of the above
• thanks for your attention